Nishi Family › Compare › Web CMS and Site Building

Nishi Compare · capabilities, evidence and delivery

Web CMS and Site Building

Compare capabilities and delivery evidence. Source discovery, qualification, deployment and accepted outcomes are separate; an open contract names work still to verify.

Nishi vs WordPress and Wix and Webflow and Drupal

Overview

The purpose, declared position and evidence coverage of this domain. Source presence and completed acceptance are different measures.

Where we are. A working sovereign CMS and site factory: structured content store, whole-site blueprint as one data file, requirements brief to finished copy, design-token SSOT, template corpus, admin save-preview-publish daemon, REST plus projection API, WordPress import, sha-verified export-import (exceed), sha-pinned static publish, BM25 on-site search, fire-once scheduling, comments, forms, the enforced conversion-funnel floor (exceed), e-commerce primitive, i18n, validated block rows, NL command edit with rollback, per-pattern generation, A-B assignment, WCAG-AAA enforced fleet-wide, roles FSM, never-brick deploy (exceed), multisite and SEO by construction. Behind WordPress on depth where its ecosystem leads: N-deep revisions, a visual canvas, and extensions.

Where we need to go. A CMS a non-developer runs for real sites on owned hardware with the three depth axes WordPress owns -- restore any revision, edit on a visual canvas, extend with declared organ-plugins -- without importing the foreign-code marketplace that is WordPress's attack surface.

The unit. 1 u = one measured session-leg (estate calibration: graphics R21 in one leg 2026-08-15). Local evidence: the site-factory gates (requirements 6/6, archetype 9/9, export-import 7/7) each landed as one leg; estimates are relative to those.

Where we are: 3 open rungs. Breadth is real and gated; the three open rungs are WordPress's ecosystem-depth axes. Counts measured at emit below this line.

Cost to revision depth: 1 u. CM1 restore-to-any-snapshot, additive and reversible.

Cost to a visual editor: 2.5 u. CM2 the block canvas emitted from the validated row model, round-tripping byte-identical.

Cost to extensibility: 2 u. CM3 the declared organ-extension registry -- sovereign plugins, fail-closed on undeclared hooks.

Research bar. WordPress is measured on revisions, Gutenberg block editor, 60k-plugin ecosystem, REST API. Theirs: the reference on all three open axes. Ours: CM1 CM2 CM3 measured on this page.

Research bar. Webflow is measured on visual designer, variables, CMS API. Theirs: the visual-canvas bar. Ours: CM2.

Latest recorded release

2026-10-01 · CM30

CM30 to CM33 are proven on the NAS: the referee lib gained pj_rows_ratings, the one Elo ruler over judged rows (source 3adc3c8d, 31491 B), the surface's pjj_ratings is a one-line delegation to it (source bb9a51bd, 37846 B), nx_pairjudge_gate built on the NAS as 39c12c6b (217320 B), promoted, and /api/gate_run on the promoted binary reads GREEN exit 0 with 54 of 54 teeth against the staged daemon cefa7556; the daemon itself stays staged and undeployed

Release history and work log

Release entries describe recorded changes; they do not establish that every capability passed evaluation.

61 of 68 capability source symbols found|4 of them authored stronger-claims|7 open|coverage 897/1000|adoption 47 full / 14 partial

Evidence profile — what the gaps on this board actually are

Measured by nx_swcompare_evidence, read back by nx_evprofile_lib. Every figure is a count with its denominator — there is deliberately no score, no grade and no percentage anywhere in this band, because a stored scalar is a field a seat can edit and a counted partition is not.

proven able to fail counts the gates that have a RECORDED RED — nx_gate_bite mutated the gate subject, rebuilt it, watched the gate go red, and that record is inside the shared TTL. never bitten is its complement over the same denominator: those gates ran and were green, and nothing has ever shown them able to detect anything, so their green is a statement about this run and not about the gate. green at 0/0 is a separate and much weaker observation — the gate printed GREEN on a zero denominator, so its own tooth counter says it examined nothing. A gate can be green, non-zero, and still never bitten; that is the common case and it is now visible instead of implied.

partition: grounded + unsupported = 29 vs present 29 · named + unnamed + withheld = 6 vs open 6 · both reconcile

liar-kill conj=GPQN · all four conjuncts held

graded document: BUILDROOT tree, 9017 bytes · gates map: PRIMARY · stamped 16d 19h ago · STALE against the shared TTL of 604800s — this is evidence about a build nobody can vouch for now · source ../knowledge/status/evstamp_cms.verdict

Gap class	What it is, and the work it names
STALE	This profile is older than the shared TTL, or its clock ran backwards. It is reported as unusable rather than as fresh, because being wrong in the direction of refusing is the safe direction.

This band reports the referee counts. The per-axis worklist rows — which axis is unsupported, which watch is ready to flip, which gap is unnamed — are printed by nx_swcompare_evidence cms itself and are not carried on the stamp, so this page names the classes and the producer names the rows. That split is stated rather than hidden: a count without a worklist is not actionable, and this band is honest about which half of that it is.

Production map

Follow the dependencies, declared acceptance criteria and recorded priorities. Inspect source binding before treating a rank as executable work.

Ranking source binding: STALE_OR_INVALID_BINDING. Recorded priorities require current acceptance evidence and resource checks before execution.

Recorded ranking withheld

Retained ranking is not bound to the current plan and matrix: STALE_OR_INVALID_BINDING. Historical data remains retained. Refresh the ranking after resolving the input differences. Latest execution outcome is unverified.

Declared roadmap — contract, acceptance, executor, effort

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Snapshot organ (CM0)	snap_read_file	Content snapshots plus one-step prev backups -- LANDED, listed so CM1 carries momentum	Organ	0 u
N-deep revisions with restore (CM1) after CM0	snap_restore_n	Restore any recorded snapshot; the pre-restore state is itself snapshotted first so every restore is reversible; gate proves restore-then-restore returns byte-identical content and a missing snapshot id is refused	Organ	1 u
Visual block canvas (CM2)	pb_canvas_emit	A canvas emitted FROM the validated row model (pb_row_valid) with drag, resize and reorder; gate proves a canvas edit round-trips to a blueprint that re-validates and emits byte-identical pages when nothing moved	Organ	2.5 u
Organ-extension registry (CM3)	cp_ext_load	An organ declares the hooks it serves in its header; the site builder loads it by name and refuses an undeclared hook; gate proves a declared extension renders into the page and an undeclared one is refused with the hook named	Organ	2 u

Milestones

Milestone	Rungs	Cumulative
M1 · Revision depth	CM1	1 u
M2 · Visual editing	CM2	3.5 u
M3 · Extensible site factory	CM3	5.5 u

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Showcase presentation tier (CM4)	ed_showcase_css	An image-led tier ON the one editorial system: full-bleed hero (image or seeded art), self-hosted display face, sticky nav with skip link, 4:5 image cards, dark band, footer; done when the incumbent site re-emits byte-identical and the craft ruler reads 15 of 15 on the emitted index (gospelofwest.com v3, 2026-09-21)	Organ	1 u
Design-craft ruler with file mode (CM5) after CM4	ua_kvn	nx_ui_audit grades bytes on disk (a release before golive, a mirrored rival) with 15 presence points and raw counts ADDITIVE to the hygiene score; done when the same counts are published for our page and the nike.com mirror side by side on this board	Organ	1 u
Rendered-pixel design referee (CM6) after CM5,CM13	pj_referee	A sovereign render of the emitted page judged on pixels (overlap, contrast as painted, hierarchy); done when the referee grades the live gospelofwest.com index and disagrees with at least one planted defect the bytes ruler passes	Organ	3 u
Licensed imagery pipeline (CM7) after CM4	sm_admit_image	heroimg and card images admitted only with a licence row beside them; done when an unlicensed path is refused by name and a licensed one renders in the hero	Organ	1.5 u

Milestones

Milestone	Rungs	Cumulative
M4 · Award-craft presentation	CM4,CM5,CM6,CM7	12 u

Ladder to SOTA

Two dated targets, and every rung's role toward them. Best in class is the proven, deployed leader today; the frontier is the research edge as of its month. Each target is dated by a sotabar row and judged against the current month by the bar-age ruler (FRESH, ATTESTED with the fallback month named, or refused). Each rung is substrate (what the target's own methods consume), an arm (a sovereign arm that triangulates without competing for the number), a contender (its done-rule is a number against the target) or superseded (kept as capability, retired as a claim), and carries the rungs it stands on.

best-in-class (T-BIC-WEBCRAFT), dated by SB-WEBCRAFT = 2026-10 FRESH. Emitted sites read as award-grade: image-led hero, display type, motion, footer, measured by the craft ruler on bytes and by a pixel referee on renders, with the operator's ACCEPT as the third signature ref awwwards-sep-2026

Rung	Role	Toward	Stands on	Why
CM0	substrate	T-BIC-WEBCRAFT	-	content snapshots every production site stands on
CM1	substrate	T-BIC-WEBCRAFT	CM0	restore-to-any-revision is what lets a presentation be iterated without fear
CM2	arm	T-BIC-WEBCRAFT	-	a visual canvas edits the same validated rows the tier emits from
CM3	substrate	T-BIC-WEBCRAFT	-	declared organ extensions carry the tier's future sections
CM4	contender	T-BIC-WEBCRAFT	-	the tier that carries the presentation
CM5	arm	T-BIC-WEBCRAFT	CM4	the bytes ruler that compares ours to the rival by the same counts
CM6	contender	T-BIC-WEBCRAFT	CM5,CM13	the pixel referee, the only judge of beauty on this board
CM7	substrate	T-BIC-WEBCRAFT	CM4	imagery the hero and cards consume

frontier (T-FRONT-WEBCRAFT), dated by SB-WEBCRAFT = 2026-10 FRESH. The September 2026 Site of the Day class goes past image-led pages: 3D and WebGL scene sections and GSAP-class scroll choreography (the listing names WebGL 16 times and GSAP 21 times across 25 winners); the sovereign frontier is the same scene hosted by the estate's own WASM and WebGPU engines with a CSS-only door, zero third-party script in the page ref awwwards-sep-2026

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
3D scene section rendered by the sovereign engines (CM8) after CM4	ed_sc_scene	A hero or band section that hosts a sovereign WASM or WebGPU scene from the estate's own engines behind a CSS-only fallback door, emitted from a blueprint key; done when the tier emits it, the page still grades 15 of 15 on bytes with zero third-party script, and the scene door announces which tier it took	Organ	3 u

Rung	Role	Toward	Stands on	Why
CM8	contender	T-FRONT-WEBCRAFT	CM4	the scene section is the frontier presentation itself

Milestones

Milestone	Rungs	Cumulative
M5 · Frontier presentation	CM8	15 u

best-in-class (T-BIC-BUILDER), dated by SB-BUILDERS = 2026-09 ATTESTED. The emitter does what the September 2026 builders do, from data: a design system derived from evidence for a brand profile, a section catalog wide enough to choose from, motion presets, licensed imagery, a conversational edit agent, cross-device verification, every value measured rather than typed and the whole page still zero script ref uniteai-builders-2026-09

frontier (T-FRONT-WEBGEN), dated by SB-WEBGEN-RESEARCH = 2026-09 ATTESTED. The September 2026 research frontier as a product: section layouts solved from constraints over the derived tokens, refinement grounded in a render the referee reads, agreement with expert pairwise judgement published beside every grade, design decisions proven in the emitted bytes, and agent-ready pages; no builder ships these and the estate can, because the whole loop is sovereign organs over data ref arxiv-webdesign-2026-09

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Brand and IA profile selects corpus and recipe (CM9) after CM5	sb_brand_profile	A brand profile (industry, personality, palette seed) in the .site blueprint selects the evidence corpus the tokens derive from and the section recipe; done when two profiles emit measurably different token sets from different corpora and the incumbent site re-emits byte-identical without one	Organ	1.5 u
Evidence corpus refreshed from the month's award listings (CM10) after CM5	crf_refresh	A beat mirrors the month's Site of the Day and Sites of the Year pages with their stylesheets through nx_research_fetch, re-derives the tokens and journals every n that moved; done when a re-derive after a refresh changes at least one token with its n and the journal names the source	Organ	1.5 u
Section catalog with a picker (CM11) after CM9	sa_catalog_pick	A declared section catalog (each entry with its measured tokens and done-rule) and a picker that composes a page from the brief, refusing a section whose inputs the brief lacks; done when the catalog holds at least the 28 crawled patterns as pickable entries and a brief with a missing input is refused by name	Organ	2.5 u
Motion presets as data, zero script (CM12) after CM4	ed_motion_preset	A preset table (reveal, parallax, scroll-scrub, sticky, view transitions) applied per section through CSS scroll-driven animation and view transitions; done when the showcase page carries at least three distinct presets, grades 15 of 15 on the craft ruler with zero script, and the reduced-motion guard still disables every one	Organ	1.5 u
Cross-device capture driver (CM13)	cd_capture	A sovereign CDP client over the estate websocket client that sets device metrics and captures a real browser at declared phone and desktop widths, the pixels handed to the referee; done when a 390 px capture reports innerWidth 390 and the phone render of gospelofwest is graded, ending UNMEASURED	Organ	2 u
Design decisions verified in the bytes (CM14) after CM5	ua_token_ref	A ruler tooth that refuses a defined-but-unreferenced token and a blueprint key that lands in no measured CSS property; done when it fires on a planted unreferenced token and passes the live showcase page	Organ	1 u
Agent-ready emission (CM15)	sb_agent_ready	schema.org JSON-LD, explicit action cues and dated validity emitted from the blueprint with zero script, graded by an agent-readiness tooth; done when the tooth passes on the emitted index and fails on the same page with the block removed	Organ	1 u
Constraint-solved section layout (CM16) after CM11	ed_layout_solve	A solver over the derived tokens and the brief chooses grid, grouping and whitespace per section, printing its objective beside the page; done when a planted misaligned layout scores worse than the solved one on the printed objective and the pixel referee agrees	Organ	3 u
Render-grounded refinement loop (CM17) after CM13	sb_refine_loop	The emitter re-runs with the referee's named defects until the pixel referee and the bytes ruler both pass, each iteration journaled; done when a planted defect is repaired within a declared iteration budget and a page that cannot converge is refused by name	Organ	2.5 u
Pairwise-preference referee with published agreement (CM18) after CM6	pj_pairwise	Candidates compared pairwise by the estate referee and an outside VLM oracle, ranked by a rating model, with agreement to a human-judged set published beside every grade, nx_ui_judge extended rather than twinned; done when the referee's ranking of a judged set agrees with the human ranking above a pre-declared floor and the number is on this board	Organ	3 u

Milestones

Milestone	Rungs	Cumulative
M6 · Builder parity from data	CM9,CM10,CM11,CM12,CM13,CM14,CM15	26 u
M7 · Research frontier as product	CM16,CM17,CM18	34.5 u

Rung	Role	Toward	Stands on	Why
CM9	contender	T-BIC-BUILDER	CM5	the brief-to-design-system leg the builders sell, from evidence instead of a model
CM10	substrate	T-BIC-BUILDER	CM5	the corpus every derived token stands on, refreshed with the field
CM11	contender	T-BIC-BUILDER	CM9	the section breadth the builders have and the emitter lacks
CM12	contender	T-BIC-BUILDER	CM4	the choreography Webflow sells as presets, reached with zero script
CM13	arm	T-BIC-BUILDER	-	the capture that ends UNMEASURED on phones and feeds the referee
CM14	arm	T-FRONT-WEBGEN	CM5	the no-design-theater tooth the field measures generators failing
CM15	contender	T-BIC-BUILDER	-	agent-readiness the builders reach through plugins
CM16	contender	T-FRONT-WEBGEN	CM11	the solved layout the September 2026 frontier rates above heuristics
CM17	contender	T-FRONT-WEBGEN	CM13	the render-grounded loop no builder ships
CM18	contender	T-FRONT-WEBGEN	CM6	the calibrated pairwise referee, the only honest judge of the loop

Milestones

Milestone	Rungs	Cumulative
M8 · Structured emission: composed and organic modes as data	CM19,CM20,CM21,CM22,CM23	9 u

best-in-class (T-BIC-IA), dated by SB-IA = 2026-08 ATTESTED. The documentation-class shell the best dense sites converge on: a sticky section rail beside one measured reading column at the wide breakpoint, a summary before detail, records as tables with named columns, related records joined in one place and linked both ways, every width and breakpoint derived from measured pages ref ghdocs-hello-world

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Layout tokens derived for board pages (CM19) after CM5	ua_layout_derive	The design ruler measures rail width, content width, column counts, breakpoints, sticky use and in-page navigation across the mirrored board-class pages and derives one board token set with n on every row [tailwind-responsive-design] [ghdocs-hello-world] [mdn-grid-common-layouts]; done when buildroot/knowledge/compare/board.tokens is written by the ruler and every row carries n	Organ	1 u
Responsive shell with a section rail (CM20) after CM19	sc_shell_pass	The one compare assembler wraps its four production sections in a two-column shell at the derived breakpoint: a sticky section rail beside a wide content column, card grids that fill the width, multi-column lists, tables that scroll in their own container, nothing removed [mdn-structuring-documents] [nng-layer-cake]; done when the emitted text bytes are unchanged, the page never scrolls sideways at 390 and 1440 px, and the shell reads its widths from board.tokens	Organ	2 u
Records as clean tables (CM21) after CM20	rct_end	Every record class the boards emit as prose (rank derivations, ledger notes, evidence bands, incumbents) renders as a table with named columns through one renderer, with progressive disclosure for the long ones [nng-progressive-disclosure]; done when a census of prose blocks carrying pipe or key-equals records reads zero on every published board	Organ	2 u
Relationship dossiers per rung (CM22) after CM21	sc_rung_dossier	A rung's plan row, matrix row, gate row, log rows, cited refs and design rows join by symbol into one dossier linked from the matrix cell and the rank table, so related information is read in one place [ia-arxiv-2026-09]; done when every rung on a board carries a dossier with its matrix cell and its gate verdict and both links resolve	Organ	3 u
Design modes as data: composed and organic (CM23) after CM20	dmo_mode_load	Two modes declared in knowledge/design_modes.conf, the composed mode (a designed path, framed sections, a summary first, a rail, disclosure) [wiki-japanese-garden] and the organic mode (title, body, related; restraint, no chrome) [wiki-wabi-sabi], selected per surface by a design row on the blueprint or the board; done when the site emitter renders a home page in the composed mode and a notes page in the organic mode from the same conf, gate-proven	Organ	1 u

Rung	Role	Toward	Stands on	Why
CM19	substrate	T-BIC-IA	CM5	the measured widths every shell row stands on
CM20	contender	T-BIC-IA	CM19	the documentation shell itself, on every board
CM21	contender	T-BIC-IA	CM20	records as tables, the readability half of the target
CM22	contender	T-BIC-IA	CM21	the join that no rival board makes, the exceed candidate
CM23	arm	T-BIC-IA	CM20	the two modes as data, so the composed shell and the organic page come from one conf

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Paper behind the glyphs (CM24) after CM6,CM13	cd_textfree_shot	The capture driver takes a second shot of every door with every text colour painted transparent, so the referee reads the paper each glyph pixel was painted over instead of estimating it from a mixed histogram: today the exact method proves solid paper only, and text over a photograph or a gradient keeps the conservative band means, the case the field's pixel checker also only warns on [atf-textcontrastcheck]; done when the eyebrow over the production hero photograph is graded against the pixels behind its own glyphs and still fails, a gradient running from the paper toward the ink no longer reads as solid paper, and a solid-paper box reads the same ratio both ways	Organ	2 u

Rung	Role	Toward	Stands on	Why
CM24	substrate	T-BIC-WEBCRAFT	CM6,CM13	the paper behind each glyph, which the pixel referee consumes wherever text sits on a photograph or a gradient

Milestones

Milestone	Rungs	Cumulative
M9 · Referee sees the paper behind the text	CM24	2 u

best-in-class (T-BIC-RUBRIC), dated by SB-RUBRICS = 2026-10 FRESH. Every emitted page scored the way the juries and the UX research bodies score, from data: one score per declared rubric axis beside the pixel grade, each axis mapped to the estate referee that measures it or printed UNMEASURED, the journey to the site's declared action walked in a real browser and graded against the checkout, form and navigation rows the research bodies publish, and a mirrored winner scored the same way so the gap is a number ref awwwards-evaluation-2026-09

frontier (T-FRONT-LOOP), dated by SB-RSI-LOOP = 2026-09 ATTESTED. The evaluate, refine, improve loop no builder ships: every emitted variant archived with its referee scores, the next modification chosen from the archive by score and novelty, admitted only when it beats its parent on every deterministic referee and on the journey receipt at a declared confidence, the loop's own proposal policy evolved the same way under a hash-pinned policy that the loop cannot edit and referees that live in another lane, and every accepted change carrying its held-out gain ref arxiv-selfimproving-2026-09

Milestones

Milestone	Rungs	Cumulative
M10 · Evaluate, refine, improve: the loop reaches the frontier	CM25,CM26,CM27,CM28,CM29	12.5 u

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Award and UX rubrics as referee axes (CM25) after CM6,CM14	rub_rubric_score	knowledge/design_rubrics.conf carries the juries' axes and weights [awwwards-evaluation-2026-09] [webby-judging-2026-09] [fwa-about-2026-09], NN/g's ten heuristics [nngroup-heuristics-2026-09] and Baymard's benchmark topics [baymard-research-2026-09] as data, and the rubric referee maps every axis to the estate referees that measure it (pixel referee, tokenref, bytes ruler, a11y-lite, funnel floor) printing UNMEASURED for the rest; done when the alpha and a mirrored winner are both scored on every declared axis, the mapping names the unmeasured axes, and a planted contrast regression lowers the design axis	Organ	2 u
Journey driver measures the 1-2-3 path in the browser (CM26) after CM13	cd_journey_run	knowledge/journeys.conf declares each site's path (landing, action, completion) as steps the capture driver walks at both doors, recording clicks, fields, scroll depth and settle time per step and grading the receipt against the research rows (error prevention, visibility of status, checkout and form rows) [nngroup-heuristics-2026-09] [baymard-research-2026-09] [builtformars-2026-09] [pageflows-2026-09]; done when the gospelofwest and cms-scene alpha paths complete with a friction receipt per step and a planted extra step reads worse	Organ	2.5 u
Variant archive and selection in the shape of the Darwin and Mendel Godel Machines (CM27) after CM17,CM26	sr_archive_select	Every emitted variant (blueprint, tokens, section order, presets) is archived content-addressed with its referee scores and journey receipt; the loop chooses the next parent by score with a novelty bonus and proposes a modification from the referees' named defects and the archive's comparative signals, admitting a child only when it beats its parent on every deterministic referee and on the journey receipt at a declared confidence [arxiv-godel-machine-2026-09]; done when a planted defect is repaired within a declared iteration budget, a page that cannot converge is refused by name, and the archive keeps every variant	Organ	3 u
Multivariate optimisation of the design knobs (CM28) after CM27	sb_knob_search	The knobs are the derived-token bands and the catalogue's discrete choices (section order, presets, hero mode, density); a factorial search over them is scored offline by the referee panel and the journey receipt, and online nx_cms_abtest assigns live traffic to the top variants with each journey step's completion logged and judged by Bayesian A/B arithmetic [arxiv-ui-optimization-2026-09] [arxiv-abtesting-web-2026-09]; done when the search prints its objective per variant, the chosen variant beats the incumbent at the declared confidence, and a knob outside its evidence band is refused	Organ	3 u
Guarded self-improvement of the improver (CM29) after CM27	sb_policy_pin	The loop's proposal policy is a versioned knob set evolved the same way under a hash-pinned policy (goals, scope, evaluation identity, budget, release conditions) that the loop cannot edit, with the referees in another lane, harness tampering audited and the judge held to advisor [arxiv-selfimproving-2026-09]; done when a mutation that edits a referee or the policy pin is refused by name and every accepted policy change carries its held-out gain	Organ	2 u

Rung	Role	Toward	Stands on	Why
CM25	contender	T-BIC-RUBRIC	CM6,CM14	scores every page on the axes the juries and the research bodies publish
CM26	contender	T-BIC-RUBRIC	CM13	measures the usability axis in the browser on the site's own path
CM27	contender	T-FRONT-LOOP	CM17,CM26	the archive and the selection rule
CM28	contender	T-FRONT-LOOP	CM27	the search over the knobs, offline then online
CM29	contender	T-FRONT-LOOP	CM27	the pinned policy and the evolved proposer

best-in-class (T-BIC-PAIRJUDGE), dated by SB-PAIRJUDGE = 2026-09 ATTESTED. The judging surface does what Design Arena and Code Arena do, read from their bytes: blind head-to-head pairs with random sides, one pair at a time, a vote per choice, a rating from the one Bradley-Terry-class ruler with a margin, win rate and standings per candidate, the next pair chosen where a vote moves the ranking most, a four-candidate session bracketed to a final ranking, and the running page itself beside the stills; every verdict a row in the referee's grammar the estate re-reads ref designarena-home-2026-09

frontier (T-FRONT-PAIRJUDGE), dated by SB-PAIRJUDGE = 2026-09 ATTESTED. The September 2026 research frontier as an estate capability: a machine co-judge anchored to the human rows with its per-judge sensitivity and bias corrected and its vote cast only where its agreement with humans clears the floor at that score gap, the agreement published as a curve over the gap, Krippendorff alpha across judges beside the pairwise agreement, the judge's dwell kept as gap evidence, leaderboard robustness under leave-one-out and flip perturbation with rank intervals, per-axis verdicts and per-judge preference profiles; no arena publishes its judge's calibration beside its leaderboard, and the estate can because the rows, the referee and the ruler share one grammar ref arxiv-arena-preference-2026-09

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Blind sides: Left and Right by a keyed coin, the click mapped back to the referee's A and B (CM30) after CM18	pjj_swapped	The pair section names no candidate; which candidate sits left is the parity of the pair's nibble of the session token, the click (L, R, tie) is mapped back to A, B, tie before the referee's one writer runs, and the side shown is recorded on a meta row; done when the gate proves a known-answer token swaps exactly the predicted pairs and a Left click on a swapped pair lands as B on the referee's row [designarena-home-2026-09]	Organ	0.5 u
Dwell recorded per verdict: the judge's time on the pair as gap evidence (CM31) after CM30	pjj_meta_row	Every verdict carries a meta row (pair, judge, date, side shown, dwell in ms from the pair's last render, door) written by one writer and read by one reader, the judge's median dwell printed on the surface; done when the gate measures a dwell at least as long as a real wait between render and post and UNSHOWN for a pair posted without a render [arxiv-arena-preference-2026-09]	Organ	0.5 u
Next pair by information: fewest judgments, then the smallest rating gap (CM32) after CM30	pjj_next_pair	The surface serves one pair at a time, the open pair whose candidates carry the fewest judgments and, among those, the smallest gap in the referee's Elo over every judged row, then the lowest ordinal; done when a known-answer row set makes the zero-gap pair of two winners the next pair where the first open pair is the wrong choice, and a tie moves no rating [arxiv-2602-09411-ksort-eval]	Organ	1 u
Standings on the surface: rating, W-L-T, win share with its Wilson interval (CM33) after CM32	pjj_standings	The referee's one rating ruler (nx_dr_elo from its start rating) over every judge's rows ranks the candidates; each row prints wins, losses, ties, the win share of decided pairs and its Wilson 95 interval from the gate base class, UNMEASURED with no decided pair, behind a disclosure until the judge is done; done when the table's rows equal the values the gate recomputes from the same rulers [designarena-home-2026-09]	Organ	0.5 u
Inter-judge agreement as Krippendorff alpha beside the pairwise agreement (CM34) after CM18	pj_kripp_alpha	The referee's receipt prints Krippendorff's alpha (nominal, over pairs judged by two or more judges) in permil beside the k-of-n agreement and its Wilson interval, in exact rational arithmetic; done when a hand-computed two-judge set over six pairs reproduces the textbook value and a single-judge set reads UNMEASURED by name [arxiv-2511-20513-designpref]	Organ	1 u
Live preview beside the stills: the candidate served on the surface (CM35) after CM30	pjj_live_route	GET /live/<k>/ serves the candidate's own index and the assets under its dir on the loopback surface so the judge interacts before voting, the way the arenas judge running apps; done when the gate fetches a candidate's page and one asset through the route, a path outside the dir is refused by name, and the pair section links both previews [arxiv-2609-15387-webcraftbench]	Organ	1.5 u
Human-anchored machine co-judge with a gap-calibrated vote (CM36) after CM34	pj_cojudge_anchor	The referee's agreement with the human rows is published as a curve over the pair's score gap; the machine casts a vote only on pairs whose gap is above the smallest gap at which its agreement clears the floor, its sensitivity and bias per judge estimated from the rows; done when the curve is on the receipt, a pair below the gap reads ABSTAIN by name and the anchored ranking of a held-out set agrees with the human ranking above the floor [arxiv-2609-19599-anchor] [arxiv-2609-15387-webcraftbench]	Organ	3 u
Per-axis verdicts: aesthetics, usability, requirement alignment and overall (CM37) after CM33	pj_axis_verdicts	The surface asks for an axis verdict beside the overall (a row per axis in the referee's grammar, the axis a declared field), the referee keeps one Elo table per axis and names the axis on which the human and the machine disagree most; done when a set judged on three axes yields three rankings and the receipt names the widest disagreement [arxiv-2609-15387-webcraftbench]	Organ	2 u
Leaderboard robustness: leave-one-out and flip perturbation with rank intervals (CM38) after CM33	pj_loo_stability	The receipt prints how often the top candidate survives dropping each single row and flipping each single decided row, and the rank interval per candidate over those perturbations; done when a set decided by one row reads FRAGILE by name and a set decided by many reads STABLE [arxiv-arena-preference-2026-09]	Organ	1.5 u
Per-judge preference profiles (CM39) after CM34	pj_judge_profile	The receipt prints each judge's agreement with the pool and with the machine and the pairs on which the judge departs from the pool; done when a planted contrarian judge reads as the lowest-agreement row and the pool's alpha drops by name when that judge is added [arxiv-2511-20513-designpref]	Organ	1.5 u
Tournament session: four candidates from one brief, brackets to a final ranking (CM40) after CM32	pjj_tournament	A session draws four candidates for one brief and runs the bracket the arenas run (two battles, winners and losers, a tie-breaker, a final ranking) on the surface, each battle a pair row; done when a four-candidate session produces a full ranking in five verdicts and the ranking equals the referee's Elo over those rows [designarena-home-2026-09]	Organ	1.5 u

Milestones

Milestone	Rungs	Cumulative
M11 · Judging at the arena bar	CM30,CM31,CM32,CM33,CM34,CM35,CM36,CM37,CM38,CM39,CM40	14.5 u

Rung	Role	Toward	Stands on	Why
CM30	substrate	T-BIC-PAIRJUDGE	CM18	blind sides, the arena's first rule; every later rung reads rows judged this way
CM31	arm	T-FRONT-PAIRJUDGE	CM30	the dwell is the gap evidence the frontier says the labels lack
CM32	contender	T-BIC-PAIRJUDGE	CM30	the next-pair rule that spends the judge's minutes where the ranking moves
CM33	contender	T-BIC-PAIRJUDGE	CM32	the standings with a margin, what the arena leaderboard shows
CM34	arm	T-FRONT-PAIRJUDGE	CM18	the agreement statistic the field reports beside every human-judged set
CM35	contender	T-BIC-PAIRJUDGE	CM30	the running page beside the still, how the arenas show a candidate
CM36	contender	T-FRONT-PAIRJUDGE	CM34	the anchored, gap-calibrated machine vote: the frontier itself
CM37	contender	T-FRONT-PAIRJUDGE	CM33	per-axis verdicts, the dimensions the September benchmark scores separately
CM38	arm	T-FRONT-PAIRJUDGE	CM33	robustness of the leaderboard the perturbation work measures
CM39	arm	T-FRONT-PAIRJUDGE	CM34	per-judge profiles, the personalisation the designer datasets call for
CM40	contender	T-BIC-PAIRJUDGE	CM32	the arena's session shape, a full ranking from one brief

Ladder verdict. LADDERED. Targets, rung roles and bars are counted in the tables below, each partition re-derived beside its total. Bars read FRESH against the current month 2026-10.

Ladder verdict, counted (17 rows, more than the 12-row disclosure bar)

Ladder verdict, counted
Class	Count	Note
targets	9	sotatarget rows on the plan
best in class	5	targets whose class is best-in-class, by code
frontier	4	targets whose class is frontier, by code
undated	0	targets naming no bar that exists
bad class	0	targets whose class is neither word
duplicate	0	target ids an earlier row already carries
rungs	41	rung rows on the plan
placed	41	rungs carrying a rungrole row
unplaced	0	rungs with no role toward any target
bad role	0	rungrole rows naming a role the ladder does not know
bad target	0	rungrole rows naming a target no sotatarget row declares
rungs, summed	41	RECONCILED: the parts sum to the total 41
substrate	8	what the target's own methods consume
arm	9	a sovereign arm that triangulates without competing for the number
contender	24	its done-rule is a number against the target
superseded	0	kept as capability, retired as a claim
placed, summed	41	RECONCILED: the parts sum to the total 41

Bar freshness, counted
Class	Count	Note
bars	8	sotabar rows on the plan
fresh	2	the bar month is the current month
attested	6	an older bar with a this-month barscan reporting no newer entry
stale	0	a this-month scan saw a newer month the bar ignores
unattested	0	nobody looked this month
malformed	0	a sotabar row missing a field
duplicate	0	bar ids an earlier row already carries
bars, summed	8	RECONCILED: the parts sum to the total 8
current month	2026-10	the month every bar is judged against

Treatment verdict. NOT DECLARED. No rung on this board declares its three-step treatment (a treat row naming the MCP row, the API route, the primitive it composes, the workflow that composes it and the agent beat that runs it); the ranker labels every rung TREATMENT-UNDECLARED until it does.

Design choices

Every design choice is a claim with its evidence beside it. A row names the choice, the error class it prevents, where it came from (research, a mined shape from the estate's own bug record, a thought leader, or the operator), the rung ids and reference keys that stand behind it, its status (adopted, proposed or retracted) and the effect it measured, or UNMEASURED. A row is refused, and says so here, when its status or source kind is outside those words or when its evidence names a key that is neither a rung on this plan nor a reference in this domain's register. The same table renders on every domain that lands design rows; the shape is shared, the rows are the board's.

Id	Choice	Error class it prevents	Source	Evidence	Status	Measured effect
GD1	A two-column shell at the wide breakpoint: a sticky section rail beside one wide content column	the single endless column that reads as a dump	research	[mdn-grid-common-layouts], [ghdocs-hello-world], CM19, CM20	adopted	published 2026-09-22: /compare/cms 333965 B carries data-shell='two-column' with DERIVED bp2=1012px rail=306px stickytop=0 (board.tokens n=5/6/5 from 20 measured sources), page_verify GREEN a11y-issues=0, page text bytes unchanged (nx_compare_garden_gate 32/32 text-identity tooth); reader-position effect UNMEASURED until the pixel referee (CM6) reads the live page
GD2	Widths, gaps and breakpoints derived from measured board-class pages, never typed	taste wearing a token's clothes	research	[tailwind-responsive-design], CM19	adopted	board.tokens 2026-09-22 from 20 measured sources (6 documentation pages + their 14 stylesheets): bp2 1012 px n=5, rail 306 px n=6, stickytop 0 px n=5, mpx 1280 px n=5, bp1 768 px n=10 -- every layout number carries its n, and the shell that reads a row with n=0 stays single column by name
GD3	Every record class renders as a table with named columns; prose is for argument, tables are for records	information dumps	research	[nng-progressive-disclosure], CM21	proposed	a record can be scanned column by column and compared row to row
GD4	Related records join by symbol into one dossier and link both ways	relationships scattered across sections	operator	CM22	proposed	one place to read a rung and everything that bears on it
GD5	Summary before detail and disclosure for the long tables, the layer-cake and progressive-disclosure patterns	the important line buried under the long one	research	[nng-layer-cake], [nng-progressive-disclosure], CM20	proposed	the first screen answers where we are
GD6	Two named modes, composed and organic, chosen per surface as data	one emitter voice for every page	operator	[wiki-japanese-garden], [wiki-wabi-sabi], CM23	proposed	serious surfaces get the composed mode, notes get the organic one

Design ledger verdict. CLEAN. Every design row is counted in the table below with its partition re-derived beside the total; a refused row is rendered with its rule and stays a claim nobody may cite until its data is repaired.

Design ledger verdict, counted (14 rows, more than the 12-row disclosure bar)

Design ledger verdict, counted
Class	Count	Note
rows	6	design rows on the plan
adopted	2	a choice in force
proposed	4	a choice awaiting evidence or acceptance
retracted	0	a choice withdrawn, kept as history
refused	0	a row a rule refused; the rule is rendered on the row
refused: bad status	0
refused: bad source kind	0
refused: dead evidence key	0
malformed	0	a short row
partition sum	6	RECONCILED: the parts sum to the total 6
unmeasured claims	0	rows whose claim names no measurement
evidence tokens	15	reference keys the rows cite
dead evidence tokens	0	keys that resolve to nothing in the register
rows with no evidence declared	0

Inspect a rung and its prerequisites

Declared nodes 41. Rank input binding: STALE_OR_INVALID_BINDING. Dependency order is authored. Implementation, acceptance evidence, authority and resource readiness are unverified. No action is recommended or dispatched here.

Plan SHA-256 a9ea7fcc838a83ba59476c5cfc925802fb44011901c42a59769cff1518687ce9. Target rows 9; role rows 41. Existing risks and release worklog retain their own scope; no node completion is inferred.

Runtime observation coverage: 0 of 41 rungs have current, identity-bound execution observations. Observation may report RED. Accepted completion remains unknown; source presence, qualification and deployment do not establish it.

Use Enter or Space on a rung to inspect its contract. Prerequisite links locate another rung in this list; open its summary to inspect it. Estimates are authored effort, not forecasts.

CM0 — Snapshot organ
Prerequisites: None declared; this does not establish execution eligibility.
Contract: snap_read_file
Acceptance: Content snapshots plus one-step prev backups -- LANDED, listed so CM1 carries momentum
Authored effort: 0. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM1 — N-deep revisions with restore
Prerequisites: CM0 (acceptance unverified)
Contract: snap_restore_n
Acceptance: Restore any recorded snapshot; the pre-restore state is itself snapshotted first so every restore is reversible; gate proves restore-then-restore returns byte-identical content and a missing snapshot id is refused
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM2 — Visual block canvas
Prerequisites: None declared; this does not establish execution eligibility.
Contract: pb_canvas_emit
Acceptance: A canvas emitted FROM the validated row model (pb_row_valid) with drag, resize and reorder; gate proves a canvas edit round-trips to a blueprint that re-validates and emits byte-identical pages when nothing moved
Authored effort: 2.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM3 — Organ-extension registry
Prerequisites: None declared; this does not establish execution eligibility.
Contract: cp_ext_load
Acceptance: An organ declares the hooks it serves in its header; the site builder loads it by name and refuses an undeclared hook; gate proves a declared extension renders into the page and an undeclared one is refused with the hook named
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM4 — Showcase presentation tier
Prerequisites: None declared; this does not establish execution eligibility.
Contract: ed_showcase_css
Acceptance: An image-led tier ON the one editorial system: full-bleed hero (image or seeded art), self-hosted display face, sticky nav with skip link, 4:5 image cards, dark band, footer; done when the incumbent site re-emits byte-identical and the craft ruler reads 15 of 15 on the emitted index (gospelofwest.com v3, 2026-09-21)
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM5 — Design-craft ruler with file mode
Prerequisites: CM4 (acceptance unverified)
Contract: ua_kvn
Acceptance: nx_ui_audit grades bytes on disk (a release before golive, a mirrored rival) with 15 presence points and raw counts ADDITIVE to the hygiene score; done when the same counts are published for our page and the nike.com mirror side by side on this board
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM6 — Rendered-pixel design referee
Prerequisites: CM5 (acceptance unverified), CM13 (acceptance unverified)
Contract: pj_referee
Acceptance: A sovereign render of the emitted page judged on pixels (overlap, contrast as painted, hierarchy); done when the referee grades the live gospelofwest.com index and disagrees with at least one planted defect the bytes ruler passes
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM7 — Licensed imagery pipeline
Prerequisites: CM4 (acceptance unverified)
Contract: sm_admit_image
Acceptance: heroimg and card images admitted only with a licence row beside them; done when an unlicensed path is refused by name and a licensed one renders in the hero
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM8 — 3D scene section rendered by the sovereign engines
Prerequisites: CM4 (acceptance unverified)
Contract: ed_sc_scene
Acceptance: A hero or band section that hosts a sovereign WASM or WebGPU scene from the estate's own engines behind a CSS-only fallback door, emitted from a blueprint key; done when the tier emits it, the page still grades 15 of 15 on bytes with zero third-party script, and the scene door announces which tier it took
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM9 — Brand and IA profile selects corpus and recipe
Prerequisites: CM5 (acceptance unverified)
Contract: sb_brand_profile
Acceptance: A brand profile (industry, personality, palette seed) in the .site blueprint selects the evidence corpus the tokens derive from and the section recipe; done when two profiles emit measurably different token sets from different corpora and the incumbent site re-emits byte-identical without one
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM10 — Evidence corpus refreshed from the month's award listings
Prerequisites: CM5 (acceptance unverified)
Contract: crf_refresh
Acceptance: A beat mirrors the month's Site of the Day and Sites of the Year pages with their stylesheets through nx_research_fetch, re-derives the tokens and journals every n that moved; done when a re-derive after a refresh changes at least one token with its n and the journal names the source
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM11 — Section catalog with a picker
Prerequisites: CM9 (acceptance unverified)
Contract: sa_catalog_pick
Acceptance: A declared section catalog (each entry with its measured tokens and done-rule) and a picker that composes a page from the brief, refusing a section whose inputs the brief lacks; done when the catalog holds at least the 28 crawled patterns as pickable entries and a brief with a missing input is refused by name
Authored effort: 2.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM12 — Motion presets as data, zero script
Prerequisites: CM4 (acceptance unverified)
Contract: ed_motion_preset
Acceptance: A preset table (reveal, parallax, scroll-scrub, sticky, view transitions) applied per section through CSS scroll-driven animation and view transitions; done when the showcase page carries at least three distinct presets, grades 15 of 15 on the craft ruler with zero script, and the reduced-motion guard still disables every one
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM13 — Cross-device capture driver
Prerequisites: None declared; this does not establish execution eligibility.
Contract: cd_capture
Acceptance: A sovereign CDP client over the estate websocket client that sets device metrics and captures a real browser at declared phone and desktop widths, the pixels handed to the referee; done when a 390 px capture reports innerWidth 390 and the phone render of gospelofwest is graded, ending UNMEASURED
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM14 — Design decisions verified in the bytes
Prerequisites: CM5 (acceptance unverified)
Contract: ua_token_ref
Acceptance: A ruler tooth that refuses a defined-but-unreferenced token and a blueprint key that lands in no measured CSS property; done when it fires on a planted unreferenced token and passes the live showcase page
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM15 — Agent-ready emission
Prerequisites: None declared; this does not establish execution eligibility.
Contract: sb_agent_ready
Acceptance: schema.org JSON-LD, explicit action cues and dated validity emitted from the blueprint with zero script, graded by an agent-readiness tooth; done when the tooth passes on the emitted index and fails on the same page with the block removed
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM16 — Constraint-solved section layout
Prerequisites: CM11 (acceptance unverified)
Contract: ed_layout_solve
Acceptance: A solver over the derived tokens and the brief chooses grid, grouping and whitespace per section, printing its objective beside the page; done when a planted misaligned layout scores worse than the solved one on the printed objective and the pixel referee agrees
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM17 — Render-grounded refinement loop
Prerequisites: CM13 (acceptance unverified)
Contract: sb_refine_loop
Acceptance: The emitter re-runs with the referee's named defects until the pixel referee and the bytes ruler both pass, each iteration journaled; done when a planted defect is repaired within a declared iteration budget and a page that cannot converge is refused by name
Authored effort: 2.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM18 — Pairwise-preference referee with published agreement
Prerequisites: CM6 (acceptance unverified)
Contract: pj_pairwise
Acceptance: Candidates compared pairwise by the estate referee and an outside VLM oracle, ranked by a rating model, with agreement to a human-judged set published beside every grade, nx_ui_judge extended rather than twinned; done when the referee's ranking of a judged set agrees with the human ranking above a pre-declared floor and the number is on this board
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM19 — Layout tokens derived for board pages
Prerequisites: CM5 (acceptance unverified)
Contract: ua_layout_derive
Acceptance: The design ruler measures rail width, content width, column counts, breakpoints, sticky use and in-page navigation across the mirrored board-class pages and derives one board token set with n on every row [@tailwind-responsive-design] [@ghdocs-hello-world] [@mdn-grid-common-layouts]; done when buildroot/knowledge/compare/board.tokens is written by the ruler and every row carries n
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM20 — Responsive shell with a section rail
Prerequisites: CM19 (acceptance unverified)
Contract: sc_shell_pass
Acceptance: The one compare assembler wraps its four production sections in a two-column shell at the derived breakpoint: a sticky section rail beside a wide content column, card grids that fill the width, multi-column lists, tables that scroll in their own container, nothing removed [@mdn-structuring-documents] [@nng-layer-cake]; done when the emitted text bytes are unchanged, the page never scrolls sideways at 390 and 1440 px, and the shell reads its widths from board.tokens
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM21 — Records as clean tables
Prerequisites: CM20 (acceptance unverified)
Contract: rct_end
Acceptance: Every record class the boards emit as prose (rank derivations, ledger notes, evidence bands, incumbents) renders as a table with named columns through one renderer, with progressive disclosure for the long ones [@nng-progressive-disclosure]; done when a census of prose blocks carrying pipe or key-equals records reads zero on every published board
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM22 — Relationship dossiers per rung
Prerequisites: CM21 (acceptance unverified)
Contract: sc_rung_dossier
Acceptance: A rung's plan row, matrix row, gate row, log rows, cited refs and design rows join by symbol into one dossier linked from the matrix cell and the rank table, so related information is read in one place [@ia-arxiv-2026-09]; done when every rung on a board carries a dossier with its matrix cell and its gate verdict and both links resolve
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM23 — Design modes as data: composed and organic
Prerequisites: CM20 (acceptance unverified)
Contract: dmo_mode_load
Acceptance: Two modes declared in knowledge/design_modes.conf, the composed mode (a designed path, framed sections, a summary first, a rail, disclosure) [@wiki-japanese-garden] and the organic mode (title, body, related; restraint, no chrome) [@wiki-wabi-sabi], selected per surface by a design row on the blueprint or the board; done when the site emitter renders a home page in the composed mode and a notes page in the organic mode from the same conf, gate-proven
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM24 — Paper behind the glyphs
Prerequisites: CM6 (acceptance unverified), CM13 (acceptance unverified)
Contract: cd_textfree_shot
Acceptance: The capture driver takes a second shot of every door with every text colour painted transparent, so the referee reads the paper each glyph pixel was painted over instead of estimating it from a mixed histogram: today the exact method proves solid paper only, and text over a photograph or a gradient keeps the conservative band means, the case the field's pixel checker also only warns on [@atf-textcontrastcheck]; done when the eyebrow over the production hero photograph is graded against the pixels behind its own glyphs and still fails, a gradient running from the paper toward the ink no longer reads as solid paper, and a solid-paper box reads the same ratio both ways
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM25 — Award and UX rubrics as referee axes
Prerequisites: CM6 (acceptance unverified), CM14 (acceptance unverified)
Contract: rub_rubric_score
Acceptance: knowledge/design_rubrics.conf carries the juries' axes and weights [@awwwards-evaluation-2026-09] [@webby-judging-2026-09] [@fwa-about-2026-09], NN/g's ten heuristics [@nngroup-heuristics-2026-09] and Baymard's benchmark topics [@baymard-research-2026-09] as data, and the rubric referee maps every axis to the estate referees that measure it (pixel referee, tokenref, bytes ruler, a11y-lite, funnel floor) printing UNMEASURED for the rest; done when the alpha and a mirrored winner are both scored on every declared axis, the mapping names the unmeasured axes, and a planted contrast regression lowers the design axis
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM26 — Journey driver measures the 1-2-3 path in the browser
Prerequisites: CM13 (acceptance unverified)
Contract: cd_journey_run
Acceptance: knowledge/journeys.conf declares each site's path (landing, action, completion) as steps the capture driver walks at both doors, recording clicks, fields, scroll depth and settle time per step and grading the receipt against the research rows (error prevention, visibility of status, checkout and form rows) [@nngroup-heuristics-2026-09] [@baymard-research-2026-09] [@builtformars-2026-09] [@pageflows-2026-09]; done when the gospelofwest and cms-scene alpha paths complete with a friction receipt per step and a planted extra step reads worse
Authored effort: 2.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM27 — Variant archive and selection in the shape of the Darwin and Mendel Godel Machines
Prerequisites: CM17 (acceptance unverified), CM26 (acceptance unverified)
Contract: sr_archive_select
Acceptance: Every emitted variant (blueprint, tokens, section order, presets) is archived content-addressed with its referee scores and journey receipt; the loop chooses the next parent by score with a novelty bonus and proposes a modification from the referees' named defects and the archive's comparative signals, admitting a child only when it beats its parent on every deterministic referee and on the journey receipt at a declared confidence [@arxiv-godel-machine-2026-09]; done when a planted defect is repaired within a declared iteration budget, a page that cannot converge is refused by name, and the archive keeps every variant
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM28 — Multivariate optimisation of the design knobs
Prerequisites: CM27 (acceptance unverified)
Contract: sb_knob_search
Acceptance: The knobs are the derived-token bands and the catalogue's discrete choices (section order, presets, hero mode, density); a factorial search over them is scored offline by the referee panel and the journey receipt, and online nx_cms_abtest assigns live traffic to the top variants with each journey step's completion logged and judged by Bayesian A/B arithmetic [@arxiv-ui-optimization-2026-09] [@arxiv-abtesting-web-2026-09]; done when the search prints its objective per variant, the chosen variant beats the incumbent at the declared confidence, and a knob outside its evidence band is refused
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM29 — Guarded self-improvement of the improver
Prerequisites: CM27 (acceptance unverified)
Contract: sb_policy_pin
Acceptance: The loop's proposal policy is a versioned knob set evolved the same way under a hash-pinned policy (goals, scope, evaluation identity, budget, release conditions) that the loop cannot edit, with the referees in another lane, harness tampering audited and the judge held to advisor [@arxiv-selfimproving-2026-09]; done when a mutation that edits a referee or the policy pin is refused by name and every accepted policy change carries its held-out gain
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM30 — Blind sides: Left and Right by a keyed coin, the click mapped back to the referee's A and B
Prerequisites: CM18 (acceptance unverified)
Contract: pjj_swapped
Acceptance: The pair section names no candidate; which candidate sits left is the parity of the pair's nibble of the session token, the click (L, R, tie) is mapped back to A, B, tie before the referee's one writer runs, and the side shown is recorded on a meta row; done when the gate proves a known-answer token swaps exactly the predicted pairs and a Left click on a swapped pair lands as B on the referee's row [@designarena-home-2026-09]
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM31 — Dwell recorded per verdict: the judge's time on the pair as gap evidence
Prerequisites: CM30 (acceptance unverified)
Contract: pjj_meta_row
Acceptance: Every verdict carries a meta row (pair, judge, date, side shown, dwell in ms from the pair's last render, door) written by one writer and read by one reader, the judge's median dwell printed on the surface; done when the gate measures a dwell at least as long as a real wait between render and post and UNSHOWN for a pair posted without a render [@arxiv-arena-preference-2026-09]
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM32 — Next pair by information: fewest judgments, then the smallest rating gap
Prerequisites: CM30 (acceptance unverified)
Contract: pjj_next_pair
Acceptance: The surface serves one pair at a time, the open pair whose candidates carry the fewest judgments and, among those, the smallest gap in the referee's Elo over every judged row, then the lowest ordinal; done when a known-answer row set makes the zero-gap pair of two winners the next pair where the first open pair is the wrong choice, and a tie moves no rating [@arxiv-2602-09411-ksort-eval]
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM33 — Standings on the surface: rating, W-L-T, win share with its Wilson interval
Prerequisites: CM32 (acceptance unverified)
Contract: pjj_standings
Acceptance: The referee's one rating ruler (nx_dr_elo from its start rating) over every judge's rows ranks the candidates; each row prints wins, losses, ties, the win share of decided pairs and its Wilson 95 interval from the gate base class, UNMEASURED with no decided pair, behind a disclosure until the judge is done; done when the table's rows equal the values the gate recomputes from the same rulers [@designarena-home-2026-09]
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM34 — Inter-judge agreement as Krippendorff alpha beside the pairwise agreement
Prerequisites: CM18 (acceptance unverified)
Contract: pj_kripp_alpha
Acceptance: The referee's receipt prints Krippendorff's alpha (nominal, over pairs judged by two or more judges) in permil beside the k-of-n agreement and its Wilson interval, in exact rational arithmetic; done when a hand-computed two-judge set over six pairs reproduces the textbook value and a single-judge set reads UNMEASURED by name [@arxiv-2511-20513-designpref]
Authored effort: 1. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM35 — Live preview beside the stills: the candidate served on the surface
Prerequisites: CM30 (acceptance unverified)
Contract: pjj_live_route
Acceptance: GET /live/<k>/ serves the candidate's own index and the assets under its dir on the loopback surface so the judge interacts before voting, the way the arenas judge running apps; done when the gate fetches a candidate's page and one asset through the route, a path outside the dir is refused by name, and the pair section links both previews [@arxiv-2609-15387-webcraftbench]
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM36 — Human-anchored machine co-judge with a gap-calibrated vote
Prerequisites: CM34 (acceptance unverified)
Contract: pj_cojudge_anchor
Acceptance: The referee's agreement with the human rows is published as a curve over the pair's score gap; the machine casts a vote only on pairs whose gap is above the smallest gap at which its agreement clears the floor, its sensitivity and bias per judge estimated from the rows; done when the curve is on the receipt, a pair below the gap reads ABSTAIN by name and the anchored ranking of a held-out set agrees with the human ranking above the floor [@arxiv-2609-19599-anchor] [@arxiv-2609-15387-webcraftbench]
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM37 — Per-axis verdicts: aesthetics, usability, requirement alignment and overall
Prerequisites: CM33 (acceptance unverified)
Contract: pj_axis_verdicts
Acceptance: The surface asks for an axis verdict beside the overall (a row per axis in the referee's grammar, the axis a declared field), the referee keeps one Elo table per axis and names the axis on which the human and the machine disagree most; done when a set judged on three axes yields three rankings and the receipt names the widest disagreement [@arxiv-2609-15387-webcraftbench]
Authored effort: 2. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM38 — Leaderboard robustness: leave-one-out and flip perturbation with rank intervals
Prerequisites: CM33 (acceptance unverified)
Contract: pj_loo_stability
Acceptance: The receipt prints how often the top candidate survives dropping each single row and flipping each single decided row, and the rank interval per candidate over those perturbations; done when a set decided by one row reads FRAGILE by name and a set decided by many reads STABLE [@arxiv-arena-preference-2026-09]
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM39 — Per-judge preference profiles
Prerequisites: CM34 (acceptance unverified)
Contract: pj_judge_profile
Acceptance: The receipt prints each judge's agreement with the pool and with the machine and the pairs on which the judge departs from the pool; done when a planted contrarian judge reads as the lowest-agreement row and the pool's alpha drops by name when that judge is added [@arxiv-2511-20513-designpref]
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
CM40 — Tournament session: four candidates from one brief, brackets to a final ranking
Prerequisites: CM32 (acceptance unverified)
Contract: pjj_tournament
Acceptance: A session draws four candidates for one brief and runs the bracket the arenas run (two battles, winners and losers, a tie-breaker, a final ranking) on the surface, each battle a pair row; done when a four-candidate session produces a full ranking in five verdicts and the ranking equals the referee's Elo over those rows [@designarena-home-2026-09]
Authored effort: 1.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.

Learning and practice paths

No structured learning path is declared for this plan. Existing research, roadmap and worklog remain available above.

Rung dossiers — every record that bears on a rung, in one place

Joined by symbol, never typed. Each dossier is assembled on every publish from the rows this board already carries: the plan's rung row, the matrix row that carries its contract symbol (its status re-measured by the one symbol ruler), the gate the organ declares and the verdict the evidence referee recorded on its last run, the reference marks in the rung text, the design rows that name the rung and the work log filed against it. The capability matrix links each cell to its dossier and each dossier links back to its cell and forward to the rungs it stands on; a dependency naming no rung on this plan is written as UNRESOLVED rather than linked. Absences are named: NO-MATRIX-ROW, NO-GATE-DECLARED, UNRECORDED.

Gate verdicts: UNSTAMPED — no per-gate verdict artifact is published for this domain yet (nx_swcompare_evidence cms writes it beside its stamp); every gate below reads UNRECORDED until it runs.

Gate declarations: knowledge/organ_gate.conf read from this working directory (1 rows) and from the estate root (196 rows); rows in both counted once: 0; declarations read: 197. Every declaration in either copy counts, so a stub beside the publisher can never hide the root.

CM0 Snapshot organ · cell PRESENT · gate NO-GATE-DECLARED

Contract	`snap_read_file` · executor Organ · 0 u · stands on nothing
Definition of done	Content snapshots plus one-step prev backups -- LANDED, listed so CM1 carries momentum
Matrix cell	Content snapshots and revisions · `runtime/_hdl_build/nx_cms_snapshot.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_cms_snapshot`; declare `nx_cms_snapshot TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

CM1 N-deep revisions with restore · cell PRESENT · gate NO-GATE-DECLARED

Contract	`snap_restore_n` · executor Organ · 1 u · stands on CM0
Definition of done	Restore any recorded snapshot; the pre-restore state is itself snapshotted first so every restore is reversible; gate proves restore-then-restore returns byte-identical content and a missing snapshot id is refused
Matrix cell	N-deep content revisions with restore · `runtime/_hdl_build/nx_cms_restore_lib.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_cms_restore_lib`; declare `nx_cms_restore_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-22 land Shipped 2026-09-22: nx_cms_restore_lib (snap_record, snap_list, snap_has, snap_restore_n over the NXSNAP1 archive lib, the site editor's own .v<id> revision naming and the store's atomic writer; refusals SR_RC_* by name) and nx_cms_restore_gate 38 of 38 GREEN on the NAS, promoted 154e0f1a, registered, rostered, in cms.gates and _offc; sources buildroot/runtime/_hdl_build on both trees; matrix symbol snap_restore_n flipped from its contract 2026-09-22 measure nx_cms_restore_gate on the NAS 2026-09-22: first_id 1790056177 and second_id 1790056178 (two records inside one second, the bump path), pre-restore id 1790056179, first archive 76 bytes for three files, ledger rows 5 after two records and three restores, ids listed 7 with two planted malformed archives; restore-then-restore byte-identical on a text file, a binary file with NUL CR LF and a high byte, and a markup file; a file of 8388609 bytes refused TOO-LARGE-FOR-PACKER with nothing recorded 2026-09-22 lesson two incumbents already held the halves of this rung (the NXSNAP1 archive one step deep, the site editor's N-deep .v<epoch> naming for its blueprint): composing them cost one lib and no second parser or naming scheme; the base packer reads each file into an 8 MiB reserve and would truncate a larger one in silence, so the new door measures every file with lseek END first and refuses by name rather than inheriting the silence; a refused restore is proven to write nothing by comparing the bytes after every refusal, not by reading the refusal code alone 2026-09-22 queue owed on this rung: wire snap_record and snap_restore_n into nx_cms_admin (the daemon still exposes the one-step snapshot and restore only; a revisions browse page in the shape of the site editor's history page is the restore UI WordPress has) and retire the base packer's silent truncation at its root by sizing the per-file read from the file; both are daemon deploys through the health-checked door, so they wait for a deploy window rather than a promote 2026-09-23 land FINISH ROW CLOSED -- snap_restore_n IS WIRED INTO THE SHIPPING PROGRAM: nx_cms_admin (buildroot/runtime/_hdl_build, both trees) imports nx_cms_restore_lib; ca_snap_names is the ONE list of the four site files the one-step archive and every revision pack; POST /admin/snapshot records revision <rid> beside snapshot.nxsnap (the 303 body names it, a refused revision is named and never fails the snapshot); GET /admin/revisions lists every recorded revision newest first with a guarded restore form each; POST /admin/restore/rev rev=<id> (session + CSRF, tested before the prefix-matching W4 route) restores through snap_restore_n, answers 303 with the prerestore id or a 409 naming the SR_RC refusal, and re-renders only after a successful unpack; the editor's Backup fieldset links the revisions page. PROOF ON THE WIRE: nx_cms_snapshot_gate extended with K7-K12 (revision file exists, listed, restore by id byte-exact, the undo of the undo, unknown id 409 by name with the live page untouched, unauthenticated 403), builds the daemon --build-only (a gate never twin-refreshes the serving root), D001-migrated onto nx_gate_verdict (every row a gv_check tooth, refusal rows neg-control-, legacy CMSGATE line restated from the base counters in the cms_gate.log dialect, gv_kv revision_id/prerestore_id, gv_verdict last): 13/13 GREEN on the NAS, SHIPPED 52fbc9f0 (loss d31dd1cc + review 24a30bae), organ_gate.conf nx_cms_admin -> nx_cms_snapshot_gate (the convention gate is a laptop dev fixture), cms.gates row. DAEMON SHIPPED through nx_organ_ship: PROVE by the declared gate 13/13, content GREEN, promoted 6dd75952 (prev 516115 B banked in the .prev slot), offc twin, registered. RELEASE STAGE: the promoted binary is READY; restarting any live nx_cms_admin instance onto it is the release owner's step (EC68), not this seat's 2026-09-23 queue the ranker's FINISH row (state=LIB-GATE-ONLY importers=1) lags the wiring: nx_compare_rank reads the importers census, which nx_importcensus derives from knowledge/store/ecograph_full, and the graph is rebuilt by nx_eco_graph_build on its daily beat (off this seat's tool surface); the new import edge nx_cms_admin -> nx_cms_restore_lib lands in the census on the next beat followed by nx_importcensus, and the row flips by itself -- no seat edit is owed, only the beat

CM2 Visual block canvas · cell LANDED · gate NO-GATE-DECLARED

Contract	`pb_canvas_emit` · executor Organ · 2.5 u · stands on nothing
Definition of done	A canvas emitted FROM the validated row model (pb_row_valid) with drag, resize and reorder; gate proves a canvas edit round-trips to a blueprint that re-validates and emits byte-identical pages when nothing moved
Matrix cell	Visual block canvas (drag, resize, reorder) · `runtime/_hdl_build/nx_cms_canvas_lib.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_cms_canvas_lib`; declare `nx_cms_canvas_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-23 land the visual block canvas as a library and its referee: runtime/_hdl_build/nx_cms_canvas_lib.nx (pb_canvas_emit + pbc_*: the validated 12-grid row model on nx_editstack_lib, gestures RESIZE/SWAP/MOVE conserve pb_row_valid by construction, canonical layout text round-trips byte-identical, page fragment carries {{key}} placeholders for crd_render, no-script gesture canvas, layout.txt + layout.ops persistence with head cursor, commit flattens) and nx_cms_canvas_gate 52/52 GREEN on the NAS, shipped ed6a8082 through nx_organ_ship (registered, cms.gates + tool_schemas rows); composed the incumbents the spend gate named (nx_editor_canvas's stack adoption, nx_cms_pagebuilder's ruler, nx_cms_store's atomic writer) rather than a client-side JS canvas the estate rejects; daemon wiring (GET /admin/canvas, POST /admin/canvas/op, {{layout}} splice before the one renderer, layout files in every snapshot) in flight on nx_cms_admin with K13-K18 wire teeth on nx_cms_snapshot_gate 2026-09-23 lesson the wire gate read 19/19 on its own ship and 17/19 one minute later on the daemon's ship: its /tmp fixture site persists between runs and the first run's commit had left layout.txt and layout.ops there, so the two teeth that begin from a site with NO layout met one (a gate that inherits its predecessor's fixture measures the predecessor); fixed at setup by moving both files aside exactly as the stale draft and snapshot already are -- the estate's idempotent-fixture law, paid for again; the magic ratchet also refused the first cut of the six teeth (typed reserves, the csrf length, row ids, a file mode) and every bound was derived instead: the capture reserve named once and read by the buffer line too, form bodies sized from their parts, the marker length read from the marker, MODE_0644 from the syscall layer, row ids from the base class counter 2026-09-23 land WIRED INTO THE SHIPPING CMS: nx_cms_admin SHIPPED 70ea37d7 (GET /admin/canvas = the no-script editor over the committed layout with its gesture log replayed, POST /admin/canvas/op = init / resize / swap / move / undo / redo / commit with every refusal the lib's own name and 403 without session or csrf, ca_render_publish splices the COMMITTED layout's fragment at the template's {{layout}} marker before the one renderer so the columns' {{key}} placeholders fill in the same pass, layout.txt + layout.ops ride every snapshot and revision); nx_cms_snapshot_gate SHIPPED a4954f17 as the daemon's declared referee, 19/19 GREEN over real HTTP (K13-K18: init form, init 303, canvas with bar and zero script, lone-column resize refused 409 by name, 403 without session, commit reaches the live page through crd_render); every bound derived, fixture idempotent; the live admin instance restart onto 70ea37d7 is the release owner's step (EC68), reported READY; the ranker's FINISH row for CM2 will read LIB-GATE-ONLY until the eco-graph beat and nx_importcensus see the nx_cms_admin import edge, as CM1's did 2026-09-23 queue the subject-level mutation bite of the canvas referee is owed and off this seat's tool window (nx_gate_bite registered, not surfaced): `nx_gate_bite nx_cms_canvas_gate buildroot/runtime/_hdl_build/nx_cms_canvas_lib.nx 2 aim pbc_apply` (in-process gate, no subject keyword needed) -- until it runs, the 52/52 is a gate that has fired on correct data plus its own planted-width bite, not a mutation-proven one; also owed: the live nx_cms_admin restart onto 70ea37d7 by the release owner (EC68), and the FINISH label flip after the eco-graph beat and nx_importcensus see the nx_cms_canvas_lib import edge

CM3 Organ-extension registry · cell LANDED · gate NO-GATE-DECLARED

Contract	`cp_ext_load` · executor Organ · 2 u · stands on nothing
Definition of done	An organ declares the hooks it serves in its header; the site builder loads it by name and refuses an undeclared hook; gate proves a declared extension renders into the page and an undeclared one is refused with the hook named
Matrix cell	Plugin and extension marketplace · `runtime/nx_cms_plugins.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_cms_plugins`; declare `nx_cms_plugins TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-23 land nx_cms_plugins (runtime/nx_cms_plugins.nx 17862 B sha 11d90d05, both trees byte-identical) + knowledge/cms_extensions.conf (timeout_ms row with its calibration; fail-closed with no ext rows yet) + nx_cms_plugins_gate 35/35 GREEN shipped a4beecb0 (promoted, _offc twin, registered): in-process teeth over a /tmp fixture whose extension binary is the gate itself (header csv, declared vs undeclared, registry allowlist, timeout row, fork only after both say yes, script and nonzero-exit refusals after the fork, splice inside main / before body / append / byte-for-byte with none), then nx_site_build built --build-only and forked on fixture blueprints (declared extension accepted and spliced inside its main landmark; undeclared hook and unregistered organ refused by name with no index written); nx_site_build wired (extroot and extension rows, cp_ext_load before any byte is written, cp_splice_write on all three tiers) and SHIPPED 922e5289 through nx_organ_ship: prove=GREEN by nx_cms_plugins_gate (organ_gate.conf declaration, closure bank CURRENT), adopt content=GREEN behave=IDENTICAL (WEAK: a usage-line probe; the behavioural evidence is the gate's end-to-end teeth), promoted + _offc twin, pre-change live ffc54653 317960 B banked by hash; contract symbol REPOINTED cp_plugin_load -> cp_ext_load on this rung and the matrix watch row (the lib's vocabulary is ext throughout, cp_ext_new/load/emit; a second name for the same loader would be a stub) -- the watch flips on the next cms regen; cms.gates + tool_schemas rows landed 2026-09-23 lesson the row-counter tooth passed n=51 for a 57-byte fixture on the gate's first ship (34/35): a hand-counted length beside a literal is a second copy of the literal's shape and drifts silently (rule 11's own law, paid for again inside a gate) -- the fixture is now one const and the tooth derives its length with pg_slen, so the count cannot disagree with the bytes; and nx_fs_write append refuses a row without its trailing newline (four rows bounced in one batch, file unchanged each time) -- end every row 2026-09-23 queue (1) the gate re-ship with the derived fixture length is job 1790143327 -- read prove= on its receipt; (2) nx_gate_bite nx_cms_plugins_gate buildroot/runtime/nx_cms_plugins.nx 2 subject nx_site_build is off this seat's surface, run it from a seat that has it; (3) the first REAL extension organ + its ext row in knowledge/cms_extensions.conf (the registry ships fail-closed and empty: no site can load an extension until an owner registers one, and the owner column names who); (4) a live site blueprint carrying extension rows goes through the release owner (EC68), never by hand; (5) FINISH labels clear on the eco-graph beat + nx_importcensus (LIB-UNCENSUSED is a lag, not a gap)

CM4 Showcase presentation tier · cell PRESENT · gate NO-GATE-DECLARED

Contract	`ed_showcase_css` · executor Organ · 1 u · stands on nothing
Definition of done	An image-led tier ON the one editorial system: full-bleed hero (image or seeded art), self-hosted display face, sticky nav with skip link, 4:5 image cards, dark band, footer; done when the incumbent site re-emits byte-identical and the craft ruler reads 15 of 15 on the emitted index (gospelofwest.com v3, 2026-09-21)
Matrix cell	Showcase presentation tier (image-led hero, display type, full-bleed, dark band) · `runtime/_hdl_build/nx_nishi_editorial.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_nishi_editorial`; declare `nx_nishi_editorial TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-21 land gospelofwest.com production v3 = tier showcase; nx_site_build promoted 81f8b82b, nx_nishi_editorial 0d5ffe7c; andelinwest.site byte-identical 7 of 7; page_verify GREEN; fonts served from the release with LICENSE 2026-09-21 land gospelofwest.com production v4 (footer headings h3, a11y heading-skip closed); nx_site_build promoted 69326cf0, _offc installed; apex and www both serve 27,156 B; www lifecycle row relabelled v4 2026-09-21 lesson Our own pages derive a FLAT scale (ratio 1.053, display 21 px): the ruler measured the kid work before any eye did; the emitter now refuses the showcase tier without a tokens file and every size on the page is a slot the derive filled, so a hand value has nowhere to live 2026-09-21 lesson A band that blends two kinds (text columns and page containers) derives a 608 px page from column widths; the ruler names the source values per row, so the band was corrected as DATA and the derive re-run -- the fix lives in the conf, never in a hand value 2026-09-21 lesson An extreme is not evidence: the derive now takes each source's most-used negative letter-spacing and the median across sources (-0.03 em, n=4), and the emitter applies it to display sizes only (h1, section h2) because tracking is an optical-size property in every measured source; and a median on an even count is the mean of the two middles -- the first derive took the upper one and read a 162 px display from sources reading 56 and 162

CM5 Design-craft ruler with file mode · cell PRESENT · gate UNRECORDED

Contract	`ua_kvn` · executor Organ · 1 u · stands on CM4
Definition of done	nx_ui_audit grades bytes on disk (a release before golive, a mirrored rival) with 15 presence points and raw counts ADDITIVE to the hygiene score; done when the same counts are published for our page and the nike.com mirror side by side on this board
Matrix cell	Per-page design-craft ruler, bytes on disk or live, rival-comparable · `runtime/_hdl_build/nx_ui_audit.nx` · PRESENT decl
Gate	`nx_ui_layout_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) `nx_ui_audit_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-21 measure nx_ui_audit live gospelofwest.com hygiene 12 of 12; file mode craft 15 of 15 on the showcase index vs 12 of 15 on the kit index (font_face 1 vs 0, uppercase 26 vs 1, letter_spacing 34 vs 3, aspect_ratio 1 vs 0, full_bleed 2 vs 0, token_defs 73 vs 28); nike.com mirror pending the NAS promote 2026-09-21 measure Awwwards September 2026 listing read from the mirror: 25 Site of the Day entries, WebGL named 16 times, GSAP 21 times; the ranker refused LADDER-PARTIAL NO-FRONTIER until this frontier target was declared 2026-09-21 measure nx_ui_audit promoted edb606b0 with file mode and the craft block; same ruler on three pages: gospelofwest v4 index craft 15 of 15 (img 0, svg 5, script 0, external src 0, font_face 1, reduced_motion_guard 2) -- nike.com home craft 6 of 15 (img 31, svg 115, alt 31, loading 31, aspect_ratio 40, script 53, external src 21, font_face 0 inline, skip_link 0, landmark_main 0) -- awwwards listing craft 9 of 15 (img 63, svg 104, script 11, external src 39); the honest gap the counts name is photography (CM7), the exceed the counts name is zero script and zero external assets 2026-09-21 measure EVIDENCE LOOP LIVE on the NAS: nx_ui_audit derive over four mirrors (nike, awwwards listing, apple, stripe) wrote gospelofwest.tokens -- body 18 px (n=2), consecutive-size ratio 1.066 (n=1), display 162 px (n=1, the listing's largest fixed size), spacing base 12 px and section 48 px (n=2, step 1.319), line-height 1.50 body / 1.20 display, breakpoints 640 and 1068 px (n=4), content width 1255 px (n=2), tracking -0.15 em (n=1), motion 800 ms, radius 24 px; apple and stripe contributed only breakpoints because their CSS is external (extract reads bytes on disk), so their stylesheets are the next sources 2026-09-21 measure tokens re-derived from 8 sources after the measure_px band was split from text columns (900..1600 px page containers): body 14 px n=6, ratio 1.214 n=4, t6 36.9 px, display 162 px n=2, spacing 12 px base n=4 and 48 px sections n=2 (step 1.319), line-height 1.40 body n=5 and 1.25 display n=2, breakpoints 640 and 1068 px n=8, page 1200 px n=3, tracking -0.15 em n=4, motion 800 ms n=4, radius 8 px n=6; v5 re-emitted from these tokens into releases/v5 and previewed at https://gospelofwest.com/v5 through a symlink under the live v4 release (no golive, no browser popup) 2026-09-21 retract the v5 preview at gospelofwest.com/v5 is RETRACTED and its symlink under the live v4 release removed (404 now): the operator's eye read it as squished and a mess, and a public path on the production host is not the alpha/beta/production methodology -- a candidate reviews on the alpha plane (nishifamily.com/alpha/SITE/), stabilises on /beta/SITE/ with a browser check, and reaches production only on the operator's eye 2026-09-21 measure headless Edge captures of v5 and v4 at 1440 px (operator-permitted, no window): every v5 heading letter-crushed -- the trk token was the single MOST NEGATIVE letter-spacing any source set (-0.15 em, n=4) applied to every heading down to the 17 px wordmark and 21 px card titles 2026-09-21 measure ruler extended with two evidence rows: fam (display face class by majority of the sources naming a generic family: sans, n=3) and dch (median h1 length inside the h1_chars band 8..60, n=2); the emitter picks the self-hosted face by class, ships it with its OFL LICENSE into RELEASE/fonts, and fits the display size per page as td x dch / len so a longer headline holds the sources' line count 2026-09-21 retract the phone-overflow reading of the first 390 px captures is RETRACTED: headless Edge and Chrome clamp a 390 px window to 504 and 512 px (innerWidth probe) and crop the shot, so the overflow was the capture's, and a 390 px iframe control is refused by the edge (frame-ancestors); phone rendering stays UNMEASURED until a capture driver sets device metrics 2026-09-21 land v6 emitted from gospelofwest_v6.tokens (8 mirrors, true median; nx_ui_audit promoted 62741fa8, nx_site_build 2ff7ba05): body 14 px n=6, ratio 1.178 n=4, display 109 px n=2 fitted to 93 px for the 35-char headline (dch 30 n=2), spacing 10 px base n=4 and 44 px sections n=2, line-height 1.40 n=5 and 1.22 n=2, breakpoints 640 and 1046 px n=8, page 1200 px n=3, tracking -0.02 em n=4 on display only, face sans n=3 self-hosted with its LICENSE; craft 14 of 15 (img 0), hygiene 12 of 12, page_verify GREEN a11y 0; served for review at https://nishifamily.com/alpha/gospelofwest (the release bytes copied into the alpha dir, sha 954bf52e identical to releases/v6/index.html; the ad strip below the footer is the nishifamily edge's own, not the release's); headless desktop capture read by the seat: no crushed glyphs, a two-line sans headline, coherent index -- an interpretation, not a grade; phone UNMEASURED; LIVE stays v4 until the operator's eye, then /beta/gospelofwest with the worklog on this board, then golive 2026-09-22 measure research leg 2026-09-21: two arXiv listings (webdesign 98 results, aesthetics 440), 8 abs pages and 8 vendor and roundup pages mirrored through nx_research_fetch and read from the bytes; 20 refs pinned by sha256; Relume app, the Awwwards annual page and the Framer press release refused as mirrors (JS shells, status 0); envato and figma trend pages mirrored but not yet read, so not cited 2026-09-23 land nx_ui_audit given its execution surface: UI-MAP rows now end with ts=<epoch> (both trees; SHIPPED dcb15fac via nx_organ_ship, prove GREEN by the organ_gate.conf declaration nx_ui_audit->nx_ui_layout_gate, contentdiff GREEN, behaviour IDENTICAL over file sites/gospelofwestcom/current/index.html); clock row uiauditshowcase 86400 declared over that release page (desired plane, folds live at the scheduler's next window) so knowledge/status/ui_debt_map.log becomes a time spine; the ranker's REGISTERED-DARK label clears once the live clock plane carries the row 2026-10-01 measure October attestation of the board's eight bars (the NAS clock crossed into October and nx_compare_rank refused BAR-UNATTESTED): nx_frontier_scan attest cms read the five arXiv listings and appended five barscan rows (fetches 136, unattestable 3: the Awwwards listing, the unite.ai roundup and the Awwwards evaluation page carry no dated source row); the seat re-fetched and read those three (sha 8a0318d6, a831a117, 85ae70e6) and adjudicated the two listings the attester flagged NEWER: SB-WEBCRAFT moves to 2026-10 (Colonia Zacamil, Oct 01, 2026 Site of the Day), SB-RUBRICS to 2026-10 (living page unchanged), SB-UIOPT to 2026-03 (fatigue-aware VR interface layout optimisation 2603.26031, missed by the September reading), SB-IA stays 2026-08 (every September entry a homonym, named on the barscan), SB-BUILDERS attested at 2026-09 (the roundup is still September's); the attester reported CLASSES UNDECLARED for cms, so a cls row for the board is the mechanism that stops a homonym setting its bar month next time 2026-10-01 lesson a bar month is a measurement with a clock on it: the ruler derives the current month from the NAS clock in UTC, so a board attested on the last day of a month is unattested the next morning, and the attester cannot judge subject without the board's class rows (cls

CM6 Rendered-pixel design referee · cell LANDED · gate NO-GATE-DECLARED

Contract	`pj_referee` · executor Organ · 3 u · stands on CM5, CM13
Definition of done	A sovereign render of the emitted page judged on pixels (overlap, contrast as painted, hierarchy); done when the referee grades the live gospelofwest.com index and disagrees with at least one planted defect the bytes ruler passes
Matrix cell	Rendered-pixel design referee on emitted pages · `runtime/nx_page_pixjudge_lib.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_page_pixjudge_lib`; declare `nx_page_pixjudge_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-21 queue beauty is UNMEASURED by any estate referee today; nx_browser is a window organ and the seat may not pop a browser; the pixel referee is the contract 2026-09-21 lesson A hygiene ruler that reads 12 of 12 on both the kit page and the showcase page discriminates nothing; the craft axis and the rival's raw counts beside ours are what separate them, and beauty still waits on the pixel referee 2026-09-21 queue A tokens file with n=1 rows is thin evidence and says so per row; widen the corpus with the winners' stylesheets and Site of the Day pages before the pixel referee grades the render 2026-09-21 queue capture driver: a sovereign CDP client over the estate's websocket client that sets device metrics and captures a real browser at phone and desktop widths -- the oracle stays outside the build path and the pixel referee reads its output 2026-09-22 lesson the estate already carries nx_ui_judge, an Awwwards-axis rubric judge over the ship manifest with an 800 permil competency floor, registered with no cap ever minted: a dark incumbent the referee rung extends rather than twins 2026-09-23 queue dependency edge DECLARED on the rung (deps CM5,CM13): the referee reads the capture driver's real-browser pixels (the oracle outside the build path) -- the sovereign renderer nx_browser_render is a 99 KB lib with one mention of grid, so a sovereign render of the showcase tier would grade a picture no browser paints; the ranker's DEP-PULL now prices CM13 ahead of CM6 instead of the seat overriding the order 2026-09-23 measure referee on the live gospelofwest.com index captured by the motion-settled driver: phone door, the eyebrow p (11 px, white over the hero photograph) paints 3.00:1 against the 4.5 bar; desktop door 3.64:1; the h1 paints 5.07 and 5.78 against the large bar 3.0; every other text box passes and unmeasured=0 on both doors; the ONE defect on both doors is the eyebrow, and no bytes ruler can see it: a photograph is not a token pair 2026-09-23 measure nx_uiq_contrast (the bytes ruler, run on the NAS through /api/organ_run) reads the same page FAIL fails=6 on --ink x --ink-soft 2.06, --ink-faint x --ink-soft 1.67, --accent x --ink-soft 1.28 in both themes: pairs a name heuristic makes (a muted-text token read as a surface) that never paint as text on a surface; the two rulers disagree in both directions and only the pixels adjudicate 2026-09-23 lesson A SHOT TAKEN BEFORE THE PAGE'S TIME-DRIVEN MOTION SETTLES READS AS UNPAINTED TEXT: the first phone capture caught the hero mid-reveal (ink=0 on the paragraph, 158 px on the h1, the CTA painted entirely in its own bg) and the referee reported two contrast defects that were the instrument's; the driver now waits for finite document-timeline animations (scroll-driven timelines and infinite iterations excluded, -1 at settle_ms), the receipt carries motion_running, and the referee refuses an UNSETTLED capture by name rather than grade it 2026-09-23 lesson the second method class for a contrast kernel is the WCAG definition worked by hand beside it (white on rgb 200 = 167, on 153 = 285, on 17 = 1888): the kernel truncates (284 for 2.849) so the gate holds those to +-1 and 167 exactly; a KAT computed by the kernel under test would have proved transcription 2026-09-23 land nx_page_pixjudge SHIPPED sha c38adac8ed4442e0ededec6fca4c4d5942af8c14c11a1ba00865808f1cfa0ad1 (nx_organ_ship: PROVE nx_page_pixjudge_gate 30/30 GREEN on the NAS through the organ_gate.conf declaration, the five browser legs announced E2E-NOT-RUN-HERE; first promote; offc twin; REGISTERED) -- sources on both trees byte-identical: nx_page_pixjudge_lib.nx 6303c29c 22471 B, nx_page_pixjudge.nx e978e1c4 9007 B, nx_page_pixjudge_gate.nx 73078451 29860 B, knowledge/pixjudge.conf (bars as data with WCAG 2.2 citations, ink tolerance and floor calibrated on the fixture, the e2e fixture rows); the laptop gate reads 35/35 with the real browser (phone and desktop captures of the fixture page graded DEFECTS contrast_fails=2 unmeasured=0); the matrix row repointed to the lib that declares pj_referee 2026-09-23 land CM13 extended for the referee: nx_cdp_capture writes <label>.boxes (the DOM box map, CD-BOXES line, nboxes on the receipt) and waits for the page's finite time-driven motion before each shot (scroll-driven timelines and infinite iterations excluded, -1 UNSETTLED at settle_ms, motion_running as the receipt's 15th field); nx_cdp_lib cdv_boxes_rows + the receipt row; nx_cdp_capture_gate 35/35 on the laptop (15-field round trip, every-door-settled reads exactly 0 on the still fixture); the re-ship carries one enumerated loss, the old usage sentence superseded by the one naming the three appended fields 2026-09-23 queue EMITTER: the showcase eyebrow (11 px white over the hero photograph) paints 3.00:1 phone / 3.64:1 desktop against the 4.5 bar on the live gospelofwest.com index -- the referee's one live finding; the remedy is in nx_site_build's hero (a scrim behind the eyebrow, or a heavier eyebrow with its own paper), proven by re-running the referee on a fresh capture, never by loosening the bar 2026-09-23 queue BYTES RULER: nx_uiq_color's uiq_is_surface reads any token whose name carries soft as a SURFACE, so --ink-soft (a muted text colour) is paired as paper and the page reads FAIL 6 on pairs that never paint; the referee is the adjudicator today; the heuristic wants a declared role per token (data), not a longer name list 2026-09-23 queue nx_gate_bite over nx_page_pixjudge_gate with subject nx_page_pixjudge (and the CM13 gate with subject nx_cdp_capture) is off this seat's surface; the planted controls bite in-process (grey-on-grey ruler control, hero contrast, containment, 2 px brush) and the mutation run is owed from a seat that carries nx_gate_bite 2026-09-25 measure referee on the live gospelofwest alpha (CM13 capture, phone and desktop): desktop DEFECTS contrast_fails=2, both the 12 px mono eyebrows (hero eyebrow and section kicker) painting 3.21 and 3.94 to 1 against 4.5; phone DEFECTS contrast_fails=3 overlap_pairs=5, where two fails are the same eyebrows (3.76 and 4.11) and one fail plus the 5 overlaps are hidden navigation links the referee still measures (declared 14.74 to 1, painted 1.00, OWN-PAPER); by hand the ink-faint on ground pair is about 4.26 to 1, so the eyebrow fails before anti-aliasing 2026-09-25 queue REFEREE: skip boxes that are not painted -- hidden navigation links read as contrast fails and overlaps on the phone door on both alphas; the capture's box map should carry rendered visibility and the referee should grade only painted boxes, with a neg-control that a visible link over its own paper is still measured 2026-09-25 queue EMITTER: the showcase eyebrow and kicker ink (ink-faint on ground, about 4.26 to 1 declared, 3.2 to 4.1 painted) fails the 4.5 bar on both alphas, the no-photograph case of the eyebrow row above; the remedy is a derived eyebrow ink (the lightest palette ink that meets the bar against its ground in each theme), proven by a fresh referee capture, never by loosening the bar 2026-09-25 land the referee now grades only PAINTED boxes: nx_cdp_capture records the browser's own Element.checkVisibility() as a 14th box field (1 painted, 0 not, -1 cannot say; nx_cdp_lib CDV_BOX_FIELDS 14), and nx_page_pixjudge_lib gives an unpainted box its own role so contrast, overlap, over-image and hierarchy all skip it, counted as unpainted with boxes = text + img + unpainted in PJ-GRADE; a zeroed slot reads UNKNOWN and is graded as before, so an older 13-field map and every hand-built gate table stay judged, never silently hidden. nx_page_pixjudge_gate 44 of 44 on the laptop (38 of 38 in-process on the NAS, SHIPPED 51ea8b1e) with bites both ways and an e2e closed-menu link reading unpainted on both doors; nx_cdp_capture_gate 35 of 35 2026-09-25 measure the live gospelofwest alpha re-captured with the fresh driver and re-graded: phone boxes=18 = text 13 + img 0 + unpainted 5 (the closed menu sheet's four links and its call to action, painted=0 in the map), contrast_fails 3 to 2 and overlap_pairs 5 to 0; desktop unchanged at contrast_fails=2 (the menu is not rendered there). The two fails left on both doors are the 12 px eyebrow and kicker -- the real defect, queued above as the derived eyebrow ink 2026-09-25 lesson asking a skipped subtree for its geometry FORCES its layout: a closed details sheet answers getBoundingClientRect with real boxes while painting nothing, so a referee that trusts geometry grades invisible text. The browser's own checkVisibility is the witness; the capture records it and the referee decides, and the value that means unknown is the one a zeroed slot holds 2026-09-25 land the editorial palette's faint ink is DERIVED against the referee's own bar (source on the NAS, promote held for a sibling site seat's window): nx_ink_derive_lib walks the declared ink along its hue until its WORST ratio over ground, ground-sunk and card meets knowledge/pixjudge.conf contrast_min_ratio100; nx_site_build arms it for the editorial and showcase tiers and refuses by name when the row cannot be read or met; unarmed callers write the old bytes. Light #6E7269 -> #656860 in 79 permil steps (worst 4.52 on ground-sunk, 4.91 on ground, 5.28 on card, each worked by hand; one step back reads 4.47), dark #8A8C84 unchanged (4.85). nx_ink_derive_gate 39 of 39 on the laptop incl. fixture roots for every refusal and an AAA row that derives a darker ink 2026-09-25 measure A/B on one emitted page where only the faint ink differs, captured in the browser and graded: old #6E7269 (declared 4.26 on ground) paints 4.11 on the phone door and 3.82 on desktop, both fails; new #656860 (declared 4.91) passes the phone door (worst 4.75, OK) and paints 4.34 on desktop, a fail at 4.5. The painted estimate runs 0.88 to 0.90 of the declared ratio for 12 px mono at 1x and about 0.97 at 2x 2026-09-25 lesson a pixel referee that averages every pixel within its ink tolerance measures anti-aliasing along with colour: on thin strokes at 1x most ink pixels are partial-coverage blends, so a WCAG-compliant ink reads as a fail. The ink_tolerance row was calibrated on a 16 px sans face whose cores sit within 12 of the declared colour; 12 px mono at 1x is a second corpus it was never tested on. Fix the estimator, never darken the palette to fit the instrument's bias 2026-09-25 queue REFEREE ESTIMATOR: estimate the painted ink from the pixels nearest the declared colour (the glyph cores) and the paper from its dominant colour rather than both means, calibrated on the A/B pair above (declared 4.26 and 4.91 must read within tolerance of declared) while the true positive, the production eyebrow over the hero photograph, still fails; read how the field's pixel-based checkers pick foreground and background before choosing 2026-09-25 land the referee reads the PAPER the way the field reads pixels and the INK the way WCAG evaluates contrast: the Android Accessibility Test Framework's ContrastSwatch (exact-colour histogram, split at the unweighted mean luminance of the distinct colours, the modal colour each side) finds the paper mode under the text [atf-contrastswatch] [atf-contrastutils], and the declared text colour is held against it, because WCAG's Understanding note on 1.4.3 says to evaluate the colours from the user agent rather than the anti-aliased text as presented [wcag-understanding-1-4-3]. The exact method runs only where the pixels prove its picture -- some pixel is exactly the declared colour, and every pixel is that colour, the paper mode or a per-channel blend of the two, no threshold -- and everything else keeps the band means, whose bias only lowers a ratio; the field's own two-mode ratio is printed as atf100 and never judged, as the framework itself reports it only as a warning [atf-textcontrastcheck]. nx_page_pixjudge SHIPPED 65b3f864 (lib 02c668fd), gate SHIPPED ee72d68c at 50 of 50 on the NAS and 57 of 57 on the laptop with the browser leg, where every box of the solid-backgrounded fixture is judged exact on both doors 2026-09-25 measure eight captures graded with the production bars, measured = exact + band summing on every run: the ink fixture with the derived #656860 reads OK on both doors, worst 4.91, 9 of 9 and 8 of 8 boxes exact; the same page with the old #6E7269 fails both doors at exactly 4.25, its declared ratio; the production gospelofwest home keeps its one true defect, the eyebrow over the hero photograph, 3.64 desktop and 3.00 phone by the band means (paper mixed: 3,333 and 6,241 unexplained pixels); the alpha still carrying the old ink reads its two faint lines at 4.25 on both doors, and its twelve pre-reveal card texts stay UNMEASURED (no pixel of their declared colour, the reveals had not run). Over every solid-paper box of three pages the band means had read 0.67 to 0.98 of the declared ratio and the field's two-mode ratio as low as 0.31, because on thin text its foreground mode is an edge blend 2026-09-25 retract row 1790354901 read the derived ink as failing the desktop door at 4.34 against 4.5 and queued an estimator change: the number was the band means' anti-aliasing bias, not the page -- the same capture paints its glyph cores exactly #656860 (29 pixels) on exactly #EFEFE9, and the exact method reads 4.91 on both doors. The estimator change it queued is the land row above 2026-09-25 lesson a fixture whose foreign colour sat outside the ink-paper interval on green and blue only let a mutant that disabled the red-channel check survive (nx_gate_bite aim pj_box_swatch site 82): three stripes, each outside on exactly one channel, killed it, and the mutants at the unexplained-count site were killed too. A per-channel rule needs a per-channel fixture, asserted as the fixture's own precondition before any outcome 2026-09-25 queue a SCANNABILITY and WAYFINDING axis for the referee, read from the box map it already loads (longest block, list share, first-record offset, rail presence, current-item marking), reported beside contrast and graded against measured benchmark pages -- the axis the bytes ruler cannot have 2026-09-25 land THE DERIVED FAINT INK IS LIVE: nx_site_build e6a81423 SHIPPED through the ship loop. The 4 lost runs are the static palette literals, now assembled around the derived token (allow_loss efc73062); the behaviour change was reviewed from the outputs, not the counts (review 6a557794, evidence knowledge/status/harness/evidence-nx_site_build-1790372800.txt): on the gospelofwest showcase workload the candidate's stdout is the live stdout plus the INK receipt line (nx_srcdiff B-SUPERSET, 55 lines kept, 1 added), and each of the 5 emitted pages equals the live page with #6E7269 swapped for #656860, byte for byte. nx_ink_derive_gate f721b47f 39 of 39 on the NAS, closure banked, on the SLOW roster by measurement (2,999 ms; its trial at the fast roster's 1,800 ms bound was killed at 1,848 ms). Both alphas rebuilt through their recipes: gospelofwest 12 of 12, cms-scene 24 of 24, page_verify GREEN on both 2026-09-25 measure both rebuilt alphas captured with the text-free shot and graded by the production bars: gospelofwest OK on both doors, contrast_fails 0, worst 4.91, 14 of 26 desktop and 10 of 18 phone boxes read by the paper method (the same alpha with the old ink read its two faint lines at 4.25 and failed, row 1790358501); cms-scene OK on both doors, worst 4.91, 11 of 13 desktop and 9 of 15 phone boxes measured. Still UNMEASURED on the gospelofwest desktop door: 12 boxes, the three feature cards at the fold (y 733 to 890 of 900, four text boxes each) caught mid-reveal by the stagger preset, no pixel of their declared ink 2026-09-25 lesson an emitter's behaviour change is reviewed from the outputs, and the proof that nothing else moved is a substitution identity: capture live and candidate on the ship probe's own argv (the candidate through a scratch row on its stage file, reaped by nx_toolreg_reap once the promote consumed it), line-diff stdout with nx_srcdiff, and show each page equals the live page with the one derived token swapped -- a byte count says the sizes match, only the identity says the change is that token and nothing else. And a gate's roster is decided by its measured runtime (nx_gate_roster_run trial, then admit), never by where the last gate was hand-appended: the two gates hand-added earlier today measured 1,203 and 702 ms and stay on the fast roster, while this one would have read TIMEOUT on every fast beat 2026-09-25 queue MEASURE THE RESTING STATE OF SCROLL-REVEALED TEXT: the capture shoots the first paint and settles only finite document-timeline motion, so text that a scroll-driven preset reveals at the fold is UNMEASURED on every run (12 boxes on the gospelofwest desktop door). Done when the capture declares a reduced-motion door (the page's own guard disables every preset under prefers-reduced-motion) or drives each box through its view range, and those boxes read a ratio 2026-09-26 queue W11 connect visual-review seat, a referee FALSE POSITIVE on real data, classified (a first-draft hit is a bucket, not a finding): on the Connect alpha account page (desktop door) nx_page_pixjudge graded box div 190,99,1060,676 -- the page wrap, a text box only because a markup defect leaves a 54-character clause as a direct text node in it -- DEFECTS at painted 2.84 with letters=225 and exact100=1518: the F83 letter search ran over every glyph pixel inside the wrap's box, children included, and graded the wrap's declared ink #182031 against a child's paper, the Sign-out button's #2f5fd0 (2.85 by hand; the button's own white text on that blue paints 5.33 and passes). A container text box must be graded on its own text runs only, the pixels of descendant boxes the map already grades excluded; bucket name for the census: container-over-descendants. Fixture for the tooth: a div holding one bare text node beside a child button with a declared ink and a solid dark background. Until then a DEFECTS on a box whose letters far exceed its textlen is a question, not a verdict. 2026-09-26 land W13 SEAT ALPHA-UI-AND-PROD (connect lane, debt 1790438899) extended nx_ink_derive_lib ADDITIVELY (buildroot/runtime/_hdl_build, 6821 to 9358 B, sha256 aefcd17b): inkd_conf_bar reads the referee's bar row from conf text (the key followed by the field separator at a line start; the value an unsigned decimal ending at the next field, a CR or the line end; a comment line and a longer key sharing the prefix are not the row) and returns the value or a named refusal INKD_E_ROW or INKD_E_VALUE, both named by inkd_code_name; inkd_meet and every existing function are unchanged. First consumer: nx_connect_appd, which derives the Connect app's --faint in both themes over --bg --surface --surface-2 at startup (light 8b94a6 to 686F7C, dark 6a7488 to 80899A); proven by nx_connect_ui_gate T13 (six teeth incl. neg-controls, 39/39 on a fresh compile, SHIPPED 1734504731e0aded) and the live alpha pixel referee (the footer 471 where it read 284). HONEST LIMIT: the nx_ink_derive_gate re-run after the edit read 39/39 on a BUILD-CACHE HIT (key 0f93a9af, job 1790456373), so it is evidence about a cached artifact, not a fresh compile of the edited lib; two --no-cache re-runs were deferred by build admission (load1 17.15 and 16.90 over the 16.43 ceiling, jobs 1790456542 and 1790456894). The edited closure was compiled fresh by the connect ui gate and the connect daemon, which exercise inkd_meet inkd_worst inkd_walk inkd_hex inkd_hex6 inkd_ratio and inkd_conf_bar. QUEUE for the cms owner: nx_site_build sb_ink_arm still reads the same row through sb_cg sb_fld sb_uint, so two readers of one bar row now exist; migrating sb_ink_arm onto inkd_conf_bar makes it one (not done here: nx_site_build is the cms lane's file), then re-run nx_ink_derive_gate with --no-cache. 2026-09-26 land THE SCALE-STEP AXIS: pj_hierarchy convicts only a text box LARGER than the h1, so a display sized at the section heading read OK (measured: a planted variant with td at t6 painted h1 32 px beside h2 32 px, hierarchy_violations=0). nx_page_pixjudge_lib pj_hierarchy_step counts the non-h1 text boxes the h1 does not lead by one step of the type scale (h1 x 1000 below font x hierarchy_min_ratio_permil), disjoint from the strict axis, UNMEASURED without an h1 or a bar; the bar 1178 permil is knowledge/pixjudge.conf's new row, calibrated from the derived profile's own ratio row (n=4); the grade line carries hierarchy_step_short, a short box flags HIER-STEP and counts toward DEFECTS. Two abstentions beside it: TEXTFREE-MISS (glyph cores painted while the text-free shot removed nothing under them) and NO-CORE (ink inside the tolerance band with no pixel at the declared colour: a scroll-driven reveal mid-fade), each UNMEASURED, flagged and counted on its own axis, never a manufactured verdict. Gate nx_page_pixjudge_gate 88 of 88 on the laptop, 81 of 81 on the NAS (0286aafd): planted h2 at the h1 size flips the verdict on the new axis alone, an h2 one step below is the positive control, agreeing shots and a faded run abstain and a lone abstaining box reads UNPROVEN 2026-09-26 measure the new axis on real pages before the emitter fix: the loop fixture at the phone door read h1 35 px beside h2 32 px (one short, 35000 below 32 x 1178) and the LIVE gospelofwest alpha at the phone door read h1 32 beside h2 32 (one short); both desktops read 0 short (109 and 93 px displays). The showcase clamp bottomed at t6, so every phone-door display sat within a step of its section headings by construction, and the fitter sent every long headline there on purpose 2026-09-26 land THE DISPLAY FLOOR: nx_nishi_editorial ed_td_floor puts the clamp's low end and the fitter's floor one step of the scale above t6 (t6 over t5, read off the derived tokens), in whole pixels and whole centirem rounded up (a floor carried at centipixel precision and printed through the two-decimal rem writer lost a pixel at each end and still read one short: 37 px against 37.7); nx_site_build 3569418a SHIPPED (nx_cms_plugins_gate 35 of 35, nx_ink_derive_gate 39 of 39 rebuilt and shipped 79348f42, stdout IDENTICAL on the gospelofwest workload), the gospelofwest alpha rebuilt in place (DISPLAY index td derived 10900 fitted 9342); the loop fixture now reads 0 short at both doors (phone h1 38 px) 2026-09-26 measure the served gospelofwest alpha after the display floor, captured from the laptop and graded by the shipped referee 28dcc838: desktop h1 93 px, phone h1 38 px, hierarchy_step_short 0 at both doors, contrast_fails 0, worst ratio 4.91, textfree_miss 0, no_core 0, verdict OK at both doors; the declared journey (land, Start reading, first theme card) OK at both doors at 2 actions. Two lane facts measured on the way: the alpha is SERVED from the published copies the release recipe's nx_site_publish rows write under nishifamily/alpha/gospelofwest, so a rebuild of sites/gospelofwestcom/alpha changes nothing served until its five pages are published (done by hand as an alpha-stage direct rollout, the recipe's own rows 3 to 11; nx_plan_run answers managed-owned-definition-required for the recipe plane); and the nx_site_build MCP row emitted the pre-floor clamp (1.98rem) twice while the promoted root binary emitted the floor (2.38rem) into a /tmp probe -- the MCP row forks a copy the ship's promote and offc install did not refresh (the third-twin class); the served pages were rebuilt through /api/organ_run of the promoted root and published

CM7 Licensed imagery pipeline · cell LANDED · gate NO-GATE-DECLARED

Contract	`sm_admit_image` · executor Organ · 1.5 u · stands on CM4
Definition of done	heroimg and card images admitted only with a licence row beside them; done when an unlicensed path is refused by name and a licensed one renders in the hero
Matrix cell	Licensed imagery pipeline for site heroes and cards · `runtime/_hdl_build/nx_site_media_lib.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_site_media_lib`; declare `nx_site_media_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-22 land SHIPPED, referee grade: gate GREEN 46 of 46 on the NAS and the laptop, awaiting operator acceptance. nx_site_media_lib (sm_admit_image): a hero or card image the blueprint names is admitted only with an image row beside it (image, path, licence id, source url, origin) and a SHIP_OK verdict from the estate's ONE provenance ledger (nx_asset_prov_lib over knowledge/provenance/assets.jrnl and the rights table knowledge/model_license.conf), the file hashed in the release, the ledger row written once per sha and licence (a relicence is a new row, never an edit), every refusal naming the path and the reason (no row, unreadable file, malformed row, unknown licence, non-redistributable licence, conditional or unverified licence which the ledger calls REVIEW and a person settles); nx_site_build admits the hero and every card image before emission and takes a medialedger row as the ledger door; the rights table gained two estate classes as data, client-owned and estate-generated, beside the existing cc0-1.0, each verified because the class is the estate's own design decision. Promoted: nx_site_media_gate fc6c235f 106,776 B registered, rostered, in _offc and declared in cms.gates; nx_site_build 16bedaaf 304,152 B (contentdiff GREEN lost 0 gained 113 over five added sources) and in _offc 2026-09-22 lesson the incumbent ledger writes the claim row before the verdict judges it (the nx_asset_prov order), so a refused licence still leaves one journal row per sha and licence and the last row wins: a cc0 re-admit after three refused claims recovers the image, and the gate asserts that recovery rather than a clean journal; and a CC-BY image stays unadmitted by design until a credit renderer exists, because the rights table marks attribution CONDITIONAL and a conditional right is a person's decision, never the emitter's 2026-09-22 queue the credit renderer for attribution licences (a rendered source and licence line in the colophon and a cc-by-4.0 row added only after its legal code is mirrored and read) is the next imagery data row; the imagery ladder the site program declared stands (public domain, then cc0, then the estate's own render, then attribution last), and a real photograph on gospelofwest waits on the client's own media under client-owned 2026-09-22 measure done-rule met: an unlicensed path is refused by name (the forked promoted emitter on a showcase blueprint with heroimg media/hero.bin and no image row exits 1 printing MEDIA REFUSED image=media/hero.bin no licence row) and a licensed one renders in the hero (the same blueprint with image rows exits 0, the receipt admits the hero and the card, the index carries figure class heroimg with src media/hero.bin at offset 21,010 and the scratch ledger carries the cc0 row); the admission gate reads 46 of 46 on the laptop and on the NAS; the byte-identity control: gospelofwest_v7.site emitted by the promoted emitter and by the banked previous one (nx_site_build_prev, the CM11 build) gives index.html sha 954bf52e 24,192 B on both, the same bytes as the shipped v6 release, so a blueprint without images is untouched by this rung

CM8 3D scene section rendered by the sovereign engines · cell LANDED · gate NO-GATE-DECLARED

Contract	`ed_sc_scene` · executor Organ · 3 u · stands on CM4
Definition of done	A hero or band section that hosts a sovereign WASM or WebGPU scene from the estate's own engines behind a CSS-only fallback door, emitted from a blueprint key; done when the tier emits it, the page still grades 15 of 15 on bytes with zero third-party script, and the scene door announces which tier it took
Matrix cell	3D scene section hosted by the sovereign engines (zero third-party script) · `runtime/_hdl_build/nx_nishi_scene.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_nishi_scene`; declare `nx_nishi_scene TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-23 land scene section chain shipped on the NAS: nx_wasm_vm memory.size and memory.grow with nx_wasm_vm_memgrow_gate 094d3f53 21 of 21; nx_nishi_scene_gate 0da88eb5 96 of 96; nx_site_build c692a0f9 then the mount build; nx_ui_audit 5972a972 (external script, never a self-contained first-party one, costs the 15th point; ui_layout_gate 37 of 37); module knowledge/scene/nx_scene_hero.wasm d71d6e68 57281 B on both trees; catalog knowledge/scene_catalog.conf 52ec6401 2026-09-23 measure done-rule on production bytes: gospelofwest v8 re-emitted by the promoted emitter is byte-identical to its release, 7 of 7 generated files; the same blueprint plus one scene row grades 14 of 15 like the control 14 of 15, every craft count equal or higher, script_ext 0; the missing point on both is imagery because knowledge/provenance/assets.jrnl holds 7 rows and none is an image 2026-09-23 measure the full-score clause is now a gate tooth: both fixture pages carry the engine's own native frame as an estate-generated hero under a scratch media ledger, base 15 of 15 and scene 15 of 15, each external script shape 14; bite: the scene page without its hero row reads 14 and the gate goes RED 93 of 96 while the base tooth holds 2026-09-23 measure first alpha through plan-site-release-cmsscene- (24 rows: 9 registry rows by nx_store_put put, draft or asset, build, audit, 3 prepare-path, 9 publishes index last, page_verify), 24 of 24 ok, page_verify GREEN; but the host router injected an estate ad aside and a sendBeacon view script (served 35363 B for a 34368 B release) because the page lacked the router's nx-ad-optout marker; gospelofwest.com production carries no ad, its alpha preview carries both 2026-09-23 land the ads row (value optout): nx_site_ads_lib writes the router's own marker as a body comment right after ed_doc_open (that opener closes the head itself, so a meta there would land in the body); refused by name for another value or a tier that does not write it; nx_site_ads_gate 945a412c 21 of 21 reads the page with the router's own hr_has_optout, bite killed 1 of 2 (the survivor only silences the lib's stdout); after the re-run the served index equals the release, 0 ad slots, 1 script 2026-09-23 measure in a real browser (CM13, Edge, phone 390 and desktop 1440) the door read css-only refused sovereign-cpu:failed:error: http 404; the capture profile's cache held /alpha/scene/orbit.wasm and /alpha/fonts/LiberationSans-Bold.ttf for a page at /alpha/cms-scene/: the edge's clean-URL form (hr_canon strips a directory's trailing slash) serves the index at /alpha/cms-scene, so every relative href on it resolves against /alpha/; the module, the display face and the nav were broken on every alpha and beta preview index, gospelofwest's included; nx_page_verify stayed GREEN (no browser, and its relative-asset law reads src attributes only) 2026-09-23 land a MOUNTED release: the emitter derives the mount from the path part of its domain argument (nishifamily.com/alpha/cms-scene gives /alpha/cms-scene/, sb_mount) and switches its one prefix pair to it, so every href, the face url() and the module data-href are root-relative from the mount; a bare domain keeps every byte; nx_site_mount_gate 20 of 20 on the laptop resolves every reference against the landing URL the router's own hr_canon computes: mounted index 17 refs 0 escapes, mounted card page 18 refs 0 escapes, the bare release served under the mount 17 of 17 escape (the neg-control is the defect itself) 2026-09-23 measure CM6 referee on the capture: desktop DEFECTS contrast_fails=1 overlap 0 hierarchy 0 (the 12 px eyebrow at 3.22 to 1, the showcase token defect CM6 named on gospelofwest); phone contrast_fails=2 overlap_pairs=6, where the 6 overlaps and one contrast fail are the collapsed menu's hidden links, whose boxes the capture records and the referee measures over the heading 2026-09-23 lesson a done-rule that names a score must name the base it starts from: production showcase pages read 14 of 15 because no image can earn the imagery point, so the clause was untestable there until the gate built its own full-score base 2026-09-23 lesson a preview is the release only if the edge serves it the way the release was written: relative hrefs on a directory index and a canonical form that strips the directory's slash cannot both hold, and a static verifier with no browser passes a page whose face and module both 404 2026-09-23 lesson a behaviour probe over a gate that prints its per-run fixture directory reads DIFFERS between two runs of the same binary: run the live binary twice, mask the run token, then locate the probe's offset inside it before writing the review 2026-09-23 queue gospelofwest's alpha recipe builds a mounted release (domain nishifamily.com/alpha/gospelofwest into its own release dir) and carries the ads row (value optout), so the preview stops loading no face and showing an estate ad its production never serves; the same for every alpha or beta built by nx_site_build 2026-09-23 queue nx_page_verify's relative-asset law also reads CSS url() and script-fetched data-href on directory pages, so it stops passing a page whose face and module resolve against the parent 2026-09-23 queue the capture box map carries rendered visibility and the pixel referee skips boxes that were not painted, so a collapsed menu's links stop reading as overlaps 2026-09-23 queue the scene loader's refusal names the URL it tried, so a 404 names its path instead of costing an investigation 2026-09-23 queue poster frame for the css-only door: the scene's first frame rendered natively in the site's palette for both themes through nx_png_write_rgb, admitted as estate-generated, so the door shows a picture and the page earns the imagery point 2026-09-23 queue WebGPU and WebGL2 passes in the module; frame time measured by CM13 in a browser (frame_ms 40 is provisional); the browser's digest compared with native 753114568; module memory sized from its framebuffer (365 pages today) 2026-09-23 queue workstreams.conf binds site-release to one site's plan (site-release-gospelofwest); a second site needs a per-site plan id rather than a second row with the same key 2026-09-23 measure in a real browser after the mounted release (CM13, Edge, one 1440 by 1700 frame): the band takes its canvas column and the sovereign CPU pass draws the core and three moons; the door line reads door sovereign-cpu refused webgpu-compute:failed:pass-absent-in-module, webgl2:failed:pass-absent-in-module, sovereign-cpu:taken threads=1 module-single-thread; the served index equals the release, 1 script, 0 ad slots, 17 root-relative references 2026-09-23 measure CM6 referee on that frame: DEFECTS contrast_fails=2 overlap_pairs=0 text_over_image=0 hierarchy_violations=0; both fails are the small uppercase eyebrows (3.22 and 3.80 to 1 against 4.5), the showcase token defect CM6 already named, not the scene; the scene band itself draws no finding 2026-09-23 land shipped to alpha at https://nishifamily.com/alpha/cms-scene/ (release sites/cmsscene/releases/v1, recipe plan-site-release-cmsscene-), referee grade DEFECTS on the eyebrow contrast only, awaiting acceptance; gates rostered: nx_nishi_scene_gate, nx_wasm_vm_memgrow_gate, nx_site_ads_gate, nx_site_mount_gate (nx_site_mount_gate b57e93b9 20 of 20 on the NAS) 2026-09-23 queue the scene reads nearly monochrome because all four bodies take neutral roles (ground, accent-wash, rule, card), and the CPU pass draws the core's rim without anti-aliasing; both are visible at 1440 and both are the next quality rungs for the scene, to be judged by the referee panel and the operator, not by the seat 2026-09-25 measure the gospelofwest alpha at nishifamily.com/alpha/gospelofwest was an indexable duplicate of the live client site: nishifamily.com robots.txt disallows /preview/ and not /alpha/, and no alpha page carried a robots meta; its release recipe also built for the production domain, so its relative links resolved against /alpha/ under the edge's slash strip, and the router injected an estate ad and a view beacon into a page whose own copy says no ads and no trackers 2026-09-25 land nx_site_stage_lib (51a757e0): a release STAGE is declared on the build line by its recipe; stage=alpha or stage=beta writes a robots noindex meta inside every page head on every tier (ed_doc_open for editorial and showcase, sk_head for the kit, the CM15 static-switch pattern, no syscalls so the kit keeps its contract); stage=production and no token are the bytes the emitter always wrote; a misspelt value, an empty one or two tokens refuse the build before any page exists. nx_site_build d62bb0fc SHIPPED with one enumerated loss (the kit's head-close literal, split in two so the meta can sit between); nx_site_stage_gate 7d341970 47 of 47 on the NAS, registered 2026-09-25 measure controls for the stage change: andelinwest (kit tier) re-emitted 7 of 7 byte-identical without a stage on the laptop; the pre-ads gospelofwest v8 blueprint (showcase) re-emitted 7 of 7 byte-identical to releases/v8 on the NAS by the promoted emitter; the ship probe on the kit workload read IDENTICAL; every armed head and page in the gate is exactly the 38-byte meta longer than its plain twin 2026-09-25 land gospelofwest alpha rebuilt through its own recipe (plan-site-release-gospelofwest-, putmany of rows 1 to 11): gospelofwest_v8.site gains the ads optout row, the build goes to sites/gospelofwestcom/alpha (production serves a different release, current is not releases/v8), domain nishifamily.com/alpha/gospelofwest, stage=alpha; 12 of 12 steps ok, page_verify GREEN; served: the noindex meta inside the head, the opt-out marker, zero beacons, 19 links mounted under /alpha/gospelofwest/. The cms-scene recipe row 10 declares stage=alpha too: 24 of 24 steps ok, the published index carries the meta once 2026-09-25 lesson a release built for one host and published under another's path is two defects wearing one symptom: its links resolve against the wrong directory, and its content is indexed as a duplicate of the host it was built for. An alpha is a preview, and the recipe that builds it must SAY so, because nothing downstream can tell a preview from a mounted production site by its path 2026-09-25 queue EDGE: rung E6 filed on edge.plan with its matrix row (directory indexes keep their trailing slash, the root remedy; debt 1790351077 is its recurrence row beside the still-open 1785865143); the router-side X-Robots-Tag for preview prefixes the emitter does not build (today /alpha/beach.html) is debt 1790351101 for the sites daemon's owner 2026-09-25 measure nx_site_stage_gate bite (nx_gate_bite over nx_site_stage_lib, 2 sites): valid_mutants=1 killed=1, restore byte-identical, both twins restored; rostered in knowledge/gateroster.conf. Gates re-shipped because this leg moved their closures: nx_cms_plugins_gate d645f836, nx_nishi_scene_gate b40b55b9 (behaviour review: the live gate differs from itself, its fixture directory is named from the clock) 2026-09-25 measure DECLARED, NOT PROMOTED: the stage lib moved the import closure of every nx_sclass_kit and nx_nishi_editorial importer, so these served binaries read behind their source until their own lanes rebuild them -- nx_connect_page, nx_connect_lds_page, nx_connect_reconnect_page, nx_connect_youth_page, nx_fin_stack_page, nx_dataeng_compare_page, nx_site_archetype and the meal pages. Their emitted bytes are unchanged (the unarmed switch writes nothing: 7 of 7 byte-identical on the kit tier, 7 of 7 on the showcase tier), so a rebuild is a no-op on output. Re-shipped this leg because they are this lane's: nx_site_build, nx_site_stage_gate, nx_cms_plugins_gate, nx_nishi_scene_gate, nx_site_agentready_gate, nx_page_pixjudge, nx_page_pixjudge_gate, nx_cdp_capture (through /api/promote with the loss named: the ship loop needs a successful representative probe and this organ needs a browser the NAS does not have), nx_cdp_capture_gate

CM9 Brand and IA profile selects corpus and recipe · cell PRESENT · gate NO-GATE-DECLARED

Contract	`sb_brand_profile` · executor Organ · 1.5 u · stands on CM5
Definition of done	A brand profile (industry, personality, palette seed) in the .site blueprint selects the evidence corpus the tokens derive from and the section recipe; done when two profiles emit measurably different token sets from different corpora and the incumbent site re-emits byte-identical without one
Matrix cell	Brand profile names the EVIDENCE the design derives from (corpus and recipe), persisted in the blueprint · `runtime/_hdl_build/nx_site_brand_lib.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_site_brand_lib`; declare `nx_site_brand_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-22 queue the ladder to SOTA declared as data: M6 builder parity from evidence (CM9 to CM15) then M7 the research frontier as product (CM16 to CM18), every rung with its role toward a dated target; the ranker prices it after this beat 2026-09-22 lesson a future contract pointed at an EXISTING organ reads DARK-CONTRACT on this board (the 09-06 mechanism cannot tell a capability nobody built from one built under another name), so every open contract now points at an absent lib file (nx_site_brand_lib, nx_ui_corpus_lib, nx_site_catalog_lib, nx_nishi_motion_lib, nx_nishi_layout_lib, nx_site_refine_lib, nx_ui_tokenref_lib, nx_site_agentready_lib, and the 08-19 rows nx_cms_restore_lib, nx_cms_canvas_lib); the refs gate's provenance check refused 10 rows whose mirror was fetched under a different url than the row cited (a paper row may not cite a listing as its mirror; an arXiv feed re-fetched under the row url keeps its byte count and changes its sha because the feed embeds a per-request id), all 10 repaired and the gate reads cms clean except the acquitted Framer benign hop 2026-09-22 land SHIPPED, referee grade: gate GREEN 29 of 29 on the NAS, awaiting operator acceptance. nx_site_brand_lib (sb_brand_profile): a profile row (industry and personality) selects the corpus and recipe from knowledge/design_corpora.conf (award-product 8 mirrors, builder-marketing 5 mirrors, both recipe showcase) and forks the ONE ruler nx_ui_audit derive --corpus over it through nx_tool_run, writing profile.corpus and profile.tokens beside the pages; an explicit tokens row wins and says so. Promoted on the NAS: nx_ui_audit 523aab7e 642,528 B (derive --corpus, contentdiff GREEN lost 0), nx_site_build efd1bb76 226,488 B (contentdiff lost 1 = the reworded no-tokens refusal, named), nx_site_brand_gate a2c7cf83 88,360 B registered, rostered and declared in cms.gates; _offc installed for both organs; sources pushed through the content door with CAS 2026-09-22 measure done-rule met: two personalities on two planted corpora derive body 1600 vs 1400 and display 9600 vs 16000 centi-px, the same corpus derives the same sizes twice, seven refusals named (unknown or empty personality, empty industry, absent mirror, unimplemented recipe, missing recipe row, unreadable conf); the incumbent path is untouched: andelinwest.site emitted by the promoted binary and by the banked previous binary hash identically on all 7 files (index 13,289 B d36748f2, sitemap 573 B, robots 68 B, four practice pages 12,105 B each), and the showcase tokens 2026-09-22 lesson a corpus whose pages carry no content width derives a tokens file the emitter refuses (mpx is a required row): the first smoke refused on it, so the gate now asserts the derived file carries every emitter-required row, a derive that succeeds into a refusal being the vacuous half of the contract; and the live sites/andelinwest docroot is not this emitter's output (7,164 B index against 13,289 B emitted, hand-kept contact, legal, schedule and llms.txt), so the honest control was LIVE versus the banked .prev binary on the same blueprint, never the live folder 2026-09-22 measure REAL CORPUS ON THE NAS: gospelofwest_v7.site (the v6 blueprint with the profile row faith-community, award-product in place of its tokens row) emitted through /api/organ_run derives profile.tokens byte-identical to gospelofwest_v6.tokens (sha e289f005, 2,331 B, 8 sources, t1 1400 n=6, td 10900 n=2, mpx 120000 n=3) and an index.html byte-identical to the shipped releases/v6/index.html (sha 954bf52e, 24,192 B): the profile path reproduces the hand-listed derive exactly, from data 2026-09-22 queue OWED: the scratch registration nx_site_build_prev (elf nx_site_build.elf.prev, registered for the A/B) has no unregister route (/api/tools/unregister answers 404), so it must be reaped before the next nx_site_build promote rotates .prev under it; a tools/unregister verb is an engineshift gap 2026-09-25 land a third personality in knowledge/design_corpora.conf, added by the connect SITE seat and used by the Connect alpha front door (nishifamily.com/alpha/connect-faith): social-trust = signal.org, hinge.co and meetup.com home pages plus the stylesheets they link (8 corpus rows, recipe showcase, additive only, mirrored 2026-09-25 by nx_research_fetch); derived n=4 over the stylesheets: t1 1600 ratio 1101 td 6200 s1 1200 s6 4400 lhb 150 bp1 67250; the lesson it carries: for a rival-class corpus the home pages alone measure almost nothing (every modern home ships its type in linked CSS), so each corpus row set must include the linked stylesheets or the derive reads n=0 on type and spacing

CM10 Evidence corpus refreshed from the month's award listings · cell LANDED · gate UNRECORDED

Contract	`crf_refresh` · executor Organ · 1.5 u · stands on CM5
Definition of done	A beat mirrors the month's Site of the Day and Sites of the Year pages with their stylesheets through nx_research_fetch, re-derives the tokens and journals every n that moved; done when a re-derive after a refresh changes at least one token with its n and the journal names the source
Matrix cell	Evidence corpus refreshed from the month's award listings (winners' pages and stylesheets), tokens re-derived with n per row · `runtime/_hdl_build/nx_corpus_refresh.nx` · LANDED decl
Gate	`nx_corpus_refresh_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-22 queue the corpora conf is the substrate CM10 refreshes: a beat that mirrors the month's Site of the Day and Sites of the Year pages with their stylesheets through nx_research_fetch, appends corpus rows, re-derives and journals every n that moved 2026-09-26 queue the listings the corpus refresh will read beside the Awwwards Site of the Day page: the Ecommerce Design Awards log [eda-home-2026-09], Siteinspire's daily showcase [siteinspire-2026-09] and Land-book's landing-page gallery [landbook-2026-09], each mirrored this month; a listing row names the page and the beat mirrors the winners it names with their stylesheets, so the three design philosophies the field converges on (immersive 3D storytelling, precision SaaS minimalism, accessible consumer clarity) become personalities in design_corpora.conf derived from named evidence, never typed 2026-09-27 land THE CORPUS REFRESH BEAT, SHIPPED: nx_corpus_refresh (entry crf_refresh; decisions in nx_corpus_refresh_lib, prefix crf_) re-mirrors every listing 2026-09-27 measure DONE-RULE MET ON THE NAS, two live runs: run 1 (epoch 1790541180) mirrored both Awwwards listings (604733 and 509177 B, zero static stylesheet links on either), derived over 10 sources and read BASELINE (no previous derive); run 2 (epoch 1790541232) re-mirrored them, derived over 12 and read REFRESHED moved=26 -- every moved row names the two dated listing mirrors as its sources (t1 1500 to 1700 n 8 to 10, ratio 1103 to 1064, t0 1359 to 1597, bp1 62000 to 60000, trk -9 to -15, and every n up by two) -- which is the mechanism working as declared and ALSO a defect: the second copies were byte-identical to the first (52 s apart) and doubled every n; the beat now refuses unchanged bytes by sha, the four run-2 rows were retracted in the conf with the reason beside them and the token state restored to run 1's derive (the run-2 mirror files stay in knowledge/fetched as fetched). Awwwards listings are script-rendered (zero /sites/ anchors on the served page, measured 2026-09-27), so no winners rule is declared for award-product; the listing pages' own inline styles are what the derive reads from them 2026-09-27 lesson a beat that re-mirrors the same page every run must dedupe by CONTENT, not by url: the url is the same by design (the month's listing) and only the bytes tell whether new evidence arrived; the laptop's _offc fetcher is an older generation that prints SAVED without a PROVENANCE row, so the beat reads the provenance row first and the SAVED line second and hashes the mirror itself when the fetcher did not; a while-read loop that feeds the door client eats the loop's own stdin (one file of three pushed, caught by the push log), so the client runs with stdin closed; the board contract symbol must be a declared function (main is not one): the driver became crf_refresh before the matrix cell could light 2026-09-27 queue owed: winners rules for listings that carry static winner links (Siteinspire, Land-book and the Ecommerce Design Awards log, once each is inspected and a prefix declared), a listing whose stylesheets are external (the winners' own pages) exercising the stylesheet and per-page-budget legs live, the beat's first clock-driven run in the journal (the row is in the desired plane; the scheduler folds it at its window init), and the derived-token consumers (the brand emitter derives on demand from the conf rows, so a refreshed corpus reaches every emitted site at its next build without a further step -- to be witnessed on the next alpha build) 2026-09-27 measure run 3 (epoch 1790541786, the beat's first MCP invocation under its own cap, on the promoted dedupe build a003223d): both listings had CHANGED since run 1 (604980 against 604733 B and 509037 against 509177 B, new shas), so same=0, both admitted as new evidence, derive over 12 sources, REFRESHED moved=26 naming the two dated mirrors; the SAME-BYTES refusal is therefore proven on the gate's fixtures and NOT yet witnessed live (an unchanged listing between two beats is the case it exists for, and Awwwards moved between every pair of runs today); the four run-2 rows stay retracted, so the corpus holds 12 sources: 8 hand-mirrored plus the two dated listing copies of run 1 and run 3

CM11 Section catalog with a picker · cell LANDED · gate NO-GATE-DECLARED

Contract	`sa_catalog_pick` · executor Organ · 2.5 u · stands on CM9
Definition of done	A declared section catalog (each entry with its measured tokens and done-rule) and a picker that composes a page from the brief, refusing a section whose inputs the brief lacks; done when the catalog holds at least the 28 crawled patterns as pickable entries and a brief with a missing input is refused by name
Matrix cell	Section catalog wide enough to choose from: catalog-conditioned generation binding brief data to declared section components · `runtime/_hdl_build/nx_site_catalog_lib.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_site_catalog_lib`; declare `nx_site_catalog_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-22 land SHIPPED, referee grade: gate GREEN 64 of 64 on the NAS (65 of 65 on the laptop, where the template sources live), awaiting operator acceptance. The section catalog as DATA: knowledge/site_sections.conf declares each of the 12 crawled section tokens per tier (renderer or shell or none, required brief keys, token slots, done-rule); knowledge/site_catalog.conf holds the 28 crawled patterns as pickable entries, generated by nx_site_catalog build from the 30 template sources through the nx_tpl_ingest extractors, keys content-addressed as the corpus and the published v- pages, rows sorted, no timestamp, so a rebuild is byte-identical; nx_site_catalog_lib sa_catalog_pick composes a page from the brief by id, archetype or auto and refuses by name (a missing input names the section and the key, no renderer names the section, unknown id, malformed row, undeclared token), the last refusal kept as state a gate reads; nx_site_build reads the pattern and catalog rows and the showcase index follows the picked seq (hero leads even when a pattern lacks it, sub-pages keep their shape), byte-identical without a row. Promoted: nx_site_catalog 64ebc295 78,128 B registered and in _offc, nx_site_catalog_gate a8f29abb 103,144 B registered, rostered, in _offc and declared in cms.gates, nx_site_build c36b9fe4 257,744 B (contentdiff GREEN lost 0 gained 90) and in _offc 2026-09-22 measure done-rule met: the live catalog carries 28 pattern rows from 30 sources (dups 2, skipped 0, partition sums) and on the NAS all 28 ids are the published sites/nishifamily/factory variation pages (28 of 28); a brief with a missing input is refused by name (section proof missing step; section ft missing footer; section price missing plan in a tier that renders pricing); the emitter proves order in bytes: the default index puts the services section at 21,034 before the band at 21,521, the proof-first pattern puts the band at 21,035 before the cards at 21,279; the real gospelofwest brief with pattern auto picks v-ecommerce-5595255323 (hd>nav>hero>cta>ft, 2 sections) after 28 considered, because 19 of 28 patterns carry a form, gallery, video, pricing or faq section the showcase tier cannot render yet and the covered renderable set ties at two sections; 9 of 28 patterns are fully renderable by the showcase tier today 2026-09-22 lesson the corpus seg-store the 28 patterns were extracted into (knowledge/index/tplcorpus) survives on neither tree, but the 30 template sources on the laptop and the 28 published pages on the NAS do, and the keys are content-addressed, so the catalog was rebuilt from the sources and cross-checked against the published names instead of the store (a derived store can be lost while its inputs and its receipts remain); and a gate that resolves an organ from a root ladder must put the local build dir before the _offc twin on the laptop, where _offc is a fossil store: a July nx_site_build.elf there read the kit index and failed two teeth until the ladder was ordered serving root, then _build, then _offc 2026-09-22 queue auto composition on a real brief is thin by construction until the showcase tier renders form, gallery, video, pricing and faq sections (19 of 28 patterns need one); the tie-break among covered patterns is catalog order, and an industry-to-archetype row on the brand profile is the next data row so a faith-community brief prefers a landing pattern over an ecommerce one; a tools unregister verb is still owed for the scratch nx_site_build_prev row, which now points at the CM15 binary after this promote rotated .prev

CM12 Motion presets as data, zero script · cell LANDED · gate NO-GATE-DECLARED

Contract	`ed_motion_preset` · executor Organ · 1.5 u · stands on CM4
Definition of done	A preset table (reveal, parallax, scroll-scrub, sticky, view transitions) applied per section through CSS scroll-driven animation and view transitions; done when the showcase page carries at least three distinct presets, grades 15 of 15 on the craft ruler with zero script, and the reduced-motion guard still disables every one
Matrix cell	Motion presets: scroll-driven, hover and reveal choreography chosen per section, zero script · `runtime/_hdl_build/nx_nishi_motion_lib.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_nishi_motion_lib`; declare `nx_nishi_motion_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-22 land SHIPPED, referee grade: gate GREEN 51 of 51 on the NAS and the laptop, awaiting operator acceptance. knowledge/motion_presets.conf declares six presets as data (reveal, stagger, parallax, scrub, sticky, view-transition) and one selector per section; nx_nishi_motion_lib (ed_motion_preset) emits each preset over the derived tokens from the conf alone: a view preset is wrapped in a supports test for animation-timeline view() with its keyframes and a reduced-motion guard, a layout preset pins with no guard, the page preset opts the document into view transitions with its duration on the motion token; five refusals name the preset, the section, the unreadable conf, the malformed row and a template without a selector; nx_site_build reads motion rows (motion, section, preset) from the blueprint and emits one style block after the showcase CSS on the index and every sub-page, and a blueprint without motion rows emits no preset. Promoted: nx_nishi_motion_gate 079b27af 77,560 B registered, rostered, in _offc and declared in cms.gates; nx_site_build ffc54653 317,960 B (contentdiff GREEN lost 0 gained 37) and in _offc; nx_site_catalog_gate a8a6b1d9 64 of 64 and nx_site_media_gate afc81b8d 46 of 46 rebuilt on the shared fixture lib nx_site_fixture_lib and re-promoted 2026-09-22 measure done-rule met: the fixture showcase page with five motion rows carries five distinct presets on the index and on the sub-page (index_distinct_presets 5 of live_presets_declared 6), the promoted craft ruler reads 15 of 15 with script_tags 0, and the reduced-motion guard covers every preset and the built-in motion (index_reduced_motion_guards 7); the plain control without motion rows carries no mv- keyframes and no view-transition rule; an unknown preset makes the emitter exit 1 naming the preset; the gate reads 51 of 51 on the laptop and on the NAS 2026-09-22 lesson a feature block behind a switch makes the switch-off run a neg-control of the whole gate, not only of the new teeth: the first fixture read 14 of 15 because the imagery point wants an img and the fixture had none, so the motion fixture now carries a licensed hero through the CM7 door, and the three site gates share one fixture lib because a fixture spelled in three gates drifts three ways 2026-09-22 measure byte-identity control: gospelofwest_v7.site emitted by the promoted emitter (ffc54653) and by the banked previous one (nx_site_build_prev, the CM7 build) into /tmp/gw_cm12_live and /tmp/gw_cm12_prev gives index.html sha 954bf52e 24,192 B on both, the same bytes as the shipped v6 release, so a blueprint without motion rows is untouched by this rung; /compare/cms republished 290,851 B (regen GREEN, forkfresh 4 of 4), page_verify GREEN a11y 0; the scratch nx_site_build_prev registration now points at the CM7 binary after this promote rotated .prev 2026-09-22 measure 2026-09-22 CM12 and CM15 are on the gospelofwest alpha (v8 blueprint: motion page view-transition, feat stagger, proof parallax, cta reveal; agentready Organization): the index carries 3 mv- keyframe blocks and itemscope 2, page_verify GREEN a11y 0, published through the owner's plane plan-site-release-gospelofwest-; CM11 pattern auto deliberately not applied (it picks a five-section pattern that drops feat and proof until the industry to archetype map lands); CM7 waits on a licensed family photo 2026-09-22 lesson a capability shipped on the board is not on the product until the release lane republishes it, and a lane a seat runs by hand has no owner and no recipe: the lane is now the publisher's plane with a RACI, rerunnable by any seat or agent through nx_plan_run site-release-gospelofwest or nx_orchestrate_raci site-release run

CM13 Cross-device capture driver · cell LANDED · gate UNRECORDED

Contract	`cd_capture` · executor Organ · 2 u · stands on nothing
Definition of done	A sovereign CDP client over the estate websocket client that sets device metrics and captures a real browser at declared phone and desktop widths, the pixels handed to the referee; done when a 390 px capture reports innerWidth 390 and the phone render of gospelofwest is graded, ending UNMEASURED
Matrix cell	Cross-device render verification (a real browser at phone and desktop widths, the captures the referee reads) · `runtime/nx_cdp_capture.nx` · LANDED decl
Gate	`nx_cdp_capture_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-22 queue next rung by rank: the capture driver, a sovereign CDP client over nx_ws_client that sets device metrics on an outside headless browser (the oracle stays outside the build path) and hands 390 px and 1440 px pixels to the referee; its grading half waits on CM6, and CM6 waits on grid layout in nx_browser_render (the sovereign renderer lays out box, flex and float over real TTF and has no grid, so it cannot lay the showcase page out yet), declared here as a browser-board dependency rather than solved as a cms rung; the ranker's near-tie between CM6, CM13 and CM12 was a missing signal, not an order 2026-09-23 measure laptop gate nx_cdp_capture_gate 32/32 GREEN (in-process over nx_cdp_lib, then the subject end to end on the fixture page: phone innerWidth=390 ihdr=780x1688 png=16281 B, desktop innerWidth=1440 ihdr=1440x900; the absent-browser neg-control refuses rc=3 with no receipt); live https://gospelofwest.com/ on the final binary: phone door innerWidth=390 ihdr=780x1688 811532 B, desktop innerWidth=1440 ihdr=1440x900 906323 B, devtools answered in 1023 ms, settled in 353 ms, verdict=OK -- the 2026-09-22 retraction (headless clamps a 390 px WINDOW to 504) is closed because Emulation.setDeviceMetricsOverride sets the VIEWPORT; the phone render is captured and UNMEASURED until the CM6 pixel referee grades it 2026-09-23 land nx_cdp_capture (the wire half: bounded loopback connect via nx_connect, the INCUMBENT upgrade nx_websocket_client_upgrade, frames via nx_websocket_frame, launch and reap of the host browser, one navigation per capture, per door Emulation.setDeviceMetricsOverride + double-rAF innerWidth + Page.captureScreenshot, receipt rows) + nx_cdp_lib (conf rows, every DevTools reply through the estate's JSON DOM nx_value_parse_json, headers via nx_http_header_find, image dims via nx_img_dims, RFC 6455 length arithmetic, the metrics rule, the receipt row) + nx_cdp_capture_gate + knowledge/cdp_capture.conf (the oracle candidates, the doors and every budget with its calibration, as data); laptop build sha b2a06615 (112334 B), gate 9e448a5e (96471 B); pushed to the NAS through the content door: lib cbdd2008 20413 B, driver 52e59b7b 35561 B, gate dc632324 22979 B, conf 6309cb84 2617 B at the root knowledge/; organ_gate.conf declares nx_cdp_capture -> nx_cdp_capture_gate; the DOM's string-loop bound derived from the source length on both hosts (nx_value_parse_json sha de5b0c97 16655 B) 2026-09-23 lesson A FIXTURE THAT HAND-PLACES TWO BYTES OF A THREE-BYTE VALUE TESTS THE FIXTURE, NOT THE LIB: the 64-bit frame-length tooth read RED because 70000 (0x011170) was written into two bytes and the parser honestly read 4464; every length byte is now derived from the value and a tooth asserts the fixtures genuinely need the 16- and 64-bit forms. AN INCUMBENT'S OWN CAP BIT THE PAYLOAD CLASS: nx_value_parse_json's string loop stopped after 1,048,576 iterations and returned the truncated span with a clean verdict, while a phone screenshot reply carries 1,123,404 base64 bytes -- the bound is now the source length (a string cannot outrun its source), nx_json_complete_gate 10/10 on the new source, and this gate carries a 1,123,404-byte reply as a tooth so the cap cannot return in silence (nx_value_parse_json_test cannot run here: its GBIF fixture file is absent on the laptop, exit 20 before and after the change). Incumbents were found only by asking nx_spendgate PER HELPER (json, image dims, header find); the organ-level query named none. NishiLang out-params are mmap'd cells handed by pointer; the laptop build lane roots at nxc2/runtime 2026-09-23 queue NAS ship through nx_organ_ship (its PROVE reads SKIP on the NAS by name: no browser row of the conf resolves there, so the laptop is the proving host and its 32/32 is the evidence); the cms.gates row for the gate; CM6 pj_referee over the capture receipts (the phone render of gospelofwest is UNMEASURED until the pixel referee grades it); a NAS-side browser row when a headless oracle exists there; nx_cdp_eval (graphics GR53, the incumbent CDP witness this driver composed its upgrade lane from) is laptop-only and ABSENT on the NAS -- a flow-back row for its owner; the gate's bite run (nx_gate_bite is off this seat's surface) 2026-09-23 lesson A GATE WHOSE END-TO-END WITNESS IS STRUCTURALLY ABSENT ON THE PUBLISHING HOST MUST ANNOUNCE THAT LEG, NEVER gv_need IT: the first NAS ship refused exit 21 because the gate's verdict read SKIP there (no browser row of the conf resolves on the NAS) and the ship loop reads any nonzero gate exit as a refusal; the witness is now optional and announced (E2E-NOT-RUN-HERE reason=no-browser-row-of-the-conf-resolves-on-this-host, e2e_ran=0 among the values), the in-process teeth carry the NAS verdict and the laptop run carries the end-to-end leg (33/33 GREEN, e2e_ran=1) -- the 2026-09-02 law applied; gate source now sha 5daef5be 23696 B on both hosts 2026-09-23 land NAS SHIPPED via nx_organ_ship: nx_cdp_capture sha 798f2e25 (build src_path buildroot/runtime/nx_cdp_capture.nx), PROVE GREEN 26/26 on the NAS through the organ_gate.conf declaration (the end-to-end leg announced E2E-NOT-RUN-HERE, e2e_ran=0; the laptop run carries it at 33/33), ADOPT-GATE no-live-elf = first promote (announced), harness token h0423172feb601dfe rows=11, PROMOTE live_sha == staged_sha, OFFC-TWIN installed from the promoted binary, REGISTERED callable (schema rows appended to knowledge/tool_schemas.conf for the organ and the gate; nx_toolreg_reconcile owed for tools/list discovery), DOC DOCUMENTED hits=3 2026-09-23 measure closing measures: nx_cdp_capture_gate SHIPPED sha 4b5aad68 (prove GREEN self, promoted, offc twin, registered); nx_toolreg_reconcile registered=2 (both rows on tools/list discovery, the 47 no-schema residue is not this lane's); scoped regen cms published index.html 374713 B (prev 369396) api.json 90811 B forkfresh=4 verdict=GREEN; nx_page_verify https://nishifamily.com/compare/cms VERDICT=GREEN status=200 a11y-issues=0; nx_compare_rank cms rungs=24 done=14 open=10 DARK-CONTRACT=0, CM13 done via cd_capture with FINISH REGISTERED-UNAUTHORISED at rank time -- cap minted the same minute (allow nx_cdp_capture,nx_cdp_capture_gate exp 1792778435 nonce 1790186500, laptop file ~/.nishi/cdp0923.cap), so the label clears on the next rank; cms.gates row appended (total 11612 B) 2026-09-23 land nx_cdp_capture re-promoted for the referee: candidate 07310c9c61ff2216e8854e04b6d1d40113a253899d7662f8b24026dcbc3332dc (140232 B, prev banked) -- nx_organ_ship BUILT it, PROVED nx_cdp_capture_gate 26/26 GREEN on the NAS (35/35 with the browser on the laptop), the content ruler named ONE lost run (the usage sentence superseded by the one naming boxes 2026-09-26 queue W11 connect visual-review seat, three measurements for the capture lane: (1) the shared profile C:\Users\elder\AppData\Local\Temp\nx_cdp_prof carries a SAVED credential for nishifamily.com -- the alpha login painted an autofilled handle and a masked passphrase while the served HTML carries no value (count 0 over 11278 B), and a clean profile of the seat's own painted the placeholders with the same referee grade; a capture that inherits another run's browser state is not a first visit, bakes an operator handle into evidence PNGs, and keeps a password in a TEMP profile every seat shares; debt 1790438914 (a fresh user-data-dir per capture, the password manager off in the launch argv). (2) a full-height door on a scratch conf (same oracle and widths, 390x6000 dsf 2 giving ihdr 780x12000 in 930537 B, and 1440x3600) captured and graded cleanly and measured the fold cards the standard doors leave UNMEASURED (the connect-faith about pages 48 of 48 and 54 of 54 boxes measured) -- a candidate door row for pages whose content lives below the fold, with the cover-range caveat that a view() reveal completes only in the upper part of a tall viewport. (3) the referee's text_over_image flag reads every text box on a page whose body carries a CSS gradient (the capture maps body as an img box): 19 of 19 and 29 of 29 on the connect alpha home, so the flag discriminates nothing there; the verdict is unaffected because the paper method grades. 2026-09-26 queue W11 connect visual-review seat, two more gaps the capture lane owes, both named from this review: (1) NO DARK-SCHEME DOOR -- both connect alphas ship a dark theme (the front door @media prefers-color-scheme dark plus :root[data-theme=dark]; the app's default palette is dark and light is the media query), and every capture this seat took is the light scheme because the driver never calls Emulation.setEmulatedMedia; the dark palettes stay UNMEASURED on the painted page (declared only: front door --ink-faint #8A8C84 on #14161A about 4.85, app --faint #6a7488 on #0b0e15 4.11); owed: a door column or conf row carrying the colour scheme. (2) THE RECEIPT NAMES THE REQUESTED URL, NOT THE DOCUMENT THE PIXELS CAME FROM -- a capture of a local page that posts a form painted the POST response (https://nishifamily.com/alpha/connect/register, the register result) while capture.receipt and CD-SETTLED both say file:///...submit.html; the driver already reads location.href (CD_EVAL_HREF) but prints it only on a settle timeout. Owed: a document field on every receipt row, so a page that navigates after load is filed under the subject that was actually graded.

CM14 Design decisions verified in the bytes · cell LANDED · gate UNRECORDED

Contract	`ua_token_ref` · executor Organ · 1 u · stands on CM5
Definition of done	A ruler tooth that refuses a defined-but-unreferenced token and a blueprint key that lands in no measured CSS property; done when it fires on a planted unreferenced token and passes the live showcase page
Matrix cell	Design decisions verified in the emitted bytes: every token referenced, every blueprint claim landed (no design theater) · `runtime/_hdl_build/nx_ui_tokenref_lib.nx` · LANDED decl
Gate	`nx_ui_audit_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-26 land DESIGN DECISIONS VERIFIED IN THE BYTES, SHIPPED: nx_ui_tokenref_lib (ua_token_ref) reads every custom property a page defines inside its style blocks and every var() reference on the page, refuses a defined-but-unreferenced property by name unless a marker row exempts it, judges every blueprint row the claims table (knowledge/design_claims.conf: fields, field:i, needle, prefixed, narrowed by key=value and key@i=value, per surface) declares and refuses a claim that does not land, and reports every derived token row; nx_ui_audit tokenref is the verb; nx_ui_audit_gate 1531b98c 38 of 38 on the NAS with the live leg over the rebuilt alpha (35 of 35 in-process on the laptop, the live leg SKIPs there by name), declared in organ_gate.conf for nx_ui_audit and the lib 2026-09-26 measure the ruler's first real run REFUSED yesterday's showcase alpha on three findings the eye had passed: the type scale defined --t4 and nothing referenced it, the blueprint's heronote row never landed on the index, pagenote never landed on a sub-page, and the section_trust kicker, heading, lead and note landed nowhere (the band carried the sigs under the steps' heading); nx_site_build 96c3feb3 fixes all four (the unreferenced token gone, the note under the hero actions, a trust group heading the sigs), both alphas rebuilt (gospelofwest 12 of 12, cms-scene 24 of 24), and the live leg reads CLEAN on the index and a sub-page 2026-09-26 lesson a blueprint is a set of claims and the emitted bytes either carry them or do not: a row the emitter silently drops is design theater with the operator's own words, and a token the scale defines that no rule reads is the same defect in CSS; both were invisible to every ruler that counted ingredients, because a count of definitions rises when theater is added. The claims table is data so a new key is a row, never a rule, and a key no row names is counted unclaimed rather than acquitted 2026-09-26 land two rulers grown for the loop: nx_ui_tokenref_lib gains the suffixed:<literal>:<i> claim kind (field i then the literal) and design_claims.conf's pages row moves from prefixed:/:1 to suffixed:/:1, because a root-relative mount writes /mount/notes/x and a relative site writes notes/x and both carry notes/ (the prefixed kind read the relative form MISSING on every file-door fixture); nx_ui_audit's craft block now publishes its absent points by name ("missing":"img ...") through ua_pt, the same expression that adds each point, so the refine loop acts on a point BY NAME and an unnamed point refuses. nx_ui_audit_gate 41 of 41 on the NAS (9bdd7b58, promoted with its 6 renamed teeth and fixture pages enumerated as the loss): a mount href and a relative href both land, a bare /notes anchor does not, and exactly the pages row is missing

CM15 Agent-ready emission · cell PRESENT · gate NO-GATE-DECLARED

Contract	`sb_agent_ready` · executor Organ · 1 u · stands on nothing
Definition of done	schema.org JSON-LD, explicit action cues and dated validity emitted from the blueprint with zero script, graded by an agent-readiness tooth; done when the tooth passes on the emitted index and fails on the same page with the block removed
Matrix cell	Agent-ready emission: structured data, action cues and temporal validity signals by construction, zero script · `runtime/_hdl_build/nx_site_agentready_lib.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_site_agentready_lib`; declare `nx_site_agentready_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-22 land SHIPPED, referee grade: gate GREEN 35 of 35 on the NAS against the promoted ruler, awaiting operator acceptance. nx_site_agentready_lib (sb_agent_ready): an agentready row (a schema.org type and an updated date, data in the blueprint) arms the editorial emitters to write Microdata (a WebPage scope, itemprop name and description on the hero, a publisher of the declared type on the wordmark, a dateModified time in the colophon) and a data-action cue on every call to action and card (ed_sc_act_open), zero script by construction; nx_ui_audit grades an additive 4-point agent block (itemscope, itemtype schema.org, itemprop, datetime, action cues, key agent_points). Promoted: nx_ui_audit 3f489573 645,016 B (contentdiff GREEN lost 0), nx_site_build 2608b3a0 234,680 B (contentdiff lost 3, the three template fragments restructured into ed_sc_act_open, bytes proven identical), nx_site_agentready_gate 4bbec1a5 128,104 B registered, rostered and declared in cms.gates; _offc installed for both organs 2026-09-22 measure done-rule met: the armed fixture page carries the scope, five itemprops, the dated time and six cues (13,461 B) while the plain page (13,034 B) carries none and is byte-identical before arming and after disarming; the ruler reads agent 4 of 4 on the armed page and 0 of 4 on the plain page and the zero-script craft point survives; the real gospelofwest blueprint with the row emits itemscope 2, itemprop 5, data-action 10, datetime 1, script 0 and grades agent 4 of 4, craft 14 of 15 (imagery, the CM7 gap); the tokens-only path re-emitted the smoke byte-identical 7 of 7 on the laptop 2026-09-22 lesson JSON-LD is a script element to every byte ruler including ours, so the honest agent-ready emission is Microdata on the elements the page already writes (a script tag would have cost the zero-script exceed for a second syntax of the same facts); and the first NAS gate run read 33 of 35 because the LIVE ruler was still the pre-agent build: two ruler teeth failing on the promoted subject before its ruler was promoted is the gate reading the right subject, not a flake, and it read 35 of 35 the moment the ruler promote landed 2026-09-26 queue W11 connect visual-review seat: ed_sc_foot (nx_nishi_editorial) appends the dated validity signal (a time element reading Updated <date>) straight after the colophon text with one space, so an armed release whose colophon ends without punctuation paints one run-on line -- measured on the alpha front door https://nishifamily.com/alpha/connect-faith: CONNECT ALPHA -- INDEPENDENT; NOT AFFILIATED WITH OR ENDORSED BY THE CHURCH OF JESUS CHRIST OF LATTER-DAY SAINTS OR ANY CHURCH UPDATED 2026-09-25. One fix in the emitter (a separator or its own line before the time element) covers every armed showcase site; nx_site_agentready_gate's tooth checks the time element's presence, not its separation, so a separation tooth rides the fix.

CM16 Constraint-solved section layout · cell LANDED · gate UNRECORDED

Contract	`ed_layout_solve` · executor Organ · 3 u · stands on CM11
Definition of done	A solver over the derived tokens and the brief chooses grid, grouping and whitespace per section, printing its objective beside the page; done when a planted misaligned layout scores worse than the solved one on the printed objective and the pixel referee agrees
Matrix cell	Constraint-solved section layout (grid alignment, grouping, whitespace and colour harmony solved over the derived tokens) · `runtime/_hdl_build/nx_nishi_layout_lib.nx` · LANDED decl
Gate	`nx_nishi_layout_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-28 measure the section layout solver on the estate's fixture tokens (14 px body, 74 ch measure, 1200 px container, spacing s1..s7 4..64 px): two cards solve to two columns (objective 125: rows 125, every other term 0), three steps to three columns (83: measure 0, rows 83), four cards to four columns at 1200 px (110: measure 48 + rows 62) and to two columns at a 600 px door (four columns would hold nine characters a line, three would orphan the fourth); a planted orphan (three steps forced into two columns) scores 1316 (orphan 1000 + empties 150 + rows 166) against the solved 83 on the printed objective; the pixel referee over browser-visible twins of both builds (desktop door) counts 0 defects on each -- it cannot see an orphan, and it never rates the solved layout worse. Gate nx_nishi_layout_gate 31/31 on the laptop (the pixel leg runs where cdp_capture.conf declares a browser-visible fixture root; elsewhere it is announced absent, never a precondition) 2026-09-28 lesson the objective is built from shares the tokens and the brief already carry, so only five weights and three bands had to be typed and each is a conf row with its calibration; the line-length band is derived (a third of the measure to the measure), the cell width from the container and the gap token, chars per line by the emitter's own ch convention -- and the gate re-derives the argmin by brute force over the whole band (a second method class for the solver's one claim) rather than asserting the column count a seat predicted: the first draft asserted two columns for four cards and the solver was right to choose four at 1200 px on those tokens 2026-09-28 land nx_site_build SHIPPED 74132c46 through nx_organ_ship (build; four declared gates proven on fresh closure banks: nx_cms_plugins_gate 35/35, nx_ink_derive_gate 39/39, nx_design_modes_gate 22/22, nx_nishi_layout_gate 30/30 on the NAS where no browser resolves; content ruler RED adjudicated: the 3 lost runs are the three stylesheet rules the solver replaces, each superseded by its custom-property twin, loss token 2a7810b5; the bare probe was refused as incomplete once a loss is approved, so the representative probe drives both binaries on the layout gate's showcase fixture site, DIFFERS STRONG 976 to 1773 bytes at the LAYOUT conf line, reviewed under NISHI-SHIP-BEHAVIOR-REVIEW/1 record a706f4eb; promoted, offc twin installed, registered); nx_nishi_layout_gate SHIPPED 08776c47 on its own lane (first promote, 30/30, registered; its first ship read 26/30 because the emitter beside it was still the pre-solver build -- the emitter ships first, the gate proves against the fresh build); knowledge/layout_solve.conf on the NAS; organ_gate.conf rows nx_nishi_layout_lib and nx_site_build -> nx_nishi_layout_gate; import census refreshed so the finish state reads from the graph. Scope stated: only showcase-tier blueprints (tokens row) are solved; the kit-tier client sites and the gospelofwest alpha are byte-untouched; a showcase client re-emit carries the solved grid through the release lane, never by this seat

CM17 Render-grounded refinement loop · cell LANDED · gate UNRECORDED

Contract	`sb_refine_loop` · executor Organ · 2.5 u · stands on CM13
Definition of done	The emitter re-runs with the referee's named defects until the pixel referee and the bytes ruler both pass, each iteration journaled; done when a planted defect is repaired within a declared iteration budget and a page that cannot converge is refused by name
Matrix cell	Render-grounded refinement loop (emit, render, judge the render, re-emit until the gate passes) · `runtime/_hdl_build/nx_site_refine_lib.nx` · LANDED decl
Gate	`nx_site_refine_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-26 queue the operator's standing direction 2026-09-26: no navel gazing -- the referees exist so the loop can improve what is emitted; the spine is CM17 with CM26 as its browser-side receipt and CM27 as its memory, then CM28 and CM29; every rung's proof is a planted defect repaired and a refusal by name, never a score that only rises 2026-09-26 land THE LOOP IS BUILT AND PROVEN: runtime/_hdl_build/nx_site_refine_lib.nx carries sb_refine_loop (CM17) and sr_archive_select (CM27), nx_site_refine is the CLI (loop <site> <out_root> <url_base> <journey> [budget=N] [refconf=], archive <out_root>), nx_site_refine_gate reads 62 of 62 GREEN on the laptop with the browser leg. EVALUATE: every variant is emitted by nx_site_build and judged by the estate's referees forked as separate binaries (nx_page_pixjudge at every capture door, nx_ui_audit file and tokenref, nx_cdp_capture journey), their fields the score vector and their own vocabulary the defect names. REFINE: a named defect maps through knowledge/refine_remedies.conf to one knob move, a token row moved one step (the larger of the page's own scale step t6 over t5 and the referee's bar) inside a declared evidence band, or a blueprint row set from the catalogue; a content-owned class is named as residual, a class the table does not declare refuses by name. IMPROVE: every variant archived content-addressed with its vector and verdict, the next parent the best admitted by objective then novelty, a child admitted only when no defect or score axis worsens and one improves, an axis the parent measured that the child cannot a rejection never an acquittal; the budget is charged per move and derived from the remedy rows when the caller gives none 2026-09-26 measure end to end on the laptop: a showcase fixture with a PLANTED flat type scale (t5 3022, step 1049 permil against the referee's 1178 bar) reads hierarchy_step_short=2 at the root (h1 35 px beside h2 32 px at the phone door), the loop moves td 10900 to 12840 by the referee's bar, the child reads 0 short with every other axis equal and is ADMITTED, and the run closes CONVERGED-RESIDUAL naming craft_missing:img (the imagery point is the client's photograph); budget=0 refuses REFUSED-BUDGET with the root judged and nothing moved (bite-proven against the converging run); a remedy table without the class refuses REFUSED-UNDECLARED:hierarchy_step_short. Archive rows 2, receipt rows iter 2026-09-26 lesson the loop's first runs were refused three times by its own referees, and each refusal was a defect in a referee or a driver, never in the loop: a display sized at the heading read OK (the strict hierarchy axis), a moved child failed contrast on a card caught mid-fade at the fold (the band means judged blends), and the file-door journey died on a clean-URL link (chrome-error); each became an instrument change with its gate tooth before the loop was believed. A loop that repairs what a referee cannot see is theatre; the referees were grown until the loop's convergence meant something 2026-09-26 queue OWED after this leg: nx_site_refine and its gate shipped to the NAS with a cap and a registered row (the browser leg SKIPs there by name, the in-process teeth run); CM28 sr_knob_search (the factorial search over the token bands and catalogue choices, offline by the referee panel then online through nx_cms_abtest) and CM29 sr_policy_pin (the hash-pinned policy the loop cannot edit); the cms-scene alpha rebuild through its release recipe (nx_plan_run now requires an owned workflow definition: managed-owned-definition-required); the referee's remaining blind spot named by its own counts: the phone-door card straddling the fold abstains NO-CORE on every capture, so the pixel contrast of scroll-revealed text at the fold is UNMEASURED until the capture settles reveals (row 1790373803) 2026-09-26 land SHIPPED to the NAS: nx_site_refine_gate e76f2914 (56 of 56 in-process on the NAS, the browser leg announced SKIPPED-BY-NAME there, 62 of 62 on the laptop) and nx_site_refine 30fabd5b (first promote, registered, offc twin installed); the referees it forks are all live at their shipped shas: nx_page_pixjudge 28dcc838 (scale-step axis, TEXTFREE-MISS and NO-CORE abstentions), nx_cdp_capture 1bab135a (file-door clean-URL resolution, cd_journey_run), nx_ui_audit dde9e060 (craft missing list, suffixed claims), nx_site_build 3569418a (display floor); confs on the NAS: refine_remedies.conf, refine_referees.conf (conf rows carry candidate paths for the two knowledge trees), pixjudge.conf hierarchy_min_ratio_permil, design_claims.conf pages suffixed:/:1, organ_gate.conf rows for the loop and its lib

CM18 Pairwise-preference referee with published agreement · cell LANDED · gate UNRECORDED

Contract	`pj_pairwise` · executor Organ · 3 u · stands on CM6
Definition of done	Candidates compared pairwise by the estate referee and an outside VLM oracle, ranked by a rating model, with agreement to a human-judged set published beside every grade, nx_ui_judge extended rather than twinned; done when the referee's ranking of a judged set agrees with the human ranking above a pre-declared floor and the number is on this board
Matrix cell	Pairwise-preference design referee (expert or VLM judged, rating-model ranked, calibrated against human critiques) · `runtime/nx_page_pairwise_lib.nx` · LANDED decl
Gate	`nx_page_pairwise_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-28 measure first real run of the pairwise referee (laptop, the browser host, the two variants the refine loop archived under nx_refine_gate/loop/variants): `nx_ui_judge pairwise <cands> -` on the awwwards platform -- candidates=2 scored=2 pairs=1 referee_judged=1 ties=0; RANK 1 v24fd5f76 elo=1516 score=992, RANK 2 v5ac366ee elo=1484 score=985 (Elo start 1500, K 32, the estate's nx_dr_elo, bit-exact); HUMAN set=- rows=0 comparable=0 agreement UNMEASURED; VLM oracle=none UNMEASURED (no image route in nx_extllm); PAIRWISE-VERDICT floor=UNCALIBRATED verdict=UNCALIBRATED-FLOOR. The rung's done-rule (agreement above a pre-declared floor on this board) is NOT met and cannot be met by a seat: it needs the operator's judged page pairs (pair 2026-09-28 lesson extend, never twin: the referee is a lib (nx_page_pairwise_lib, prefix pj_) composed from three incumbents found by nx_capsearch before a line was written -- the one rubric referee (nx_page_rubric_lib, which grew a SILENT per-platform score by delegating the printed platform row and the compared number to ONE function, rub_rubric_score_into, so the number on the receipt and the number in the comparison cannot drift), the estate's deterministic integer Elo (nx_dr_elo, start 1500, K 32, bit-exact) and the gate base class's Wilson bounds (gv_wilson_permil at z95) -- and it rides nx_ui_judge as a verb, the incumbent the matrix row names, whose no-argument path stays byte for byte (the gate proves it). Three shapes worth naming: an unmeasured candidate must rank LAST, never on its untouched start rating above a rated loser (the first draft would have ranked it second); the floor is a conf row whose calibration is DERIVED by the referee itself (the inter-judge agreement printed from any judged set with two judges) because no seat may type the bar a referee is held to; and a mechanical rewrite of a helper's call sites rewrote the helper's own body into a self-call -- the crash guard named the unbounded recursion and nx_addr2line named the line in one call, a defect the estate's tooling turned into a two-minute fix 2026-09-28 land the pairwise-preference referee SHIPPED on the NAS: nx_ui_judge 55105965 through nx_organ_ship (the incumbent competency judge EXTENDED with the `pairwise` verb, never twinned; content ruler GREEN; the bare probe compares the no-argument path, byte-identical; gate proven 41/41 from the promoted nx_page_pairwise_gate once its sibling ladder preferred the staged candidate), nx_page_pairwise_gate SHIPPED 588446bd (first promote, registered, declared on cms.gates), runtime/nx_page_pairwise_lib.nx (pj_pairwise, the symbol the matrix row watches), nx_page_rubric_lib grew rub_platform_score (the silent per-platform score by the one arithmetic; its consumers nx_page_rubric and nx_site_refine read BEHIND until their lanes re-ship, behaviour unchanged), knowledge/pairwise_referee.conf on the NAS (platform awwwards; floor UNCALIBRATED with its calibration rule), organ_gate.conf rows nx_page_pairwise_lib, nx_ui_judge and nx_page_rubric_lib -> nx_page_pairwise_gate. Status: BUILT and PROVEN, the matrix cell flips to PRESENT on the next publish, the rung stays OPEN by its done-rule -- agreement with a human-judged set above a pre-declared floor is on this board only when the operator's judged pairs exist (two judges on the same pairs calibrate the floor); the VLM oracle arm reads UNMEASURED until nx_extllm carries an image route 2026-09-28 queue two arms measured for feasibility on 2026-09-28 and declared as rungs rather than half-built: (1) the VLM oracle arm needs an image route on the ONE egress door (nx_extllm_call): a capture PNG must be a DECLARED public file (a release row with its sha256, the C-by-default classification refuses an undeclared capture), the provider's terms row must admit the JUDGE use, and the request gains a base64 image content part -- every conjunct of the egress law, so it is an egress-board rung, not a patch; until then the oracle set is any pairs file in the referee's grammar written by any outside judge. (2) A NAS capture door for the browser-dark rungs (CM13, CM24, CM26, and the served-page judge on a beat): the sovereign render core (nx_browser_render: br_layout to br_shot_png over the Liberation faces the NAS already holds) can paint a page to PNG without Chrome, but the pixel referee also reads the .boxes text-box record the DevTools driver emits, so the door is the render core emitting the capture grammar (png + boxes + paper) behind a renderer=sovereign row in cdp_capture.conf with its parity to Chrome measured by nx_parity_judge_gate before any verdict rides on it -- a browser-board rung. Neither blocks the human-judged set: the judging surface runs on the laptop, the browser host 2026-09-28 land THE JUDGING SURFACE (the input only a human can give, made into minutes of clicking): runtime/nx_pairjudge_lib.nx (the pure core: every unordered pair of the referee's candidates in the referee's own order, what a judge already ruled on in either orientation, a no-script page with one form per open pair showing both captures side by side at a chosen door, the routes GET / and GET /img/<k>/<door>.png and POST /verdict with every refusal named: 403 csrf, 400 pair or verdict, 409 already judged, 404 elsewhere) + runtime/nx_pairjudge_serve.nx (a thin loopback daemon: bind 127.0.0.1:<port>, read, route, write, exit after a request budget) + nx_pairjudge_gate 30/30 on the laptop (real loopback end to end on a port proven free in a window; the budget exits the daemon with 0, no orphan). Every verdict lands through the referee lib's ONE row writer pj_pair_row (a pipe becomes a slash, a control byte a space), so the surface and the referee's reader share one grammar by construction. ALPHA LIVE on the laptop, the browser host: nx_pairjudge_serve 8098 knowledge/judged/cms.cands knowledge/judged/cms.pairs elder 0 desktop -> http://127.0.0.1:8098/ with four captured pages (alpha-gospelofwest, rival-nike-live, the two refine variants) = six pairs; a second judge runs the same command with another judge name; then nx_ui_judge pairwise knowledge/judged/cms.cands knowledge/judged/cms.pairs prints the agreement and the inter-judge floor candidate. NAS: nx_pairjudge_gate promoted f4de3003 DIRECTLY through /api/promote with the one lost run named (a renamed tooth) because the ship loop's probe cannot qualify against an incumbent that crashes by construction (the first NAS build's fixed fixture port 8097 was taken there and the gate walked into a null row; fixed: a free port found in a window, the wire teeth skipped by name when none is free, no row dereferenced when none landed); nx_ui_judge re-shipped 09eb7c3d against the new writer; nx_pairjudge_serve BUILT and STAGED 2290d307 on the NAS, declared `daemon` in organ_kind.conf and deliberately NOT deployed there -- it serves loopback on the host that holds the captures and the browser, so a NAS instance would have nothing to show; the stage slot is named here so the drift board's new row has an owner and a reason 2026-09-28 measure THE OPERATOR'S FIRST JUDGED SET, 2026-09-28: six pairs over four candidates (the gospelofwest alpha, the live nike.com home, two refine-loop variants), judge elder on the alpha surface at 127.0.0.1:8098; the referee (nx_ui_judge pairwise, awwwards platform score, Elo start 1500 K 32) agrees on 2 of 6 (333 permil, Wilson 97 to 700) and ranks alpha 1531, refine-v24fd5f76 1530, refine-v5ac366ee 1485, nike 1454, where the operator ranks nike first (3 wins of 3), refine-v24fd5f76 over refine-v5ac366ee, and the alpha tied with both variants; floor UNCALIBRATED (one judge, inter-judge UNMEASURED); verdict UNCALIBRATED-FLOOR, exit 4 2026-09-28 lesson the referee inverts the operator's top and bottom because its platform score is the bytes rulers' hygiene and craft counts (the nike home carries 53 scripts and 21 external assets and reads craft 6 of 15, so it ranks last) while the operator's eye judges photography, composition and confidence; a 2-of-6 agreement is the measurement CM18 exists to publish and it says the estate's judge does not yet see what the operator sees: the pixel referee's beauty axes and licensed imagery (CM6, CM7) are the missing signal, never a reason to move the floor; the operator's own verdict on the surface was that it is a good start and not the field's state of the art, which this milestone declares as data

CM19 Layout tokens derived for board pages · cell LANDED · gate NO-GATE-DECLARED

Contract	`ua_layout_derive` · executor Organ · 1 u · stands on CM5
Definition of done	The design ruler measures rail width, content width, column counts, breakpoints, sticky use and in-page navigation across the mirrored board-class pages and derives one board token set with n on every row [tailwind-responsive-design] [ghdocs-hello-world] [mdn-grid-common-layouts]; done when buildroot/knowledge/compare/board.tokens is written by the ruler and every row carries n
Matrix cell	Layout tokens derived for board pages · `runtime/nx_ui_layout_lib.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_ui_layout_lib`; declare `nx_ui_layout_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Design rows	GD1 A two-column shell at the wide breakpoint: a sticky section rail beside one wide content column adopted GD2 Widths, gaps and breakpoints derived from measured board-class pages, never typed adopted
Work log	2026-09-22 measure The six mirrored documentation pages carried NOT ONE breakpoint, track or sticky rule between them (nx_ui_audit derive REFUSED: type ratio n=0) because every one links its CSS externally; the layout evidence is in the stylesheets, so fourteen were mirrored and pinned beside the pages: MDN [mdn-css-page-layout] [mdn-css-reference-layout] [mdn-css-left-sidebar] [mdn-css-reference-toc] [mdn-css-global], GitHub Docs [ghdocs-css-1] [ghdocs-css-2] [ghdocs-css-3] [ghdocs-css-4] [ghdocs-css-5], Tailwind [tailwind-css-1] [tailwind-css-2], NN/g [nng-css-1] [nng-css-2]. The corpus is data: buildroot/knowledge/compare/board_layout.corpus (20 sources). 2026-09-22 lesson A page whose CSS is external contributes only what its bytes carry -- a derive over pages alone reads the field as carrying no layout at all, which is a false NOTHING, not a measurement. Mirror the stylesheets a page links and pass them as sources; the derive prints every source's value so the corpus can be checked row by row. 2026-09-22 queue nx_ui_audit has no gate of its own: nx_organ_ship refused NO-GATE-FOUND, and the ship was re-run with gate=nx_ui_layout_gate (the in-process proof of the extracted measurement base class nx_ui_measure_lib and of ua_layout_derive) plus a behaviour probe of the untouched file verb on the live cms page. OWED: nx_ui_audit_gate -- the file, extract, derive and layout verbs end to end over fixture pages, either by forking the freshly built binary (gk_run_capture) or, the OO direction, by moving the verb bodies into nx_ui_audit_lib so an in-process gate can drive them; declare the pair in knowledge/organ_gate.conf when it lands. 2026-09-22 land nx_ui_measure_lib.nx (the audit's measurement base class, extracted verbatim: value parser, bands + ua_bands_use seam, distinct-value table, um_median_field, um_bp_scan legacy+range syntax, the ONE token-row writer) + nx_ui_layout_lib.nx (ul_track_scan, ul_sticky_scan, ul_toc_count, ul_measure, ua_layout_derive) + nx_ui_audit layout verb over ua_verb_sources (derive and layout share ONE source reader); nx_ui_audit shipped sha 943dccb7 with gate=nx_ui_layout_gate 37/37 (sha f108708f) and a behaviour probe of the file verb IDENTICAL 2184 B; bands rail_px 160..360 px and sticky_top 0..120 px in knowledge/design_evidence_ranges.conf; the design-architecture plane (ux-owned) ran 4 of 4: audit file, layout --corpus, derive --corpus, page_verify GREEN. 2026-09-22 measure board.tokens DERIVED over 20 sources (6 pages + 14 stylesheets): bp2 1012 px n=5, rail 306 px n=6, stickytop 0 px n=5, mpx 1280 px n=5, bp1 768 px n=10, toc present in 4 of 20 sources (the pages carry the marker, the stylesheets cannot). board_type.tokens now derives too (t1 16 px n=7, ratio 1.16 n=5, s1 10 px n=12, s6 60 px n=3, mot 150 ms n=9, rad 6 px n=9, fam sans n=5) where the pages alone REFUSED at ratio n=0. 2026-09-25 queue add the benchmark's page and stylesheets to the board-class derive corpus beside MDN, GitHub Docs and Tailwind, so board tokens are derived with it in the population, never copied from it 2026-09-26 land the watch row named a file that never existed (runtime/_hdl_build/nx_ui_layout_lib.nx) while the derivation lived in runtime/nx_ui_layout_lib.nx and board.tokens carried its six rows with n; the row is repointed, nothing rebuilt

CM20 Responsive shell with a section rail · cell LANDED · gate NO-GATE-DECLARED

Contract	`sc_shell_pass` · executor Organ · 2 u · stands on CM19
Definition of done	The one compare assembler wraps its four production sections in a two-column shell at the derived breakpoint: a sticky section rail beside a wide content column, card grids that fill the width, multi-column lists, tables that scroll in their own container, nothing removed [mdn-structuring-documents] [nng-layer-cake]; done when the emitted text bytes are unchanged, the page never scrolls sideways at 390 and 1440 px, and the shell reads its widths from board.tokens
Matrix cell	Responsive shell with a section rail · `runtime/nx_compare_garden_lib.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_compare_garden_lib`; declare `nx_compare_garden_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Design rows	GD1 A two-column shell at the wide breakpoint: a sticky section rail beside one wide content column adopted GD5 Summary before detail and disclosure for the long tables, the layer-cake and progressive-disclosure patterns proposed
Work log	2026-09-22 queue 2026-09-22 the ux role is Accountable for design-architecture (RACI row, crew row ux design-architecture nx_ui_audit, binding design-architecture to nx_plan_run) and runs plan-design-architecture-cms- (measure the emitted board with the design ruler, derive board.tokens from six board-class mirrors, verify the page); the seat tutors by landing the design rows GD1 to GD6 with their evidence and by building the shell as the first landing under that ownership; nothing is deleted to structure the page 2026-09-22 lesson a page can carry every fact and still read as a dump: structure is a layout and a join, not a cut; the three complaints (one endless column, records as prose, relationships scattered) are one architecture defect with three faces, and the remedy is measured widths, one table renderer and one join by symbol 2026-09-22 measure nx_organ_ship nx_swcompare_matrix: the content ruler read lost_from_live=2 (SS-COMMIT REFUSED: segid= and (seq1730). NOTHING -- seg-store lib messages from a sibling's shared-lib change, SOURCE-ADDED nx_scorecard_receipt_lib.nx + nx_store_seed_checked.nx, not the shell) and GAINED 55 (every shell string among them); approved by the exact evidence record allow_loss_v1=38dc0c2b6833582c1c7b1b45069b6a31a54cd32b09c5f89f3c12edb899b22980 and re-run. The behaviour probe on cms html is EXPECTED to differ by exactly the wrappers and the shell CSS: that difference is the rung. 2026-09-22 land garden shell LIVE: nx_swcompare_matrix 04c61be8 (1121032 B) on the serving root, _offc and buildroot/_offc (all three hashed); nx_compare_garden_lib + nx_compare_garden_gate 32/32 (9a5cbe18); /compare/cms republished 333965 B via nx_compare_regen_scoped --domain cms (the EC67 per-call organ, no shared scope file); page carries data-shell='two-column' x1 and data-bp2='1012' x1; page_verify GREEN a11y-issues=0; GD1 adopted 2026-09-22 measure ship loop: rebuild is byte-stable (jobs 1790095253 and 1790095572 both built 04c61be8; job 1790094509 built ba342a09 before a sibling shared-lib change landed); behaviour probe live-vs-staged cms html DIFFERS +761 B at offset 9053 = the shell; review record knowledge/status/harness/review-nx_swcompare_matrix-cm20-shell.txt sha 6eafcc80 over observation 56363174 (bound to ba342a09, superseded by the rebuild); the promote that landed 04c61be8 is journaled in organ_ship.jrnl at 1790095687 SHIPPED GREEN by a concurrent organ_ship run, not by this seat's reviewed run (which read artifact-hash-unavailable once the stage was consumed) 2026-09-22 lesson an exact-artifact review binding cannot land while a sibling ships the same target: each run rebuilds the candidate, the review binds to one candidate, and a concurrent promote consumes the stage; the loss token is deterministic across byte-stable rebuilds so approvals converge only in a quiet window -- coordinate a shared generator through its RACI lane, never by racing; a cap's second field is its expiry (the lane cap expired at 1790092533 and read as scope denials all afternoon); regen.scope is the EC67 race in the wild (re-written to lang under this seat), so the scoped regen is called with --domain on argv from now on 2026-09-22 queue owed: nx_ui_audit_gate declared in organ_gate.conf; CM21 sc_record_table (records as tables); CM22 sc_rung_dossier (join related rows by rung id); CM23 dm_mode_load (garden vs wabi-sabi as data); operator ACCEPT row for the shell in the accept- plane is a separate signature from this land 2026-09-22 measure closure ran as the ux-owned plane nx_plan_run design-architecture-cms extended to 8 rows (5 nx_eco_graph_build, 6 nx_importcensus, 7 nx_compare_regen_scoped --domain cms, 8 page_verify): 8/8 COMPLETED in 41 s, graph 21707 nodes 61605 edges, census asof 1790096972, /compare/cms republished 340919 B with exactly one nx-shell wrapper (tag-anchored count 1; the other 3 mentions are board text), page_verify GREEN a11y 0; nx_compare_rank cms reads rungs=24 done=10 open=14 with CM20 out of the FINISH list; RACI design-architecture ux=A engineer=R migrated 243 rows 8/8 GREEN; debt 1790096923 filed for the TSV->store migrate that cannot express a removed row 2026-09-25 queue the benchmark is this rung's done-rule made concrete: a section rail that marks the current section, a contents rail with one link per record, the rails folding into one breadcrumb line on phone; and the shared emitter's section heading must never outrank the page title -- the referee's hierarchy tooth fires on this board today (h2 40 px over h1 37 px), so the fix is in the one generator every board uses, shipped through its lane

CM21 Records as clean tables · cell LANDED · gate NO-GATE-DECLARED

Contract	`rct_end` · executor Organ · 2 u · stands on CM20
Definition of done	Every record class the boards emit as prose (rank derivations, ledger notes, evidence bands, incumbents) renders as a table with named columns through one renderer, with progressive disclosure for the long ones [nng-progressive-disclosure]; done when a census of prose blocks carrying pipe or key-equals records reads zero on every published board
Matrix cell	Records as clean tables · `runtime/nx_compare_record_lib.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_compare_record_lib`; declare `nx_compare_record_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Design rows	GD3 Every record class renders as a table with named columns; prose is for argument, tables are for records proposed
Work log	2026-09-22 measure DARK-CONTRACT cleared before the build: cms.matrix rows 54 and 55 repointed from runtime/nx_compare_garden_lib.nx (an existing organ, so the ruler read the missing symbol as DARK) to the absent lib files the rungs will land in, runtime/nx_compare_record_lib.nx (sc_record_table) and runtime/nx_compare_dossier_lib.nx (sc_rung_dossier); CM23 already pointed at absent runtime/nx_design_modes_lib.nx; cms republished 336568 B via nx_compare_regen_scoped --domain cms; nx_capsearch (record table render prose disclosure columns) names no incumbent renderer for record prose (nx_record_tsv_gate is a TSV bridge, nx_design_ledger_lib renders one ledger), so sc_record_table is novel and composes the ledger's table shape; baseline structure on cms: 23 tables, 12 meth prose blocks, 7 sections, 0 pre 2026-09-22 land nx_compare_record_lib (rct_begin/rct_row/rct_row_word/rct_row_sum/rct_end: a capture renderer in the page's own plan-table shape with the row count DERIVED and published as data-rows, disclosure above the conf bar as a details summary; rct_census_list: the record-as-prose ruler that NAMES every offending block) + nx_compare_record_gate 60/60 promoted ab445398 registered offc-twin + nx_record_prose CLI (live d8cc7d64 count-only; the listing candidate c4d6b806 is staged behind the loop's zero-exit probe until the cms page reads CLEAN); the four bars are threshold rows in knowledge/design_evidence_ranges.conf (record_kv=3 record_pipes=2 record_named=5 disclose_rows=12, each with its calibration); adoption source-landed at all five prose-record sites: ladder+bars and treatment verdicts (nx_swcompare_lib plan_pass), scorecard tally, design ledger verdict (nx_design_ledger_lib), rival-claim provenance (nx_swcompare_matrix), every partition re-derived by that ruler's own sum and read RECONCILED or LEAK on the page 2026-09-22 measure baseline census on the LIVE cms page before the matrix ships (nx_record_prose, thresholds SET): blocks=17 records_prose=5 kv=9 pipes=42 named=39 partition=3 tables=23 2026-09-22 lesson a row count handed to a renderer beside the rows is a second copy of the rows' shape and the two drift, so the renderer captures the rows and derives the count; and a count without a worklist is not actionable, so the census prints RECORD-BLOCK id= kv= pipes= named= partition= text= for every offender before its verdict line (the first cut of this very instrument printed a count and nothing else) 2026-09-22 lesson the ladder's target counters are not a partition (best in class and frontier count by code, undated, bad class and duplicate by state) so the table prints no sum row for targets; rungs, roles, bars, treatment rows and design rows each have a sum function in their own ruler (ld_partition_sum, ld_role_sum, bf_partition_sum, trt_partition_sum, dl_partition_sum) and the page re-derives every one beside its total 2026-09-22 queue closes when nx_swcompare_matrix and nx_swcompare_sota are promoted, buildroot/_offc refreshed, cms regen'd and nx_record_prose reads records_prose=0 on every published board; the sota generator ship was refused NO-GATE-FOUND (declare its gate in organ_gate.conf first); owed: the regen prints records_prose= per domain as a receipt line; nine dated candidate copies of the emitter in buildroot/runtime still carry the old prose and are lifecycle EC71 work, not this rung 2026-09-22 measure done-rule on the LIVE cms page after both generator ships: nx_record_prose sites/nishifamily/compare/cms/index.html reads blocks=17 records_prose=0 kv=0 pipes=0 named=7 partition=0 tables=28 thresholds=SET verdict=CLEAN, against the 1790100651 baseline of records_prose=5 kv=9 pipes=42 named=39 partition=3 tables=23 on the same page; the seven named counts left are sentences under the record_named bar, not records 2026-09-22 land both generators SHIPPED through nx_organ_ship with the field discovery receipt as a table (fi_summary_table in nx_swcompare_lib copies the field 2026-09-22 land THE CENSUS RIDES THE PUBLISHED LINE (the queue row's owed regen receipt): nx_compare_regen 9aa402ca and its scoped twin nx_compare_regen_scoped c88a8c42 both SHIPPED through nx_organ_ship (nx_forkfresh_gate 45/45, content GREEN one-way gain, behaviour reviews review-nx_compare_regen-cm21-records-receipt.txt 026feb7a and review-nx_compare_regen_scoped-cm21-records-receipt.txt eea93aca over the scoped cms publish as the representative probe): rg_records_receipt censuses the exact bytes each published .html page with rct_census, the same ruler nx_record_prose and nx_compare_record_gate run, prints records_prose= on that page's published line and records_pages= records_prose_total= records_uncal= before verdict= (additive, never fails); witnessed on the promoted scoped twin: published cms/index.html (349553 bytes) records_prose=0 and records_pages=1 records_prose_total=0 records_uncal=0 verdict=GREEN, forks 4/4 FRESH, agreeing with the CLI's independent reading; owed: a fixture-tree tooth for the receipt in a regen gate (today the proof is the probe observation plus the CLI agreeing), and the fleet total from one full regen run 2026-09-22 lesson the contract was DARK while the capability shipped: matrix row 54 and this rung named sc_record_table, a symbol nx_compare_record_lib never declared (its entry is the capture, and rct_end is the function that emits the table), so the cell read DARK-CONTRACT (organ present, symbol absent) after four promotes; repointed BOTH the row (_ABSENT_:rct_end) and the rung symbol (rct_end) to the real emitting entry instead of minting a name-only alias, because the flip must reward the entry point that delivers the capability, never the name; the cell flips on the next regen of cms 2026-09-22 measure fleet witness from the full nishi_compare_regen run 1790106496 with the receipt in the path: published=255 fails=0 forks 4/4 FRESH records_pages=129 records_prose_total=10 records_uncal=0 verdict=GREEN; 120 of 129 published pages read records_prose=0; the residue is 9 pages carrying 10 records in TWO classes the census names by block: the sota page's Measured bindings verdict paragraph (named=6, one record each on llm foodscience water livingdocs cleanserve lang smallos researchreach) and sharehub's two head-to-head measured blocks (Measured on both sides kv=5; Where to beat them named=7); so the done-rule holds on cms and on 120 boards and NOT yet on every board, and the worklist is those two emitters in the sota generator, the same rct_begin/rct_row/rct_row_sum/rct_end conversion the five matrix sites got 2026-09-22 lesson the emitter of the Measured bindings block is not the organ its name suggests: the sota-class boards (llm foodscience water livingdocs cleanserve lang smallos researchreach) are rendered by nx_swcompare_matrix, which imports nx_swcompare_sota_lib, while nx_swcompare_sota never imports that lib and carries its own copy of the paragraph in main (spelled consistent= stale_under=, dormant: no published page carries that shape); converting the lib alone left the sota candidate byte-equal on that block (the staged elf kept the old fragment and none of the new caption, which is how the mistake was caught before approval); both copies now render the rct table with the partition sum, and the lib-vs-program duplicate is EC71 lifecycle work for the sota lane, named here rather than merged blind 2026-09-23 land residue conversion SHIPPED through nx_organ_ship, both trees: nx_swcompare_sota 8b3d77e0 (loss set e3d83173 = 16 retired sentence fragments, review 02ab29c0) and nx_swcompare_matrix ac683187 (loss set 6011ff50 = 21 retired fragments, review 667a21a4; the cms probe's same-size diff adjudicated as clock-derived age fields since the field receipt is all-digit and fi_kv_rows is output-identical on it); the sota lib and the sota program both render the Measured bindings partition as a table with the sum beside bound, the shared lib renders Where to beat them as ppp-gap rows and the probe stamp as ppp-stamp rows through fi_kv_rows (the one key=value row emitter fi_summary_table composes); buildroot/_offc forks refreshed and hash-identical (sota 8b3d77e0, matrix ac683187), the scoped regen read forkfresh=4; witnesses of both residue classes: sharehub 313412 B records_prose=0 (CLI CLEAN named=5 tables=18, was 2) and foodscience 186456 B records_prose=0 (was 1); fleet witness owed from the full run 2026-09-23 measure DONE-RULE HOLDS ON EVERY PUBLISHED BOARD: full nishi_compare_regen run 1790126244 with the shipped generators (sota 8b3d77e0, matrix ac683187) and all four buildroot/_offc forks FRESH: published=255 fails=0 records_pages=129 records_prose_total=0 records_uncal=0 verdict=GREEN; every one of the 129 published .html receipt lines reads records_prose=0 (count of ' records_prose=' lines 129, count of ' records_prose=0' lines 129, difference 0); the nine residue pages of run 1790106496 (llm foodscience water livingdocs cleanserve lang smallos researchreach sharehub, ten records) are all zero, and the publisher receipt and the nx_record_prose CLI agree on the two witness classes (sharehub, foodscience); the residue was the sota page's Measured bindings paragraph and sharehub's two head-to-head blocks, now nx_compare_record_lib tables 2026-09-23 land THE OWED PUBLISHER REFEREE EXISTS: nx_compare_regen_gate (buildroot/runtime, both trees) drives the PROMOTED nx_compare_regen and nx_compare_regen_scoped on a /tmp fixture tree in --domain mode and proves the records receipt against an independent rct_census over the bytes each run published: planted meth-block record reads records_prose=1 == census, the clean page 0 == 0 (gv_bite both ways), the promoted nx_swcompare_matrix's real sota page (48010 B) reads 0 == 0 with forkfresh=1, the scoped twin reads 1 == 1, api.json lines carry no records field, summary records_pages=1 records_uncal=0; the gate doubles as the fixture generator through a symlink at the fixture's _offc/ (verdict 0, json and html above the publisher's floors, export 2) so no second organ was built; admission refusal (exit 4) reads SKIP. 22/22 GREEN on the NAS (job 1790129414), SHIPPED 6788e559 through nx_organ_ship (PROVE self 22/22, first promote, offc twin, registered), declared in knowledge/organ_gate.conf for nx_compare_regen and nx_compare_regen_scoped (the nx_forkfresh_lib row keeps its own proof) and on cms.gates. First cut read 19/22: the generator refused a sota fixture with no bound rows and my echo filter hid its words -- the fixture now mirrors nx_sota_drift_gate 1:1 and any nonzero publisher exit echoes its whole stdout 2026-09-23 lesson A LINE FILTER OVER A PUBLISHER'S STDOUT HID THE GENERATOR'S REFUSAL AND SHOWED ITS DOWNSTREAM SYMPTOM: leg C first read ATOMIC-FAIL code=-107 syscall=-2 on the scope receipt, one hop after the real event (the promoted generator refused a sota fixture with no bound rows in its verdict mode, and GATE RED plus the generator's own words were not among the echoed needles). The gate now prints the publisher's whole stdout on any nonzero exit; a gate that reports a failure without the diagnostic the subject already captured is unactionable 2026-09-23 queue OWED, off this seat's tool surface: the subject-level mutation bite of the publisher through the new gate -- nx_gate_bite nx_compare_regen_gate buildroot/runtime/nx_compare_regen.nx 2 subject nx_compare_regen aim rg_records_receipt (the aim keeps the mutants inside the receipt function; every mutant build deploys the publisher to the serving root, so run it in a quiet window with compare_regen.progress reading domain=DONE and the harness restore verified by hash). The fixture-level bite (planted 1 / clean 0) is proven inside the gate; this is the second, subject-level proof 2026-09-25 measure BENCHMARK (operator, 2026-09-25): the Claude Code changelog [claude-code-changelog-2026-09] against this board, both captured in a real browser at both doors and graded by the same instruments. Pixel referee: ours DEFECTS on desktop by HIERARCHY -- the OVERVIEW h2 paints 40 px over a 37 px h1 -- theirs clean; contrast clean on both (worst 5.20 ours, 4.85 theirs). Box map, first viewport: longest text block 759 characters on ours against 147 to 189 on theirs, list items 0 against 3 to 7, text boxes on desktop 27 against 89 (their section rail, tabs and a one-link-per-release contents rail); on phone both titles land at 192 css px, their first release record at 577, our first section heading at 652 with the 759-character block beneath it. Bytes ruler: ours ABOVE theirs (hygiene 12 against 11, craft 10 against 7) -- it counts ingredients, and none of the qualities that separate the two pages is a count it takes 2026-09-25 lesson a ruler that grades our page above a benchmark it visibly trails is measuring presence, not the reader's task: the gap is scannability and wayfinding -- block length, list share, a contents rail, the current item marked, identifiers as code -- and every one of them is measurable from the box map the capture already writes. Grow the ruler; never tune a verdict to agree with an eye, including this seat's (its first phone reading, that our navigation ate the first screen, was refuted by the box map: both titles land at 192 css px) 2026-09-25 queue every ledger record renders the way the benchmark renders a release: an anchored chip (rung and date), the text as short bullets, identifiers as inline code; done when the box-map census reads no block in the first viewport longer than the benchmark's longest (189 characters, measured 2026-09-25) and the first record on phone no lower than the benchmark's (577 css px)

CM22 Relationship dossiers per rung · cell LANDED · gate UNRECORDED

Contract	`sc_rung_dossier` · executor Organ · 3 u · stands on CM21
Definition of done	A rung's plan row, matrix row, gate row, log rows, cited refs and design rows join by symbol into one dossier linked from the matrix cell and the rank table, so related information is read in one place [ia-arxiv-2026-09]; done when every rung on a board carries a dossier with its matrix cell and its gate verdict and both links resolve
Matrix cell	Relationship dossiers per rung · `runtime/nx_compare_dossier_lib.nx` · LANDED decl
Gate	`nx_compare_dossier_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Design rows	GD4 Related records join by symbol into one dossier and link both ways proposed
Work log	2026-09-27 lesson the NAS copies of nx_swcompare_matrix (+93 B, a sibling's pm plan surface), nx_swcompare_evidence (390 NAS-only lines: a TjBuffer stamp writer and a sealed gate binding) and four of their imports were AHEAD of the laptop and were pulled and adopted before a byte was edited; the evidence organ does not build on the laptop at all (stale sibling libs), so its edit was applied to the NAS generation by unique-line insertion and proven by the NAS build (cb4a37e7); a python heredoc through the harness mangles backslash-n anchors -- edit scripts live in files; the matrix generator SEGFAULTS on a fixture root where the estate root is absent (nx_catalog_lib cl_open_estate -> cl_estate_path with a -1 path pointer, laptop-only: the NAS run of the same fixture emits 92 KB), so the dossier gate's end-to-end leg is need-guarded and proves itself where the generator resolves beside it 2026-09-27 queue (1) the per-gate verdict record (nx_swcompare_evidence writing knowledge/status/evgates_<dom>.tsv, evg_path/evg_resolve in nx_evprofile_lib) is BUILT on the NAS (candidate cb4a37e7) and REFUSED by its release scope gate (COMPARE-RELEASE-REFUSED reason=changed-closure: knowledge/gates/nx_swcompare_evidence.release.scope pins every closure file with a decision and an explicit review reference) -- needs-intervention for the compare referee's release owner: re-adjudicate the changed closure with a review record, then ship; until then every dossier reads gate UNRECORDED and evgates UNSTAMPED by name, which is the honest state; (2) the laptop-only generator crash on a fixture root without the estate root deserves a null guard in nx_catalog_lib; (3) the sota generator renders no dossiers yet (matrix-class boards only); (4) a symbol two rungs share links its cell to the first rung only 2026-09-28 land nx_swcompare_matrix SHIPPED 603de9cd through nx_organ_ship (build, 7 of 7 declared gates proven on fresh closure banks including nx_compare_dossier_gate 24/24, content ruler RED adjudicated: 8 lost runs = 2 literal splits of the capability row and its cell close by the new id and dossier link, 6 supersessions by sibling lanes, loss token 7c7cff3a; behaviour probe on cms html DIFFERS STRONG 514186 to 705311 B, first diff at the dossier stylesheet rule, reviewed under NISHI-SHIP-BEHAVIOR-REVIEW/1 record b05d4039; promoted, offc twin installed, registered); nx_compare_dossier_gate SHIPPED d43f180e on its own lane (first promote, 24/24, registered) and declared on cms.gates; organ_gate.conf rows nx_compare_dossier_lib and nx_swcompare_matrix -> nx_compare_dossier_gate. The per-gate verdict record (nx_swcompare_evidence writing knowledge/status/evgates_<dom>.tsv beside its stamp) is BUILT-UNPROMOTED: the NAS build cb4a37e7 is refused by nx_swcompare_evidence_scope_gate as changed-closure until the release owner re-adjudicates knowledge/gates/nx_swcompare_evidence.release.scope, so every dossier reads its gate verdicts UNSTAMPED until that ships (needs-intervention, release owner; the seat does not forge the review) 2026-09-28 lesson the first live publish rendered all 30 dossiers NO-GATE-DECLARED for organs whose gates the ship loop proves every day: buildroot/knowledge/organ_gate.conf is a 354-byte stub carrying one sibling row, the regen runs from buildroot, and a CWD-first ladder returned the stub and never looked at the 44 KB declaration file at the estate root -- the two-tree trap, arriving through a file the compare emitters had never read before. A declaration is a fact and two lists of facts add, so the dossier lib now reads the UNION of the CWD copy and the root copy (rows deduplicated byte for byte, identical copies counted as one) and prints which copy answered with how many rows in the section itself; the evgates verdict record keeps its stamp's ambiguity rule because a verdict is a measurement and two different measurements refuse. Gate teeth: a stub beside the publisher cannot hide the root, no row twice, a copy without a trailing newline, comments are not declarations -- every expected count derived from the fixture texts after the NAS magic ratchet refused the first draft's hand-counted lengths (the estate's own law, paid for again) 2026-09-28 queue the organ_gate.conf census (nx_shelltool over buildroot/runtime, 26959 files, coverage complete): the readers are nx_organ_ship_lib (OSL_CONF_DEFAULT, read from the ship's working directory, the estate root), nx_rungclose (RP_GATE_CONF, a clock-stage organ whose working directory decides which copy it reads) and the dossier lib (now the union of both roots). The 354-byte buildroot/knowledge/organ_gate.conf stub is a sibling's declaration (nx_hostctl_supervise_gate) and must not be deleted; any reader that runs from buildroot and reads the CWD copy alone inherits the same blindness -- for nx_rungclose's owner to adjudicate (read both roots as the dossier lib does, or run from the estate root). The evidence organ's per-gate record stays queued behind its release scope (release owner) 2026-09-28 measure second ship: nx_swcompare_matrix SHIPPED a892f53b (content ruler GREEN, one-way gain of 11 runs, 7 of 7 declared gates including nx_compare_dossier_gate rebuilt 31/31 in the ship; behaviour on cms html 715960 to 717263 bytes reviewed under record 6039d4c9), nx_compare_dossier_gate SHIPPED e6f0ce71 on its own lane (31/31); scoped regen published cms index.html 720283 bytes (prev 709183), api.json 117188, matrix generator fork FRESH at a892f53b, nx_page_verify GREEN with 0 a11y issues. The dossier section's own provenance line: organ_gate.conf read from this working directory (1 rows) and from the estate root (154 rows), rows in both counted once 0, declarations read 155. Foot partition: dossiers=30 cells_joined=30 no_matrix_row=0, gates_recorded=0 gates_unrecorded=14 no_gate_declared=16 (sum 30), evgates=UNSTAMPED -- the 14 name their declared gate and wait for the evidence organ's per-gate record; the 16 are organs with no organ_gate.conf row (some because the declaration names the program while the matrix row names its lib, nx_cms_snapshot against nx_cms_admin's nx_cms_snapshot_gate, for their owners). nx_compare_rank cms: 29 of 30 done, open CM18, finish states 4 (REGISTERED-DARK: CM10, CM13, CM24, CM26, all laptop-executed browser or fetch organs the NAS actlog window cannot see) 2026-09-28 lesson a gate promoted at the serving root proves the INCUMBENT: the judge's ship ran its declared gate from the root (the closure bank said the root copy IS the fresh build, so no rebuild) and the gate's sibling ladder (<selfdir>/<name>.sov.elf, then <name>.elf) found the LIVE judge beside it, not the candidate the ship had staged as <name>.sov.elf.new -- 39/41 with exactly the two verb teeth failing on a judge that did not carry the verb yet. The dossier and layout gates carried the same ladder and had passed by luck (their subjects' ships happened to rebuild the gate into buildroot/_build, where the fresh subject sits). All three ladders now try the staged candidate (<name>.sov.elf.new beside the gate and at the root above buildroot/_build) before any promoted twin, re-shipped: nx_page_pairwise_gate 588446bd, nx_compare_dossier_gate 02798af7 (31/31), nx_nishi_layout_gate 9e6bdad0 (30/30). The general law: an end-to-end gate that forks its subject must prefer the artifact under test over the artifact in service, or a green proves the wrong binary

CM23 Design modes as data: composed and organic · cell LANDED · gate UNRECORDED

Contract	`dmo_mode_load` · executor Organ · 1 u · stands on CM20
Definition of done	Two modes declared in knowledge/design_modes.conf, the composed mode (a designed path, framed sections, a summary first, a rail, disclosure) [wiki-japanese-garden] and the organic mode (title, body, related; restraint, no chrome) [wiki-wabi-sabi], selected per surface by a design row on the blueprint or the board; done when the site emitter renders a home page in the composed mode and a notes page in the organic mode from the same conf, gate-proven
Matrix cell	Design modes as data · `runtime/_hdl_build/nx_design_modes_lib.nx` · LANDED decl
Gate	`nx_design_modes_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Design rows	GD6 Two named modes, composed and organic, chosen per surface as data proposed
Work log	2026-09-27 land design modes as DATA: knowledge/design_modes.conf (mode composed/organic, 8 facets nav framed summary band cta footer related disclosure, default home/notes) read by nx_design_modes_lib (dmo_mode_load refuses no-such-mode / facet-undeclared / facet-not-0-or-1 by name; dmo_mode_for = blueprint mode 2026-09-27 measure byte-identity control: the composed defaults rebuild the incumbent alpha pages IDENTICAL on all 3 pages (stage=alpha; the only earlier DIFFERS was the noindex meta the loop adds at stage=alpha, not the modes); gate 22/22 laptop and NAS; the modes conf lands in knowledge/ on both hosts and in buildroot/knowledge 2026-09-27 lesson a conf read by cwd alone is a gate-host dependency: nx_ink_derive_gate runs the emitter inside fixture roots where design_modes.conf is absent and read 34/39 until the conf was resolved on a ladder derived from the executable (cwd, exe/../knowledge, exe/knowledge, exe/../../knowledge) -- the same law as the trust-store ladder (dataio DI20), arriving through the emitter 2026-09-27 queue the disclosure facet is declared 0 on both modes and OWED (no emitted page carries an expandable section yet); no blueprint opts into organic yet, so the alphas are unchanged by construction; an organic page goes under the operator's eye and the pixel referee before any client site declares mode

CM24 Paper behind the glyphs · cell LANDED · gate UNRECORDED

Contract	`cd_textfree_shot` · executor Organ · 2 u · stands on CM6, CM13
Definition of done	The capture driver takes a second shot of every door with every text colour painted transparent, so the referee reads the paper each glyph pixel was painted over instead of estimating it from a mixed histogram: today the exact method proves solid paper only, and text over a photograph or a gradient keeps the conservative band means, the case the field's pixel checker also only warns on [atf-textcontrastcheck]; done when the eyebrow over the production hero photograph is graded against the pixels behind its own glyphs and still fails, a gradient running from the paper toward the ink no longer reads as solid paper, and a solid-paper box reads the same ratio both ways
Matrix cell	Paper behind the glyphs (a text-free capture the pixel referee reads the paper from) · `runtime/nx_cdp_capture.nx` · LANDED decl
Gate	`nx_cdp_capture_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-25 land THE PAPER BEHIND THE GLYPHS, SHIPPED: the capture driver shoots every door a second time with the text painted transparent (a style the page is given and loses; the page restored or every later door refused by name) and writes <label>.paper.png plus two receipt fields; the referee reads it and grades the declared ink against the measured paper by WCAG's failure technique F83 [wcag-understanding-1-4-3] -- the paper most like the text behind any glyph pixel, and when that fails, letter by letter (a letter is an 8-connected component of the pixels the two shots disagree on). nx_cdp_capture 63b815b8, nx_cdp_capture_gate 41024282 (27 of 27 on the NAS, 39 of 39 on the laptop with the browser leg), nx_page_pixjudge 211a29cc, nx_page_pixjudge_gate 38dd9194 (60 of 60 on the NAS, 67 of 67 on the laptop); mutants at the blend-check, unexplained-count and core-counter sites KILLED; both gates rostered. Done-rule met on its fixtures: a gradient running from the ground toward the ink passes at 4.91 by the exact method alone and fails on its own paper 2026-09-25 measure production gospelofwest home, captured with the text-free shot and graded: the h1 over the hero glow FAILS by the letter it sits over -- worst letter 2.71 desktop and 2.67 phone against the 3.0 large-text bar -- where the band means had read 5.78 and 5.07 (a defect the averaging hid); the eyebrow PASSES on desktop (quick check 5.18, worst letter 5.77, where the band means read 3.64 -- the anti-aliasing bias) and FAILS on phone (quick check 2.48, worst letter 2.60); every solid-paper box reads its declared ratio exactly, the paper method agreeing with the exact method on all 17 boxes of the ink fixture 2026-09-25 lesson two capture-driver defects found by using the driver, both fixed at the root: (1) a browser error page is complete and loaded like any page, so a certificate interstitial was captured and graded as the site with verdict OK -- the driver now reads Page.navigate's own errorText and refuses by name, and readiness requires an http, https or file document; (2) a WARM browser answers /json/version before its first tab exists, so a capture launched right after another read an empty target list -- the list is now polled on the boot budget. The first was a vantage artefact (the laptop's network was in transition; the edge's certificates verify), found only because the capture checked its own pixels 2026-09-26 land the dark contract is lit: the text-free shot is its own entry point, cd_textfree_shot in nx_cdp_capture (the style on, the paper shot through the same writer as the page shot, the style off, the restore reported), extracted with no change to a byte of output; nx_cdp_capture 3cb33eb3 promoted with lost_from_live=0, _offc twin installed, nx_cdp_capture_gate 27 of 27 on the NAS and 39 of 39 on the laptop with the browser leg

CM25 Award and UX rubrics as referee axes · cell LANDED · gate UNRECORDED

Contract	`rub_rubric_score` · executor Organ · 2 u · stands on CM6, CM14
Definition of done	knowledge/design_rubrics.conf carries the juries' axes and weights [awwwards-evaluation-2026-09] [webby-judging-2026-09] [fwa-about-2026-09], NN/g's ten heuristics [nngroup-heuristics-2026-09] and Baymard's benchmark topics [baymard-research-2026-09] as data, and the rubric referee maps every axis to the estate referees that measure it (pixel referee, tokenref, bytes ruler, a11y-lite, funnel floor) printing UNMEASURED for the rest; done when the alpha and a mirrored winner are both scored on every declared axis, the mapping names the unmeasured axes, and a planted contrast regression lowers the design axis
Matrix cell	Award rubrics as referee axes · `runtime/_hdl_build/nx_page_rubric_lib.nx` · LANDED decl
Gate	`nx_page_rubric_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped) `nx_page_pairwise_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-26 queue the research the operator handed the board is landed as data, not prose: fourteen refs pinned from mirrored primary pages and arXiv API listings read this month, knowledge/design_rubrics.conf (Awwwards D40 U30 C20 Cn10, the Webbys' seven, the FWA's bands, NN/g's ten, Baymard's topics), the bars SB-RUBRICS, SB-RSI-LOOP and SB-UIOPT with their targets T-BIC-RUBRIC and T-FRONT-LOOP, milestone M10 with CM25 to CM29; CSS Design Awards is UNFETCHABLE by the estate client today (TLS 1.3-only host) and its rubric is not landed; the paste's FWA axis labels were not in the primary bytes and are not landed either 2026-09-27 measure the served-page judge on the LIVE alpha https://nishifamily.com/alpha/gospelofwest/ (journey gospelofwest, doors desktop+phone, laptop browser host; referees nx_cdp_capture, nx_page_pixjudge, nx_ui_audit over the wire, nx_page_verify): signals measured 9/10 -- pixel.contrast 1000, hierarchy 1000, overlap 1000, overimage 1000, bytes.craft 933, hygiene 1000, journey.complete 1000, friction 1000, a11y 1000; tokenref UNMEASURED by name (a served page carries no token declaration). awwwards 992 over covered weight 70/100 (design 986, usability 1000; creativity and content UNMEASURED no-map-row), webby 995 over 4/7 axes, nngroup 983 over 2/10, baymard 966 over 1/6. Receipt knowledge/status/rubric_alpha_gospelofwest_2026-09-27.receipt 2026-09-27 measure the SAME judge on the live rival https://www.nike.com/ (journey rival-nike: one land step whose expectation is the page's own name; a rival capture conf with a 45 s settle because the estate's 15 s timed out on the cold first visit, CD-SETTLE-TIMEOUT, and the warm capture settled in 3.1 s): 9/10 measured -- pixel.contrast 894 (one failing box per door), hierarchy 1000 (the desktop door carries no h1 and reads hierarchy UNMEASURED, so it is out of that denominator), overlap 1000, overimage 789 (four text boxes over photographs on desktop), bytes.craft 333, hygiene 750, journey 1000/1000, a11y 250 (one image without alt, seven heading-level skips, no main landmark). awwwards 795 over 70/100 (design 803, usability 786), webby 826, nngroup 833, baymard 666. Receipt knowledge/status/rubric_rival_nike_live_2026-09-27.receipt. The mirrored bytes alone (rf_20699706.raw through nx_ui_audit file) measure 2/10 signals and read awwwards 400 over 40/100 -- a mirror without a browser names eight signals unmeasured, the live judge names one (receipts rubric_mirror_nike_home_2026-09-21 and rubric_mirror_awwwards_sotd_listing_2026-09-21) 2026-09-27 lesson three laws from the referee: (1) a rubric axis is the MEAN of the estate signals that measure it and prints UNMEASURED by name when none can, never 0 -- creativity, innovation, content and overall experience have no estate referee and the map says so; a platform number carries its covered weight beside it (992 over 70/100 is not 992 over 100). (2) every pixel signal carries its own denominator, summed only over the doors that measured it: a door reading hierarchy UNMEASURED contributes neither its violations nor its text boxes, or the share is a distance measured from a region that does not exist (the 2026-09-06 flood law in referee form; the fixture's phone door is exactly that case, 913 not 937). (3) a served-page judge must not end at the first refused stage: the first cut returned on capture:refused and the bytes ruler and verifier never ran, so a rival whose cold load outlived the settle budget measured 0/10 -- now every stage runs, each banks as it finishes, and why names the refused stages comma-joined; the scorer names the unmeasured, an early return manufactures them 2026-09-27 lesson the refine gate went 104/105 the moment CM23 rebuilt nx_site_build: the laptop policy pin (CM29) named the OLD evaluator sha and the loop refused REFUSED-TAMPER:build exactly as designed; the re-pin came from the gate's own fresh converging receipt under the new referees (PIN ok seal=2e7b6e42, the previous policy banked to refine_policy.conf.history) and the gate read 105/105 -- a pin that outlives its referee is the alarm, not the defect. The rival landing journey first read INCOMPLETE because its step label was prose and the driver's land expectation IS the label text (the needle in the visible text); the page's own name is the honest expectation, and the journeys conf carries it as data 2026-09-27 land nx_page_rubric_lib (rub_signals_from_dir, rub_axis_score, rub_rubric_score: the ten signals, an axis = the mean of its measured mapped signals or UNMEASURED by name, platform roll-ups with covered_weight and axes_measured) + knowledge/design_rubric_map.conf (map 2026-09-27 queue (1) a fixture-referee tooth for sr_judge_url in the rubric gate (today proven live on two URLs, not in a gate); (2) the NAS has no browser leg, so the served-page judge there reads capture:refused PARTIAL (audit + verify only) -- a browser host for the NAS is the standing gap behind CM13, CM6 and this rung; (3) creativity, innovation, content and overall experience stay UNMEASURED until a referee exists: CM18's pairwise referee with a human-judged set is the road to the first of them; (4) the rival journey is one land step; a rival's 1-2-3 path needs its own step rows before journey.friction means the same thing on both sides; (5) tokenref reads UNMEASURED on every served page: the token declaration could ride the page (a meta row the emitter already knows) so the served-page judge can measure it

CM26 Journey driver measures the 1-2-3 path in the browser · cell LANDED · gate UNRECORDED

Contract	`cd_journey_run` · executor Organ · 2.5 u · stands on CM13
Definition of done	knowledge/journeys.conf declares each site's path (landing, action, completion) as steps the capture driver walks at both doors, recording clicks, fields, scroll depth and settle time per step and grading the receipt against the research rows (error prevention, visibility of status, checkout and form rows) [nngroup-heuristics-2026-09] [baymard-research-2026-09] [builtformars-2026-09] [pageflows-2026-09]; done when the gospelofwest and cms-scene alpha paths complete with a friction receipt per step and a planted extra step reads worse
Matrix cell	Journey driver measures the 1-2-3 path in the browser · `runtime/nx_cdp_capture.nx` · LANDED decl
Gate	`nx_cdp_capture_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-26 land THE JOURNEY DRIVER, SHIPPED: nx_cdp_capture journey <name> <out> <journeys.conf> walks a site's declared path at every door of the capture conf -- land, click <selector>, fill <selector>, expect <needle> -- recording per step whether the target was found, whether it sat in the viewport at the moment of the click, how many viewport heights down it sat, the settle time after it (a click that navigates settles the new document too) and whether the expectation held in the href or the visible text, then the actions a user pays (clicks + fields + scrolls) against the journey's declared bar; INCOMPLETE, WORSE-THAN-BAR or OK per door, the worst door decides, every row in journey.receipt. One session opener now serves the capture and the journey (cd_session_open), the decisions live in nx_cdp_lib (cdv_journey_verdict, cdv_pipe_field, cdv_atoi_z, cdv_step_kind) so the gate proves them in-process. nx_cdp_capture f01a4ada promoted with lost_from_live=0, gate 43 of 43 on the NAS and 59 of 59 on the laptop where the e2e legs run: the fixture path OK within its bar, a planted extra step WORSE-THAN-BAR, a missing target INCOMPLETE naming the step 2026-09-26 measure knowledge/journeys.conf declares two paths with the funnel floor's own promise as the bar (3 actions): gospelofwest (land, Start reading, first theme card) reads OK on both doors at 2 actions (2 clicks, 0 scrolls, settle 675 and 414 ms); cms-scene (land, See the scene, first note card) reads OK on both doors at exactly 3 actions -- the scene band pushes the first card one viewport below the CTA's landing, so the path costs a scroll on every door; the friction receipt names it, nothing is tuned 2026-09-26 lesson a numeric field read as 0 on every step while the label beside it read the clicked text -- cdv_atoi returns the index after the digits, never CDV_OK, and a reader that compares to the wrong contract silently counts nothing (the direction that under-reports friction); the fix is the contract read from the lib, and the gate now holds atoi to its arithmetic. And an in-page action that navigates leaves the OLD document complete for a moment: a settle poll that returns at once on readyState measures nothing, so the driver re-reads the href after the click and settles the new document when it moved 2026-09-26 land the file door walks the links a served root walks: nx_cdp_capture's click script returns the anchor's own resolved target beside the document href, and a target the browser could not load (chrome-error) is resolved to <x>.html then <x>/index.html in that order, settled and announced (CD-JOURNEY-FILE-FORM ... landed=1), so a file:// journey completes on a clean-URL sub-page link exactly as https does (edge E6 names the served mapping); the walker carries the board's contract name cd_journey_run; the journey verb's usage line said conf=... while the driver reads a bare fifth argument (usage fixed). nx_cdp_capture_gate 62 of 62 on the laptop (a page linking `sub` beside sub.html completes, the receipt names sub.html, a link to a .html file never triggers the resolution), 43 of 43 on the NAS (8c2bdf08)

CM27 Variant archive and selection in the shape of the Darwin and Mendel Godel Machines · cell LANDED · gate UNRECORDED

Contract	`sr_archive_select` · executor Organ · 3 u · stands on CM17, CM26
Definition of done	Every emitted variant (blueprint, tokens, section order, presets) is archived content-addressed with its referee scores and journey receipt; the loop chooses the next parent by score with a novelty bonus and proposes a modification from the referees' named defects and the archive's comparative signals, admitting a child only when it beats its parent on every deterministic referee and on the journey receipt at a declared confidence [arxiv-godel-machine-2026-09]; done when a planted defect is repaired within a declared iteration budget, a page that cannot converge is refused by name, and the archive keeps every variant
Matrix cell	Variant archive and selection in the shape of the Darwin and Mendel Godel Machines · `runtime/_hdl_build/nx_site_refine_lib.nx` · LANDED decl
Gate	`nx_site_refine_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-26 land the archive and the selection rule (sr_archive_select): variant

CM28 Multivariate optimisation of the design knobs · cell LANDED · gate UNRECORDED

Contract	`sb_knob_search` · executor Organ · 3 u · stands on CM27
Definition of done	The knobs are the derived-token bands and the catalogue's discrete choices (section order, presets, hero mode, density); a factorial search over them is scored offline by the referee panel and the journey receipt, and online nx_cms_abtest assigns live traffic to the top variants with each journey step's completion logged and judged by Bayesian A/B arithmetic [arxiv-ui-optimization-2026-09] [arxiv-abtesting-web-2026-09]; done when the search prints its objective per variant, the chosen variant beats the incumbent at the declared confidence, and a knob outside its evidence band is refused
Matrix cell	Multivariate optimisation of the design knobs · `runtime/_hdl_build/nx_site_refine_lib.nx` · LANDED decl
Gate	`nx_site_refine_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-27 land THE KNOB SEARCH, SHIPPED (offline): knowledge/refine_knobs.conf declares token knobs (levels inside a named evidence band) and row knobs (catalogue choices) plus a calibrated evaluation budget; sb_knob_search sizes the full factorial before the first evaluation and refuses REFUSED-BAND:<knob>=<level> outside <band>, REFUSED-KNOB (a token the page lacks) and REFUSED-BUDGET plan+incumbent=N budget=B by name, then emits and judges every plan point with the loop's referees, prints variant rows (levels, id, Pareto admission against the incumbent with the axis, the vector) and archives them beside the loop's variants, choosing the best admitted by the objective or letting the incumbent stand (a completed search); a plan point identical to the incumbent reads INCUMBENT-SAME and is never re-evaluated. THE ONLINE HALF: nx_abtest_bayes_lib computes P(challenger beats incumbent) for two Beta posteriors by the exact closed form in the log domain over a log-factorial table on the bits-up f64 family (bit-identical on every host), and nx_site_refine abtest <experiment> <ledger> tallies the visitor ledger (hit rows per arm) against knowledge/refine_abtest.conf (incumbent, challenger, confidence permil, Beta prior) and journals CHALLENGER-WINS, INCUMBENT-HOLDS or UNDECIDED beside the ledger; nx_abtest_bayes_gate 26 of 26 SHIPPED 516e2621 (registered, offc twin, organ_gate row) 2026-09-27 measure four independent methods agree on the probability: exact rationals (Beta(2,1) against the uniform reads one third, Beta(2,1) against Beta(3,1) reads three fifths, delta 0 at 1e-12), the symmetry P(B>A)+P(A>B)=1 at a hundred visits per arm (delta 0), a log-free product evaluation of the same sum (delta 0 at 1e-10) and a 4000-cell grid integration sharing no arithmetic with the lib (delta 0 cells); 30 of 100 against 40 of 100 reads 930 permil, UNDECIDED at 950 and CHALLENGER-WINS at 900, 300 of 1000 against 400 of 1000 reads 1000 permil; an empty ledger reads 500 and stays UNDECIDED under every honest bar, a bar at or below 500 is REFUSED-BAR because an empty ledger would pass it. The e2e search on the laptop: a two-level plan over the display token read plan=2 evaluated=1 admitted=1 (the planted level is the incumbent's own point, the level one referee step up cleared the scale-step axis) and CHOSE it; a budget of one refused REFUSED-BUDGET plan+incumbent=3 budget=1 before any evaluation 2026-09-27 lesson a set of a token to the value it already holds must be a byte no-op, or the plan re-evaluates the incumbent under a note and calls it a variant (the first e2e read evaluated=2 same=0 with a REJECTED-NOGAIN twin of the root; fixed at the token setter, not by comparing vectors); the loop resumes from an existing archive by design, so a gate that inherits its last run's archive converges at iteration 0 on the old child and measures the predecessor (77 of 81 before the loop dirs were reset at setup); the magic ratchet named the two fixture literals the new teeth had spelled (50000 and 12840, kind FIXTURE) and both were derived instead (the level read back from the fixture row, the bar move from its declared const); a cap transcribed from a summary is a forged cap (the cms lane cap's nonce differed on disk) and nx_filecopy takes _cap as a sibling like every row 2026-09-27 queue the online half is arithmetic, reader and verb proven on fixture ledgers; the LEDGER PRODUCER is not built: a served site must assign arms through nx_cms_abtest (deterministic, no PII; today it has no importer but its gate) and write hit rows on journey-step completion, which is a serving-daemon change through its RACI lane; until it lands the done-rule's declared confidence is met on fixtures, not on live traffic. Also owed: a search over row knobs across tiers (editorial beside showcase), knob levels derived per site from its own tokens rather than the fixture's absolute values

CM29 Guarded self-improvement of the improver · cell LANDED · gate UNRECORDED

Contract	`sb_policy_pin` · executor Organ · 2 u · stands on CM27
Definition of done	The loop's proposal policy is a versioned knob set evolved the same way under a hash-pinned policy (goals, scope, evaluation identity, budget, release conditions) that the loop cannot edit, with the referees in another lane, harness tampering audited and the judge held to advisor [arxiv-selfimproving-2026-09]; done when a mutation that edits a referee or the policy pin is refused by name and every accepted policy change carries its held-out gain
Matrix cell	Guarded self-improvement of the improver · `runtime/_hdl_build/nx_site_refine_lib.nx` · LANDED decl
Gate	`nx_site_refine_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-27 land THE POLICY PIN, SHIPPED: the refine loop's evaluator (the four referees) and mutation operator (the remedy table) are pinned by sha256 in knowledge/refine_policy.conf (pin rows per kind, a gain row naming the held-out receipt and its root and best vectors, a seal over every non-comment row); the loop hashes what it resolved and refuses REFUSED-TAMPER:policy-seal (a pin row edited without the pin verb), REFUSED-TAMPER:<kind> (a referee or the remedy table moved, a forged seal included), policy-unreadable or policy-missing-pin:<kind> BEFORE any variant is judged, and stamps referee rows (kind, path, sha256) into every receipt; nx_site_refine pin <policy> gain=<receipt> mints a policy ONLY from a receipt that converged under exactly the live referees with an admitted child beating the root (refusals receipt-not-converged, receipt-no-gain, receipt-referee-mismatch:<kind>), banking the previous policy to <policy>.history; a referee conf without a policy row announces POLICY unpinned and proceeds (opt-in, never a fleet-wide red). Entry points sb_policy_pin and sr_policy_check in nx_site_refine_lib; nx_site_refine PROMOTED d193d81c on the NAS (contentdiff GREEN lost 0 gained 216, offc twin), nx_site_refine_gate SHIPPED 6f8d2fee (93 of 93 on the NAS, 105 of 105 on the laptop with the browser leg: pin from the real converging receipt, a moved pixjudge referee refuses before any variant is judged with no archive written) 2026-09-27 measure the laptop lane is PINNED in production: refine_policy.conf minted from the gate's converging receipt over the _build referees (seal c97b5496), declared in the laptop referee conf, and the pinned loop on the fixture read POLICY pinned ok then CONVERGED-RESIDUAL with one admitted move; the NAS referee conf stays UNPINNED and announces it, because a pin needs a converging receipt under the NAS referees and the loop's browser leg runs only where a browser row resolves; the rung's board symbol sr_policy_pin was repointed to the real entry point sb_policy_pin on both the rung and the matrix row rather than minting a name-only alias (the CM21 rule) 2026-09-27 queue owed: a NAS-side converging receipt so the NAS referee conf can declare its pin (needs the browser leg on a host with a browser row, or a served-root capture door); re-pin the laptop policy after any referee rebuild (the loop refuses by name until then, which is the point); the proposal policy evolved under the pin (the rung's second clause) is the knob search over refine_remedies.conf rows themselves, not yet built

CM30 Blind sides: Left and Right by a keyed coin, the click mapped back to the referee's A and B · cell PRESENT · gate UNRECORDED

Contract	`pjj_swapped` · executor Organ · 0.5 u · stands on CM18
Definition of done	The pair section names no candidate; which candidate sits left is the parity of the pair's nibble of the session token, the click (L, R, tie) is mapped back to A, B, tie before the referee's one writer runs, and the side shown is recorded on a meta row; done when the gate proves a known-answer token swaps exactly the predicted pairs and a Left click on a swapped pair lands as B on the referee's row [designarena-home-2026-09]
Matrix cell	Blind side-by-side pair judging surface (one pair at a time, Left and Right by a keyed coin, the click mapped back to the referee's A and B) · `runtime/nx_pairjudge_lib.nx` · PRESENT decl
Gate	`nx_pairjudge_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	2026-09-28 measure the bar for the judging surface read from this month: two arXiv listings (30 and 13 entries, digested through nx_arxiv_digest) and the Design Arena home whose served bytes carry its methodology strings (four random models per session, two randomly paired models presented anonymously side by side, a choice is one vote, winner and loser brackets with a tie breaker to a final ranking, Bradley-Terry denoted Elo from 1200 with a standard-error margin, win rate and placement); the LMArena WebDev arena is Code Arena now and its methodology post is a script shell (URL-witnessed, not cited); eleven rungs CM30 to CM40 under M11 with their roles toward T-BIC-PAIRJUDGE and T-FRONT-PAIRJUDGE 2026-10-01 measure nx_compare_rank cms after the October attestation: BARFRESH bars=8 fresh=2 attested=6 stale=0 unattested=0 verdict=FRESH; LADDER targets=9 rungs=41 placed=41 verdict=LADDERED; rungs=41 done=34 open=7 finish=4 (the four REGISTERED-DARK finish states are the laptop browser and capture organs CM10 CM13 CM24 CM26); critical path on target version 10.0 (M11): #1 CM34 pj_kripp_alpha (DEP-PULL as the prerequisite of CM36), #2 CM35 pjj_live_route, #3 CM38 pj_loo_stability, #4 CM40 pjj_tournament, #5 CM37 pj_axis_verdicts, #6 CM39 pj_judge_profile, #7 CM36 pj_cojudge_anchor; CM30 to CM33 read done because their symbols are declared in buildroot/runtime/nx_pairjudge_lib.nx (a758912a), proven 53 of 53 on the laptop and NOT yet promoted on the NAS (the ship is queued behind build admission at load1 28, runqueue 27 of 8), so their cells flip on this regen while the gate verdict record stays the laptop's until the NAS promote lands; the five open rungs the ranker prices first are being built in parallel by a five-agent workflow (operator: fan out on workflows), each a new lib and gate, the integration into the referee receipt and the surface owed after 2026-10-01 land CM30 to CM33 are proven on the NAS: the referee lib gained pj_rows_ratings, the one Elo ruler over judged rows (source 3adc3c8d, 31491 B), the surface's pjj_ratings is a one-line delegation to it (source bb9a51bd, 37846 B), nx_pairjudge_gate built on the NAS as 39c12c6b (217320 B), promoted, and /api/gate_run on the promoted binary reads GREEN exit 0 with 54 of 54 teeth against the staged daemon cefa7556; the daemon itself stays staged and undeployed 2026-10-01 lesson a gate run once green is not a deterministic gate: nx_pairjudge_gate read 53 of 53 on single runs and GREEN on only 11 of 40 consecutive runs, because three teeth asserted facts that depended on the random session token (a two-character fixture id matching inside the hex token in the pair section, in-process and over the wire, and a count of left-side markers taken over every judge's meta rows while the other judge's coin was random); the ship loop's PROVE stage caught it as 51 of 53 once build admission cleared; fixed at the root with a reader that ignores an occurrence lying wholly inside the token, a count over one judge's rows, and a planted-leak control, then 40 of 40 runs green; run a gate twenty times before shipping it 2026-10-01 queue the ship loop cannot observe a coupled gate and daemon redesign: the incumbent gate and the candidate resolve one daemon through the same ladder, the surface changed from every pair on one page to one blind pair, so the incumbent reads RED against the new daemon and the behaviour stage refuses complete-successful-representative-probe-required before any observation can be reviewed; this gate went through /api/promote with the exact sha after the loop's own BUILD and PROVE were green and its 31 lost runs were enumerated (every one superseded by the blind design: the old style block, the A and B section and buttons, hand-typed fixture rows and paths, renamed teeth, the old header and note); owed on the loop: an observation path for a superseded incumbent that fails for a named subject change 2026-10-01 measure triangulation of SB-PAIRJUDGE read from the mirrored listing's own comment fields (cmp_cms_arxiv_202609_arena_human_preference.xml, 80638 B): peer-reviewed witnesses K-Sort Eval 2602.09411 (ICLR 2026), Transitivity Meets Cyclicity 2605.17342 (ICML 2026), Personalized Benchmarking 2604.18943 (Findings of ACL 2026) and DR-Arena 2601.10504 (ACL 2026); September preprints from two separate groups ANCHOR 2609.19599 and WebCraftBench 2609.15387; Design Arena's served methodology as a vendor claim; three independent witness classes with a peer-reviewed paper among them and the month's listing read

CM31 Dwell recorded per verdict: the judge's time on the pair as gap evidence · cell PRESENT · gate UNRECORDED

Contract	`pjj_meta_row` · executor Organ · 0.5 u · stands on CM30
Definition of done	Every verdict carries a meta row (pair, judge, date, side shown, dwell in ms from the pair's last render, door) written by one writer and read by one reader, the judge's median dwell printed on the surface; done when the gate measures a dwell at least as long as a real wait between render and post and UNSHOWN for a pair posted without a render [arxiv-arena-preference-2026-09]
Matrix cell	Dwell per verdict as gap evidence (a meta row: side shown, milliseconds on the pair, door) · `runtime/nx_pairjudge_lib.nx` · PRESENT decl
Gate	`nx_pairjudge_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	no log row names this rung yet

CM32 Next pair by information: fewest judgments, then the smallest rating gap · cell PRESENT · gate UNRECORDED

Contract	`pjj_next_pair` · executor Organ · 1 u · stands on CM30
Definition of done	The surface serves one pair at a time, the open pair whose candidates carry the fewest judgments and, among those, the smallest gap in the referee's Elo over every judged row, then the lowest ordinal; done when a known-answer row set makes the zero-gap pair of two winners the next pair where the first open pair is the wrong choice, and a tie moves no rating [arxiv-2602-09411-ksort-eval]
Matrix cell	Next pair by information: fewest judgments then the smallest rating gap · `runtime/nx_pairjudge_lib.nx` · PRESENT decl
Gate	`nx_pairjudge_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	no log row names this rung yet

CM33 Standings on the surface: rating, W-L-T, win share with its Wilson interval · cell PRESENT · gate UNRECORDED

Contract	`pjj_standings` · executor Organ · 0.5 u · stands on CM32
Definition of done	The referee's one rating ruler (nx_dr_elo from its start rating) over every judge's rows ranks the candidates; each row prints wins, losses, ties, the win share of decided pairs and its Wilson 95 interval from the gate base class, UNMEASURED with no decided pair, behind a disclosure until the judge is done; done when the table's rows equal the values the gate recomputes from the same rulers [designarena-home-2026-09]
Matrix cell	Standings on the judging surface: rating, W-L-T, win share with its Wilson interval · `runtime/nx_pairjudge_lib.nx` · PRESENT decl
Gate	`nx_pairjudge_gate` · UNRECORDED — no recorded verdict for this gate (it is declared for the organ but the evidence referee's last run did not record it: it is not on this board's gates map, or the referee has not run since the per-gate record shipped)
Work log	no log row names this rung yet

CM34 Inter-judge agreement as Krippendorff alpha beside the pairwise agreement · cell WATCHING · gate NO-GATE-DECLARED

Contract	`pj_kripp_alpha` · executor Organ · 1 u · stands on CM18
Definition of done	The referee's receipt prints Krippendorff's alpha (nominal, over pairs judged by two or more judges) in permil beside the k-of-n agreement and its Wilson interval, in exact rational arithmetic; done when a hand-computed two-judge set over six pairs reproduces the textbook value and a single-judge set reads UNMEASURED by name [arxiv-2511-20513-designpref]
Matrix cell	Inter-judge agreement as Krippendorff alpha beside the pairwise agreement · `runtime/nx_pairwise_agreement_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_pairwise_agreement_lib`; declare `nx_pairwise_agreement_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

CM35 Live preview beside the stills: the candidate served on the surface · cell WATCHING · gate NO-GATE-DECLARED

Contract	`pjj_live_route` · executor Organ · 1.5 u · stands on CM30
Definition of done	GET /live/<k>/ serves the candidate's own index and the assets under its dir on the loopback surface so the judge interacts before voting, the way the arenas judge running apps; done when the gate fetches a candidate's page and one asset through the route, a path outside the dir is refused by name, and the pair section links both previews [arxiv-2609-15387-webcraftbench]
Matrix cell	Live preview beside the stills on the judging surface · `runtime/nx_pairjudge_live_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_pairjudge_live_lib`; declare `nx_pairjudge_live_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

CM36 Human-anchored machine co-judge with a gap-calibrated vote · cell WATCHING · gate NO-GATE-DECLARED

Contract	`pj_cojudge_anchor` · executor Organ · 3 u · stands on CM34
Definition of done	The referee's agreement with the human rows is published as a curve over the pair's score gap; the machine casts a vote only on pairs whose gap is above the smallest gap at which its agreement clears the floor, its sensitivity and bias per judge estimated from the rows; done when the curve is on the receipt, a pair below the gap reads ABSTAIN by name and the anchored ranking of a held-out set agrees with the human ranking above the floor [arxiv-2609-19599-anchor] [arxiv-2609-15387-webcraftbench]
Matrix cell	Human-anchored machine co-judge with a gap-calibrated vote · `runtime/nx_pairwise_anchor_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_pairwise_anchor_lib`; declare `nx_pairwise_anchor_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

CM37 Per-axis verdicts: aesthetics, usability, requirement alignment and overall · cell WATCHING · gate NO-GATE-DECLARED

Contract	`pj_axis_verdicts` · executor Organ · 2 u · stands on CM33
Definition of done	The surface asks for an axis verdict beside the overall (a row per axis in the referee's grammar, the axis a declared field), the referee keeps one Elo table per axis and names the axis on which the human and the machine disagree most; done when a set judged on three axes yields three rankings and the receipt names the widest disagreement [arxiv-2609-15387-webcraftbench]
Matrix cell	Per-axis verdicts: aesthetics, usability, requirement alignment and overall · `runtime/nx_pairwise_axes_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_pairwise_axes_lib`; declare `nx_pairwise_axes_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

CM38 Leaderboard robustness: leave-one-out and flip perturbation with rank intervals · cell WATCHING · gate NO-GATE-DECLARED

Contract	`pj_loo_stability` · executor Organ · 1.5 u · stands on CM33
Definition of done	The receipt prints how often the top candidate survives dropping each single row and flipping each single decided row, and the rank interval per candidate over those perturbations; done when a set decided by one row reads FRAGILE by name and a set decided by many reads STABLE [arxiv-arena-preference-2026-09]
Matrix cell	Leaderboard robustness: leave-one-out and flip perturbation with rank intervals · `runtime/nx_pairwise_robust_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_pairwise_robust_lib`; declare `nx_pairwise_robust_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

CM39 Per-judge preference profiles · cell WATCHING · gate NO-GATE-DECLARED

Contract	`pj_judge_profile` · executor Organ · 1.5 u · stands on CM34
Definition of done	The receipt prints each judge's agreement with the pool and with the machine and the pairs on which the judge departs from the pool; done when a planted contrarian judge reads as the lowest-agreement row and the pool's alpha drops by name when that judge is added [arxiv-2511-20513-designpref]
Matrix cell	Per-judge preference profiles · `runtime/nx_pairwise_profile_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_pairwise_profile_lib`; declare `nx_pairwise_profile_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

CM40 Tournament session: four candidates from one brief, brackets to a final ranking · cell WATCHING · gate NO-GATE-DECLARED

Contract	`pjj_tournament` · executor Organ · 1.5 u · stands on CM32
Definition of done	A session draws four candidates for one brief and runs the bracket the arenas run (two battles, winners and losers, a tie-breaker, a final ranking) on the surface, each battle a pair row; done when a four-candidate session produces a full ranking in five verdicts and the ranking equals the referee's Elo over those rows [designarena-home-2026-09]
Matrix cell	Tournament session: four candidates from one brief, brackets to a final ranking · `runtime/nx_pairjudge_tournament_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_pairjudge_tournament_lib`; declare `nx_pairjudge_tournament_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

dossiers=41 cells_joined=41 no_matrix_row=0 (sum 41) gates_recorded=0 gates_unrecorded=19 no_gate_declared=22 (sum 41 of cells_joined) evgates=UNSTAMPED

Capability comparisons

Compare the field, search individual capabilities and open their source and adoption evidence. Documented presence does not establish comparative quality.

Position map — centrality and distinctiveness

The four-quadrant map the field uses for brand strategy (Dawar and Bagga, HBR June 2015), re-derived from this matrix on every publish. Centrality is the share of the category's feature mass a player covers, each feature weighted by how many hold it; distinctiveness is the average lead over each rival on the rows the player holds; breadth is the depth-weighted share of the whole matrix (the bubble); depth is how deeply the rows held are held; momentum is the day-over-day move off the spine (green rising, red falling, grey until day two); the dashed path runs first day → previous day → today. Dividers are the category means. Axes are fitted to the field of play, so read the tick numerals, not the frame. Rival marks are documented presence, so a rival's position reads the record, never its quality. The picture grades its own readability below; the table beside it is the same data for a screen reader or a second method.

Rows. 60 feature rows mapped of 68 matrix rows; 8 substrate rows (kernels, bounds and instruments the plan's rungrole marks substrate) are the estate's scaffolding, not category features, and stay off the map

Views. 2D cut: centrality, distinctiveness · 3D cube: centrality, distinctiveness, breadth · axes registered: centrality, distinctiveness, breadth, depth, momentum · a board picks its own in knowledge/compare/cms.cdmap (cut|x|y, cube|x|y|z); absent = the HBR defaults

bubble area = breadth · ring = momentum (green rising, red falling, grey until day two) · dashed = category means · axes fitted to the field of play: centrality 750–1000, distinctiveness 0–500 of 0–1000 permil (the full range put every player in one corner)
readability of the 2D cut, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 0 · unresolved labels 0 · mark overlaps 0 (a fact of the data: two players that close are that close) · data spread 708 permil of the plot · quadrant words unseated 0
text contrast, measured with wcag2-ratio (floors from contrast.conf), light theme: labels 16.24 (floor 4.50) · notes 16.24 (floor 4.50) · quadrant words 5.89 (floor 4.50) · tick numerals 5.89 (floor 4.50) · axis titles 5.59 (floor 4.50) · dark theme: labels 13.78 (floor 4.50) · quadrant words 5.55 (floor 4.50) · axis titles 5.97 (floor 4.50) · dark classes under their floor 0 (one figure serves both themes: a dark shortfall is a token to fix, never a class to hide) · classes refused under their floor 0 (a refused class is not drawn; the scale classes are measured, never hidden) · export: SVG PNG (receipt, rendered by the estate's own rasteriser from this page)

axes fitted: centrality 750–1000 · distinctiveness 0–500 · breadth 200–800 permil · farther bubbles are painted first, nearer ones over them · the floor shadow is each bubble's (x, y) at height 0
readability of the 3D cube, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 0 · unresolved labels 0 · mark overlaps 7 (a fact of the data: two players that close are that close) · data spread 76 permil of the plot

Nishi WordPress Wix Webflow Drupal
10 panels over 5 registered axes, every pair once (the lower axis on x, the higher on y) · each panel fitted to its own field of play, first and last tick numerals shown · no labels in a small cut, the legend names the colours; the grade below reports the mark terms only (MM summed over the panels, spread averaged)
readability of the small multiples, self-graded by the layout ruler: label overlaps 0 · labels over marks 0 · off-canvas 0 · unresolved labels 0 · mark overlaps 12 (a fact of the data: two players that close are that close) · data spread 558 permil of the plot

Player	Quadrant	centrality	distinctiveness	breadth	depth	momentum	Rows held	Days on spine	First seen
Nishi	Aspirational	1000	366	611	691	0 since 2026-10-01	53	15	2026-09-15
WordPress	Peripheral	845	109	311	583	0 since 2026-10-01	32	15	2026-09-15
Wix	Peripheral	856	128	344	626	0 since 2026-10-01	33	15	2026-09-15
Webflow	Peripheral	845	146	355	646	0 since 2026-10-01	33	15	2026-09-15
Drupal	Peripheral	801	55	261	522	0 since 2026-10-01	30	15	2026-09-15

Day	Matrix rows reviewed	Players recorded
2026-09-16	30	5
2026-09-17	30	5
2026-09-18	30	5
2026-09-19	30	5
2026-09-20	30	5
2026-09-21	30	5
2026-09-22	31	5
2026-09-23	45	5
2026-09-25	45	5
2026-09-26	50	5
2026-09-27	50	5
2026-09-28	50	5
2026-10-01	60	5
2026-10-08	60	5

How to read this evidence. A Nishi source mark records whether the named symbol is declared in the selected source file. It does not establish working behavior or accepted delivery. A watching tag names an expected source entry; discovering it updates source coverage only. A dark tag means the file exists but the named symbol was not found: inspect the source and contract before assuming the capability is missing or renaming code. Peer codes are authored assessments; unassessed means no comparison is established, not absence. Qualification, deployment identity, runtime observations and acceptance must carry their own evidence. Adoption records the catalog's observed path through source, build, promotion, registration and invocation; libraries use importer evidence and gates use declared execution surfaces. A full catalog path does not establish quality, freshness or a successful user outcome. Partial and unknown states retain their stated next action. Census stamps: importers asof 1790637268, gate census asof 1790104131 (unix seconds; -1 = census absent).

Capability matrix — measured against source

◉ authored stronger claim; source present● present◐ partial○ absent? unassessed; excluded from relative comparison · click any capability for its evidence

Peer cell coverage: authored assessed 272, unassessed 0. Authored assessments are not verified outcomes.

Capability	Nishi	WordPress	Wix	Webflow	Drupal
Structured content store, portable text bundle, no SQL Measured: `cst_get` exists in `runtime/_hdl_build/nx_cms_store.nx`, verified at emit. Content serializes to a self-describing text bundle on your disk; rivals sit on MySQL or a proprietary cloud [wix-dev-docs] Adoption: `LIB-WIRED importers=23 nonval=8` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	●	●	●
Whole-site blueprint as ONE data file Measured: `sb_build_site` exists in `runtime/_hdl_build/nx_site_build.nx`, verified at emit. A .site line file emits index, pages, nav, sitemap, robots; WP splits theme plus database, Drupal config-sync is partial rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `REGISTRY-UNCERTAIN` — PARTIAL: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion. not adopted: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, DrupalCons not adopted yet: registry evidence is malformed, ambiguous or unverified: repair the registry before lifecycle promotion	●	◐	○	○	◐
Requirements brief compiles to finished copy Measured: `sa_req_load` exists in `runtime/_hdl_build/nx_site_archetype.nx`, verified at emit. A .req brief fills 10 archetypes with real copy, gate 6 of 6; no rival compiles requirements into content Adoption: `SOURCE-ONLY` — PARTIAL: source exists, never compiled: /api/build it. not adopted: source exists, never compiled: /api/build it Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, DrupalCons not adopted yet: source exists, never compiled: /api/build it	●	○	○	○	○
Design-token SSOT, retheme without markup edits Measured: `sk_head` exists in `runtime/nx_sclass_kit.nx`, verified at emit. Tokenized shell rethemes by swapping values; WP theme.json and Webflow variables are the comparators [webflow-dev-docs] Adoption: `LIB-WIRED importers=10 nonval=9` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of Wix, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	◐	●	◐
Template corpus consumed from real open templates Measured: `ti_classify` exists in `runtime/_hdl_build/nx_tpl_ingest.nx`, verified at emit. 28 deduped structural patterns crawled sovereign; WP 12k-theme breadth is the reference rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `REGISTERED-UNAUTHORISED` — PARTIAL: registered, no cap ever minted; no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked). not adopted: registered, no cap ever minted; no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) Pros Implementing source symbol found; runtime acceptance separateCons behind WordPress (leads); not adopted yet: registered, no cap ever minted; no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked)	●	◉	●	●	●
Content snapshots and revisions Measured: `snap_read_file` exists in `runtime/_hdl_build/nx_cms_snapshot.nx`, verified at emit. WP N-deep revisions with restore UI is the reference; ours is a snapshot organ plus one-step prev backups rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-WIRED importers=4 nonval=2` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of Wix, Webflow; fully adopted on the estate ladderCons behind WordPress (leads)	●	◉	◐	◐	●
Admin save-preview-publish API, validated, backed up Measured: `sed_backup_index` exists in `runtime/_hdl_build/nx_siteedit_daemon.nx`, verified at emit. Live daemon validates the blueprint, previews to a scratch docroot, publishes keeping the previous page rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `RUN-BY:daemon` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	●	●	●
REST plus GraphQL-style projection content API Measured: `cms_api_render_obj` exists in `runtime/_hdl_build/nx_cms_api.nx`, verified at emit. WP REST API is the reference [wp-rest-api]; ours emits injection-safe JSON with caller-chosen field projection [graphql-oct2021] Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of Wix, WebflowCons behind WordPress (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◉	◐	◐	●
Import from WordPress and any foreign HTML Measured: `cms_migrate_wp_import` exists in `runtime/_hdl_build/nx_cms_migrate.nx`, verified at emit. Field and value migration in; nx_site_import strips any foreign site to a portable blueprint rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-GATE-ONLY importers=2` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, DrupalCons not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◐	◐	○	◐
Tamper-evident export-import, sha-verified Source present; authored stronger claim: `cms_migrate_export` in `runtime/_hdl_build/nx_cms_migrate.nx`, verified at emit. Refuses a corrupt or tampered bundle by hash, gate 7 of 7; rival migrations move bytes silently rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-GATE-ONLY importers=2` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Source symbol found; stronger capability claim is authored; ahead of WordPress, Wix, Webflow, DrupalCons not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	◉	◐	◐	◐	◐
Static publish API, sha-pinned promote, prev-backed Measured: `ma_do_promote_content` exists in `runtime/_hdl_build/nx_mgmt_api.nx`, verified at emit. Upload then promote with confirm; concurrent sessions can never cross-publish each other's bytes rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIVE-DAEMON` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	◐	○
On-site search, BM25 Measured: `sr_cat` exists in `runtime/_hdl_build/nx_site_search.nx`, verified at emit. Gated to land the right page on 3 of 3 queries against a real practice corpus rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Webflow, DrupalCons not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◐	●	◐	◐
Scheduled publishing, fire-once cron Measured: `sg_num` exists in `runtime/_hdl_build/nx_cms_schedule_gate.nx`, verified at emit. Register, validate, tick, fire exactly once, missed jobs catch up once rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `GATE:LIVE trial=-` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	●	●	●
Comments Measured: `cmt_score` exists in `runtime/_hdl_build/nx_cms_comments.nx`, verified at emit. WP native comments are the reference rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-WIRED importers=4 nonval=2` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of Webflow; fully adopted on the estate ladderCons behind WordPress (leads)	●	◉	●	○	●
Forms handling Measured: `fg_cat` exists in `runtime/_hdl_build/nx_cms_forms_gate.nx`, verified at emit. Gated form parse and validation; WP leans on plugins rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `GATE:LIVE trial=GREEN` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	◐	●	●	●
Conversion-funnel floor ENFORCED at emit Source present; authored stronger claim: `wb_ux_compliant` in `runtime/_hdl_build/nx_web_builder.nx`, verified at emit. Refuses any blueprint missing CTA, on-site search, trust, or a 3-step path; no rival refuses to build a bad funnel Adoption: `LIB-WIRED importers=24 nonval=17` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
E-commerce cart, tax, totals primitive Measured: `ec_cart_subtotal` exists in `runtime/_hdl_build/nx_cms_ecommerce.nx`, verified at emit. Woo is a plugin; ours is integer money math, gated rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	◐	●	●	◐
Multilingual i18n Measured: `i18n_resolve` exists in `runtime/_hdl_build/nx_cms_i18n.nx`, verified at emit. Drupal core multilingual is the reference [drupal-multilingual] Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow; fully adopted on the estate ladderCons behind Drupal (leads)	●	◐	◐	◐	◉
Block rows validated responsive Measured: `pb_row_valid` exists in `runtime/_hdl_build/nx_cms_pagebuilder.nx`, verified at emit. Gutenberg and the Webflow designer are the references [webflow-dev-docs]; ours validates rows and widths, no visual canvas yet Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of Drupal; fully adopted on the estate ladderCons behind WordPress (leads), Webflow (leads)	●	◉	●	◉	◐
NL command edit with rollback (conversational edit agent over the site) Measured: `cmd_parse` exists in `runtime/_hdl_build/nx_site_command.nx`, verified at emit. Parsed intent: set-hero, add, remove, cta, rollback; ambiguous commands are a safe no-op, no model in the loop, honest and narrow. The September 2026 bar is an agent: Wix Harmony pairs conversational prompting with Aria, its built-in AI agent, beside drag-and-drop editing [wix-harmony-2026-09] and describes it as the evolution of ADI [wix-blog-aibuilder-2026-09]; Framer ships in-editor agents with reasoning and fast mode and 3D transforms this month [framer-updates-2026-09] Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, DrupalCons behind Wix (leads), Webflow (leads); not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◐	◉	◉	○
Per-pattern generation, consumed structure drives layout Measured: `sa_recipe_seq` exists in `runtime/_hdl_build/nx_site_archetype.nx`, verified at emit. 28 live variation pages each follow their source template's own section order, gate 9 of 9 Adoption: `SOURCE-ONLY` — PARTIAL: source exists, never compiled: /api/build it. not adopted: source exists, never compiled: /api/build it Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, DrupalCons not adopted yet: source exists, never compiled: /api/build it	●	○	○	○	○
A-B testing assignment Measured: `ab_assign` exists in `runtime/_hdl_build/nx_cms_abtest.nx`, verified at emit. Deterministic hash assignment, gated rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-WIRED importers=3 nonval=1` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	◐	●	◐	◐
Accessibility enforced on the emitted fleet Measured: `a11_find` exists in `runtime/_hdl_build/nx_cms_a11y.nx`, verified at emit. WCAG-AAA contrast measured on every generated page [wcag22]; rivals reach a11y via tools and plugins Adoption: `LIB-GATE-ONLY importers=1` — PARTIAL: imported only by validation organs (gates, tests, benches): wire it into a shipping program. not adopted: imported only by validation organs (gates, tests, benches): wire it into a shipping program Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, DrupalCons not adopted yet: imported only by validation organs (gates, tests, benches): wire it into a shipping program	●	◐	◐	◐	◐
Roles authz FSM, tenant isolation Measured: `rg_num` exists in `runtime/_hdl_build/nx_cms_roles_gate.nx`, verified at emit. The gate caught a real tenant-isolation bug and refused to record until fixed; WP roles are the reference rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `GATE:LIVE trial=-` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind WordPress (leads)	●	◉	●	●	●
Never-brick deploy, health watchdog, auto-rollback Source present; authored stronger claim: `ma_do_deploy` in `runtime/_hdl_build/nx_mgmt_api.nx`, verified at emit. Promote then detached health probe; a failed deploy restores the previous binary automatically Adoption: `LIVE-DAEMON` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
Multisite from one control plane Measured: `ms_cat` exists in `runtime/_hdl_build/nx_cms_multisite_gate.nx`, verified at emit. Two production domains serve live from one host config today rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `GATE:LIVE trial=GREEN` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	●	●	●
SEO sitemap and robots emitted by construction Measured: `sg_cat` exists in `runtime/_hdl_build/nx_cms_seo_gate.nx`, verified at emit. Every generated site ships sitemap.xml [sitemaps-protocol] and robots.txt [rfc9309]; Yoast-class depth is the WP reference Adoption: `GATE:LIVE trial=GREEN` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind WordPress (leads)	●	◉	●	●	●
Plugin and extension marketplace Measured: `cp_ext_load` exists in `runtime/nx_cms_plugins.nx`, verified at emit. WP 60k plugins [wp-plugins]; the honest structural gap and a deliberate sovereignty tradeoff -- the contract is a sovereign organ-extension registry (an organ declares the hooks it serves; the site builder loads it by name, fail-closed on an undeclared hook), never a foreign-code marketplace Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of Webflow; fully adopted on the estate ladderCons behind WordPress (leads), Drupal (leads)	●	◉	●	◐	◉
N-deep content revisions with restore Measured: `snap_restore_n` exists in `runtime/_hdl_build/nx_cms_restore_lib.nx`, verified at emit. Shipped 2026-09-22: restore to any recorded revision with the previous state recorded first, so every restore is additive and reversible; revisions are NXSNAP1 archives named .v<id> beside the one-step snapshot, listed newest first and ledgered append-only; every refusal is named and writes nothing (gate 38 of 38); WP still owns the restore UI, ours is the lib, its gate and the daemon wiring queued rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of Wix, Webflow; fully adopted on the estate ladderCons behind WordPress (leads)	●	◉	◐	◐	●
Visual block canvas (drag, resize, reorder) Measured: `pb_canvas_emit` exists in `runtime/_hdl_build/nx_cms_canvas_lib.nx`, verified at emit. Rows and widths validate today (pb_row_valid) but there is no visual canvas; Gutenberg and the Webflow designer are the references [webflow-dev-docs] -- the contract is a canvas emitted FROM the validated row model that round-trips to a byte-identical blueprint Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of Drupal; fully adopted on the estate ladderCons behind WordPress (leads), Webflow (leads)	●	◉	●	◉	◐
Showcase presentation tier (image-led hero, display type, full-bleed, dark band) Measured: `ed_showcase_css` exists in `runtime/_hdl_build/nx_nishi_editorial.nx`, verified at emit. The tier showcase line in a .site blueprint emits a full-bleed hero (heroimg key or seeded art), a self-hosted SIL-OFL display face, a sticky nav with skip link, 4:5 image cards, a dark statement band and a multi-column footer, zero JS, from ONE editorial system; the incumbent site re-emits byte-identical; Wix and Webflow templates are the visual bar, nike.com the reference read this month [nike-home-2026-09] Adoption: `LIB-WIRED importers=7 nonval=4` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of Drupal; fully adopted on the estate ladderCons behind Wix (leads), Webflow (leads)	●	●	◉	◉	◐
Per-page design-craft ruler, bytes on disk or live, rival-comparable Source present; authored stronger claim: `ua_kvn` in `runtime/_hdl_build/nx_ui_audit.nx`, verified at emit. nx_ui_audit file <path> grades a release BEFORE golive and a mirrored rival page by the same counts: 15 presence points (self-hosted face, display treatment, measure control, imagery, aspect boxes, sticky nav, fluid grids, full-bleed, keyframes, scroll-driven motion, reduced-motion guard, skip link, one h1, four landmarks, no script or external src) plus raw counts, ADDITIVE beside the 12-point hygiene score; no rival CMS grades its own output against a competitor page [awwwards-sep-2026] Adoption: `RUN-BY:clock` — fully adopted (top of its ladder). dossier Pros Source symbol found; stronger capability claim is authored; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
Rendered-pixel design referee on emitted pages Measured: `pj_referee` exists in `runtime/nx_page_pixjudge_lib.nx`, verified at emit. The bytes ruler cannot see beauty; the contract is a sovereign render of the emitted page judged on pixels (overlap, contrast as painted, hierarchy) that disagrees with at least one planted defect the bytes ruler passes; until it lands every craft claim on this board is UNMEASURED beauty Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Licensed imagery pipeline for site heroes and cards Measured: `sm_admit_image` exists in `runtime/_hdl_build/nx_site_media_lib.nx`, verified at emit. Shipped 2026-09-21: an image row beside the hero or card image (path, licence id, source, origin) is required and the estate provenance ledger must verdict SHIP_OK for the file's own sha256 against the rights table (cc0-1.0, client-owned, estate-generated admitted; an unknown, non-redistributable, conditional or unverified licence, a missing row, a missing file or a malformed row is refused by name); the ledger row is written once per sha and licence and a relicence is a new row; attribution licences wait on a credit renderer before a cc-by row may grant them rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Wix (leads), Webflow (leads)	●	●	◉	◉	●
3D scene section hosted by the sovereign engines (zero third-party script) Measured: `ed_sc_scene` exists in `runtime/_hdl_build/nx_nishi_scene.nx`, verified at emit. The September 2026 Site of the Day class carries WebGL scenes and GSAP-class scroll choreography (WebGL named 16 times and GSAP 21 across 25 winners in the mirrored listing); the contract is a hero or band section hosting a scene from the estate's own WASM and WebGPU engines behind a CSS-only door, emitted from a blueprint key, with the page still at 15 of 15 on the craft ruler [awwwards-sep-2026] Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	●	●	○
Design system derived from evidence, never typed (type scale, spacing, breakpoints, tracking, face, motion, radius, each with n per row) Measured: `ua_derive` exists in `runtime/_hdl_build/nx_ui_audit.nx`, verified at emit. nx_ui_audit extract measures every value a page carries and derive writes ONE token set as the median across mirrored best-in-class pages with n on every row; the showcase tier refuses to emit without it. Wix Harmony, the Webflow AI Site Builder and Squarespace Design Intelligence generate a design system from a brief with a model [wix-harmony-2026-09] [webflow-aisb-2026] [squarespace-design-intelligence], Stitch writes it as a portable DESIGN.md [stitch-designmd-2026]; ours is the one whose every value carries the count of sources that produced it and can be re-derived when the field moves, an exceed candidate until the operator's eye and the pixel referee agree Adoption: `RUN-BY:clock` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Drupal; fully adopted on the estate ladderCons behind Wix (leads), Webflow (leads)	●	◐	◉	◉	○
Section catalog wide enough to choose from: catalog-conditioned generation binding brief data to declared section components Measured: `sa_catalog_pick` exists in `runtime/_hdl_build/nx_site_catalog_lib.nx`, verified at emit. Shipped 2026-09-21: knowledge/site_sections.conf declares every crawled section token per tier with its renderer, required brief keys and token slots, knowledge/site_catalog.conf holds the 28 crawled patterns as pickable entries (ids are the published variation pages, rebuilt byte-identical from the sources) and sa_catalog_pick composes the showcase index from the brief, refusing a missing input or an unrenderable section by name; 9 of 28 patterns are fully renderable by the showcase tier today, the rest wait on form, gallery, video, pricing and faq renderers. The September 2026 production shape is generation that SELECTS from a declared component catalog and binds data to props, and a 4B model already recovers about 98 percent of frontier quality doing so [arxiv-2609-04184-a2ui]; Relume-class libraries run to a thousand sections. The contract: a declared section catalog (each entry with its own measured tokens and done-rule) and a picker that composes a page from the brief, refusing a section whose inputs the brief lacks Adoption: `LIB-WIRED importers=3 nonval=2` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Drupal; fully adopted on the estate ladderCons behind Wix (leads), Webflow (leads)	●	◐	◉	◉	◐
Motion presets: scroll-driven, hover and reveal choreography chosen per section, zero script Measured: `ed_motion_preset` exists in `runtime/_hdl_build/nx_nishi_motion_lib.nx`, verified at emit. Shipped 2026-09-22: six presets declared as data in knowledge/motion_presets.conf (reveal, stagger, parallax, scrub, sticky, view-transition) with one selector per section, emitted by nx_nishi_motion_lib over the derived tokens with a supports test for scroll-driven animation, keyframes and a reduced-motion guard per view preset and a view-transition opt-in for the page; nx_nishi_motion_gate reads 51 of 51 on the NAS with five distinct presets on the fixture page, the craft ruler at 15 of 15 and zero script. Webflow builds interactions on GSAP and saves custom actions as reusable presets for scroll, hover and other triggers [webflow-gsap-presets], and the month's Site of the Day class names GSAP 21 times across 25 winners [awwwards-sep-2026]; the estate reaches the same choreography with zero script Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind Wix (leads), Webflow (leads)	●	●	◉	◉	●
Cross-device render verification (a real browser at phone and desktop widths, the captures the referee reads) Measured: `cd_capture` exists in `runtime/nx_cdp_capture.nx`, verified at emit. The field measures 68 percent of AI-generated pages failing on at least one of nine browser-and-device pairs, 88.3 percent of symptoms whole-layout [arxiv-2608-12518-webcompat]; rivals preview breakpoints in their editors; ours is UNMEASURED on phones today because headless Edge and Chrome clamp a 390 px window to 504 and 512 px. The contract: a sovereign CDP client over the estate websocket client that sets device metrics, captures at declared widths and hands the pixels to the referee; the browser stays an outside oracle, never in the build path Adoption: `REGISTERED-DARK` — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked). dossiernot adopted: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, WebflowCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked)	●	◐	◐	◐	●
Constraint-solved section layout (grid alignment, grouping, whitespace and colour harmony solved over the derived tokens) Measured: `ed_layout_solve` exists in `runtime/_hdl_build/nx_nishi_layout_lib.nx`, verified at emit. Layouts generated and optimised under declared constraints with designer preferences were rated above heuristic and random layouts in two user studies this month [arxiv-2609-03918-asp-gui]; Wix and Webflow auto-layout inside fixed templates; ours places sections by template order. The contract: a solver (the estate already carries nx_part_solver constraint kinds and nx_chartlay_lib collision placement) that chooses grid, grouping and whitespace per section from the derived tokens and the brief, printing its objective beside the page Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	◐	◐	○
Render-grounded refinement loop (emit, render, judge the render, re-emit until the gate passes) Measured: `sb_refine_loop` exists in `runtime/_hdl_build/nx_site_refine_lib.nx`, verified at emit. One edit, one renderer-derived observation, with a submission-level hard layout gate, beats one-shot generation across frontier models [arxiv-2609-00194-redeck]; the field grades edit and repair as tasks [arxiv-2506-06251-designbench] and interaction beyond stills [arxiv-2606-00154-webigbench]. No builder ships an automated judge loop. The contract: the emitter re-runs with the referee's named defects until the pixel referee and the bytes ruler both pass, journaling each iteration on this board Adoption: `LIB-WIRED importers=3 nonval=2` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Pairwise-preference design referee (expert or VLM judged, rating-model ranked, calibrated against human critiques) Measured: `pj_pairwise` exists in `runtime/nx_page_pairwise_lib.nx`, verified at emit. UI-Bench ranks 300 generated sites from over 4,000 expert pairwise judgments with a TrueSkill-derived model [arxiv-2508-20410-uibench]; VLM pairwise contests ranked by Glicko are compared with an artist's own ranking [arxiv-2606-00112-vlm-glicko]; VLMs still show clear gaps against human aesthetic assessment [arxiv-2603-01083-aeseval] and designer critiques are the calibration set [arxiv-2407-08850-uicrit]. The contract: candidates from the refinement loop compared pairwise by the estate referee and by an outside VLM oracle, ranked by a rating model, with the referee's agreement to a human-judged set published beside every grade; nx_ui_judge (registered, no cap ever minted) already scores an Awwwards-axis rubric floor and is the incumbent to extend, never twin Adoption: `LIB-WIRED importers=3 nonval=2` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Design decisions verified in the emitted bytes: every token referenced, every blueprint claim landed (no design theater) Measured: `ua_token_ref` exists in `runtime/_hdl_build/nx_ui_tokenref_lib.nx`, verified at emit. Generative UI tools leave over 25 percent of their stated design rationales unimplemented [arxiv-2607-22928-design-theater]; the estate measured its own compare pages defining layout tokens that nothing referenced. The contract: a ruler tooth that refuses a defined-but-unreferenced token and a blueprint key that lands in no measured CSS property, scored per perspective the way WebCoderBench scores [arxiv-2601-02430-webcoderbench] Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Agent-ready emission: structured data, action cues and temporal validity signals by construction, zero script Measured: `sb_agent_ready` exists in `runtime/_hdl_build/nx_site_agentready_lib.nx`, verified at emit. An agent-ready version of one site took browser agents from 49.3 to 89.3 percent strict success and cut steps from 9.31 to 6.49 [arxiv-2607-12056-agent-ready]; WordPress and Drupal reach structured data through plugins and modules, Wix emits it natively. Ours: an agentready row (a schema.org type and an updated date, data in the blueprint) arms Microdata on the elements the emitter already writes (a WebPage scope, name, description, a publisher of the declared type, a dateModified time), a data-action cue on every call to action and card, with zero script by construction (a script tag would cost the craft point), and nx_ui_audit grades it as an additive 4-point agent block; gate 35 of 35, the plain page byte-untouched Adoption: `LIB-WIRED importers=3 nonval=2` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	◐	◐	●	◐
Brand profile names the EVIDENCE the design derives from (corpus and recipe), persisted in the blueprint Measured: `sb_brand_profile` exists in `runtime/_hdl_build/nx_site_brand_lib.nx`, verified at emit. Squarespace Blueprint derives a site's personality into palette and type from a guided brief [squarespace-design-intelligence]; persisted information architecture keeps generated interfaces aligned where regeneration without it drifts [arxiv-2608-25565-maru]. Ours: a profile row (industry and personality) selects a corpus of mirrored best-in-class pages and a recipe from knowledge/design_corpora.conf, and the tokens are derived by the one ruler forked over that corpus with n on every row (gate 29 of 29: two corpora derive two token sets, seven named refusals, an explicit tokens row wins); the rivals derive a personality from a model, ours from named evidence a second seat can re-derive Adoption: `LIB-WIRED importers=3 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Drupal; fully adopted on the estate ladderCons behind Wix (leads), Webflow (leads)	●	○	◉	◉	○
Evidence corpus refreshed from the month's award listings (winners' pages and stylesheets), tokens re-derived with n per row Measured: `crf_refresh` exists in `runtime/_hdl_build/nx_corpus_refresh.nx`, verified at emit. Today the derive reads 8 hand-mirrored sources; the Sites of the Year listing [awwwards-soty] and the Site of the Day listing [awwwards-sep-2026] name the pages, and a page whose CSS is external contributes only breakpoints until its stylesheets are fetched. The contract: a beat that mirrors the month's winners with their stylesheets through nx_research_fetch, re-derives the tokens and journals every n that moved, so the design system ages with the field instead of with the seat Adoption: `REGISTERED-DARK` — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use). dossiernot adopted: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use) Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, DrupalCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use)	●	○	○	○	○
Layout tokens derived for board pages Measured: `ua_layout_derive` exists in `runtime/nx_ui_layout_lib.nx`, verified at emit. Rail width, content width, column counts and breakpoints measured across board-class pages and derived with n per row; the builders ship fixed grids, the derivation is ours to make [tailwind-responsive-design] Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	●	●	●
Responsive shell with a section rail Measured: `sc_shell_pass` exists in `runtime/nx_compare_garden_lib.nx`, verified at emit. The two-column documentation shell on every board: sticky rail, wide content column, filled grids, scrolling tables, nothing removed [mdn-grid-common-layouts] Adoption: `LIB-WIRED importers=4 nonval=3` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	●	●	●
Records as clean tables Measured: `rct_end` exists in `runtime/nx_compare_record_lib.nx`, verified at emit. One renderer turns every record class into a table with named columns and disclosure for the long ones [nng-progressive-disclosure] Adoption: `LIB-WIRED importers=10 nonval=8` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	●	●	●
Relationship dossiers per rung Measured: `sc_rung_dossier` exists in `runtime/nx_compare_dossier_lib.nx`, verified at emit. A rung's plan row, matrix row, gate, log rows, refs and design rows joined by symbol in one place and linked both ways; no rival board joins its records [ia-arxiv-2026-09] Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Design modes as data Measured: `dmo_mode_load` exists in `runtime/_hdl_build/nx_design_modes_lib.nx`, verified at emit. Composed and organic modes declared once and chosen per surface; Wix and Webflow ship templates, not modes [wiki-japanese-garden] [wiki-wabi-sabi] Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of Wix; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	◐	●	●
Paper behind the glyphs (a text-free capture the pixel referee reads the paper from) Measured: `cd_textfree_shot` exists in `runtime/nx_cdp_capture.nx`, verified at emit. No rival builder grades its own output on pixels; the field's pixel checker estimates the paper from a mixed histogram of the text region and reports that heuristic as a warning only [atf-textcontrastcheck]; the contract is a second capture of every door with the text painted transparent, so the paper behind every glyph is measured instead of estimated and text over a photograph or a gradient is graded like text on solid paper Adoption: `REGISTERED-DARK` — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked). dossiernot adopted: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, DrupalCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked)	●	○	○	○	○
Award rubrics as referee axes Measured: `rub_rubric_score` exists in `runtime/_hdl_build/nx_page_rubric_lib.nx`, verified at emit. The field scores itself through juried competitions: Awwwards on Design 40, Usability 30, Creativity 20, Content 10 by 18+ jury members with the 3 outlying scores dropped [awwwards-evaluation-2026-09], the Webbys on seven criteria from content to overall experience [webby-judging-2026-09], the FWA on 100 points from 20+ judges with 70 the FOTD line [fwa-about-2026-09]; no builder scores its own output on those axes. The contract: knowledge/design_rubrics.conf as data and a referee that prints one score per declared axis beside the pixel grade, mapped to the estate referee that measures it or UNMEASURED by name, a mirrored winner scored the same way Adoption: `LIB-WIRED importers=3 nonval=2` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Journey driver measures the 1-2-3 path in the browser Measured: `cd_journey_run` exists in `runtime/nx_cdp_capture.nx`, verified at emit. Task-level UX is what the research bodies measure: Baymard's 200,000+ hours over checkout, product page, search, homepage and category, mobile and accounts [baymard-research-2026-09], NN/g's ten heuristics [nngroup-heuristics-2026-09], and the flow logs of Built for Mars (90+ case studies, 1,400+ examples) [builtformars-2026-09] and Page Flows [pageflows-2026-09]; no rival ships a synthetic driver that walks its own emitted page's declared path -- their analytics observe live visitors, and Webflow Optimize measures conversions on traffic [webflow-optimize-2026-09] -- so the four read No. The contract: knowledge/journeys.conf declares each site's landing, action and completion steps, the capture driver walks them at both doors and writes a friction receipt per step (clicks, fields, scroll depth, settle time) graded against the research rows, and a planted extra step reads worse Adoption: `REGISTERED-DARK` — PARTIAL: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked). dossiernot adopted: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked) Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, DrupalCons not adopted yet: callable, authorised, completed MCP invocation not established by supplied journal window; lifetime history unknown (direct execution may be recorded elsewhere; this scan does not establish lifetime use); no execution surface runs it either (clock, cron, daemon, roster, actlog and surfaced forks checked)	●	○	○	○	○
Variant archive and selection in the shape of the Darwin and Mendel Godel Machines Measured: `sr_archive_select` exists in `runtime/_hdl_build/nx_site_refine_lib.nx`, verified at emit. The Godel-machine line keeps an open-ended archive of variants and validates empirically (DGM 2505.22954), evolves the evaluators with the agents (Red Queen 2606.26294), admits an edit only when superiority is certified statistically (SGM 2510.10232) and mutates from comparative signals across the archive (Mendel 2608.07645) [arxiv-godel-machine-2026-09]; no builder keeps an archive of what it emitted with the scores it earned. The contract: every variant archived content-addressed with its referee scores and journey receipt, the parent chosen by score and novelty, a child admitted only when it beats its parent on every deterministic referee at a declared confidence, nothing deleted Adoption: `LIB-WIRED importers=3 nonval=2` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Multivariate optimisation of the design knobs Measured: `sb_knob_search` exists in `runtime/_hdl_build/nx_site_refine_lib.nx`, verified at emit. UI optimisation on arXiv is thin (cooperative design optimisation 2508.16077, Bayesian-optimised UI 2501.11829, an opportunistic bandit for UI experimentation 2006.11873) [arxiv-ui-optimization-2026-09] while online experimentation moved to calibrated abstention for A/B prediction and Bayesian A/B arithmetic [arxiv-abtesting-web-2026-09]; WordPress reaches A/B through plugins (57 in its directory) [wp-plugins-abtesting-2026-09], Wix's builder page names no experimentation feature [wix-harmony-2026-09], Webflow Optimize ships A/B, multi-variable testing and AI optimisation on live traffic [webflow-optimize-2026-09], and Drupal has a contributed A/B and multivariate module [drupal-abjs-2026-09]; none searches its own emitter's design knobs against a deterministic referee. The contract: the knobs are the derived-token bands and the catalogue's discrete choices, a factorial search scored offline by the referee panel and the journey receipt, then nx_cms_abtest assigning live traffic to the top variants with each step's completion logged, and a knob outside its evidence band refused Adoption: `LIB-WIRED importers=3 nonval=2` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	◐	○	◐	◐
Guarded self-improvement of the improver Measured: `sb_policy_pin` exists in `runtime/_hdl_build/nx_site_refine_lib.nx`, verified at emit. September 2026: evidence-gated harnesses (RegenHarness 2609.27612), regularised harness evolution (RRSI 2609.24972), hash-pinned policy with evolution confined to a code-owned catalogue and bounded knobs (GuardrailLoop 2609.12216), the judge demoted to advisor behind a deterministic verification layer (2609.02246), harness tampering and poisoned self-evaluation named as the failure classes (2609.00069, 2609.17817) [arxiv-selfimproving-2026-09]. The contract: the loop's proposal policy is a versioned knob set evolved under a hash-pinned policy it cannot edit, the referees live in another lane, a mutation that touches a referee or the pin is refused by name, and every accepted policy change carries its held-out gain Adoption: `LIB-WIRED importers=3 nonval=2` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Blind side-by-side pair judging surface (one pair at a time, Left and Right by a keyed coin, the click mapped back to the referee's A and B) Measured: `pjj_swapped` exists in `runtime/nx_pairjudge_lib.nx`, verified at emit. Design Arena presents two randomly paired models anonymously side by side and counts a choice as one vote [designarena-home-2026-09]; LMArena's Code Arena battles the same way [lmarena-codearena-2026-09]; no CMS judges its own output that way. Ours: the loopback surface nx_pairjudge_serve shows one open pair with no candidate name, which candidate sits left is the parity of the pair's nibble of the session token, and a Left or Right click is mapped back to the referee's A or B before its one writer runs (gate 51 of 51 with known-answer tokens); an exceed candidate until the operator and the nishi team judge with it Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Dwell per verdict as gap evidence (a meta row: side shown, milliseconds on the pair, door) Measured: `pjj_meta_row` exists in `runtime/nx_pairjudge_lib.nx`, verified at emit. Pairwise labels are an attention-limited measurement and response time carries gap information the labels do not [arxiv-arena-preference-2026-09]; the arenas record the vote alone. Ours writes one meta row beside every pair row (one writer, one reader) with the dwell from the pair's last render and prints the judge's median dwell on the surface; a pair posted without a render reads UNSHOWN by name Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Next pair by information: fewest judgments then the smallest rating gap Measured: `pjj_next_pair` exists in `runtime/nx_pairjudge_lib.nx`, verified at emit. K-Sort Eval picks the next comparison by expected information, balancing uncertainty and diversity [arxiv-2602-09411-ksort-eval]; Design Arena draws four models at random per session [designarena-home-2026-09]. Ours serves the open pair whose candidates carry the fewest judgments and, among those, the smallest gap in the referee's Elo over every judged row (a known-answer set where the first open pair is the wrong choice is a gate tooth) Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Standings on the judging surface: rating, W-L-T, win share with its Wilson interval Measured: `pjj_standings` exists in `runtime/nx_pairjudge_lib.nx`, verified at emit. Design Arena ranks by Bradley-Terry from 1200 with a standard-error margin, win rate and placement per model [designarena-home-2026-09], on a leaderboard page that is itself client-rendered [designarena-leaderboard-2026-09]; leaderboards move under sub-1 percent perturbation, so an interval is part of the number [arxiv-arena-preference-2026-09]. Ours: the referee's one rating ruler (nx_dr_elo, start 1500, K 32) over every judge's rows, wins, losses, ties, the win share of decided pairs with the Wilson 95 interval from the gate base class, UNMEASURED with no decided pair, behind a disclosure until the judge is done Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of WordPress, Wix, Webflow, Drupal; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Inter-judge agreement as Krippendorff alpha beside the pairwise agreement Open — watching `runtime/nx_pairwise_agreement_lib.nx` : `pj_kripp_alpha`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. Professional designers disagree on UI pairs at Krippendorff alpha 0.25 [arxiv-2511-20513-designpref] [arxiv-pairwise-ui-2026-09] and arena leaderboards are validated by interrater agreement [arxiv-arena-preference-2026-09]; the contract: alpha over pairs judged by two or more judges printed in permil on the referee's receipt beside the k-of-n agreement, UNMEASURED by name with one judge dossierwatching pj_kripp_alpha Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Live preview beside the stills on the judging surface Open — watching `runtime/nx_pairjudge_live_lib.nx` : `pjj_live_route`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. The arenas judge running applications, not stills: BigCodeArena executes the code and WebCraftBench explores the app under coverage guidance [arxiv-2609-15387-webcraftbench]; the contract: GET /live/<k>/ serves the candidate's own page and assets under its dir on the loopback surface, a path outside the dir refused by name dossierwatching pjj_live_route Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Human-anchored machine co-judge with a gap-calibrated vote Open — watching `runtime/nx_pairwise_anchor_lib.nx` : `pj_cojudge_anchor`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. ANCHOR infers a new model's human-reference score from LLM-judge comparisons by learning each judge's sensitivity and bias from historical human and judge pairs [arxiv-2609-19599-anchor]; WebCraftBench's agreement with human preference rises with the pair's score gap [arxiv-2609-15387-webcraftbench]. The contract: the referee's agreement curve over the score gap on its receipt, a machine vote cast only above the gap where the curve clears the floor, ABSTAIN by name below it dossierwatching pj_cojudge_anchor Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Per-axis verdicts: aesthetics, usability, requirement alignment and overall Open — watching `runtime/nx_pairwise_axes_lib.nx` : `pj_axis_verdicts`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. WebCraftBench scores visual aesthetics, usability and requirement alignment as separate dimensions with no model leading on all three [arxiv-2609-15387-webcraftbench]; the contract: an axis field on the pair row, one Elo table per axis, the widest human-machine disagreement named per axis dossierwatching pj_axis_verdicts Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Leaderboard robustness: leave-one-out and flip perturbation with rank intervals Open — watching `runtime/nx_pairwise_robust_lib.nx` : `pj_loo_stability`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. Bradley-Terry leaderboards are non-robust: sub-1 percent targeted drops, adds and flips change the top model and the confidence intervals [arxiv-arena-preference-2026-09]; the contract: the top candidate's survival under every single-row drop and flip and a rank interval per candidate on the receipt, FRAGILE and STABLE by name dossierwatching pj_loo_stability Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Per-judge preference profiles Open — watching `runtime/nx_pairwise_profile_lib.nx` : `pj_judge_profile`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. Personalised judges beat aggregated ones when designers disagree (DesignPref, GenUI personalisation) [arxiv-2511-20513-designpref] [arxiv-pairwise-ui-2026-09]; the contract: each judge's agreement with the pool and with the machine on the receipt, a planted contrarian judge reading lowest by name dossierwatching pj_judge_profile Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Tournament session: four candidates from one brief, brackets to a final ranking Open — watching `runtime/nx_pairjudge_tournament_lib.nx` : `pjj_tournament`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. Design Arena's session draws four models plus a backup, runs winner and loser brackets with a tie breaker to a final ranking, each choice one vote [designarena-home-2026-09]; the contract: a four-candidate session on the surface producing a full ranking in five verdicts, each a pair row, the ranking equal to the referee's Elo over those rows dossierwatching pjj_tournament Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○

Rival-claim provenance. Every Yes, Best or Part mark in a rival column is a claim about someone else's product. A cited row carries a reference mark that resolves to a pinned mirror (the refs gate measures that); an uncited row is an observation read and is badged in its evidence until a reference lands. The badge is the receipt, never the proof: a mark proves a mirror exists, not that the mirror supports the code. The rows are counted below.

Rival claims, counted
Class	Count	Note
rows with rival marks	41	rows carrying a Yes, Best or Part in a rival column
cited	24	a reference mark resolves to a pinned mirror
uncited	17	an observation read, badged until a reference lands
rows, summed	41	RECONCILED: the parts sum to the total 41

Delivery and evidence

Inspect risks, technical debt, rendered observations, experiments and references. Read scope and limitations alongside every result.

Risk register

Risk	Likelihood x impact	Mitigation
A canvas that emits a different blueprint than the one it read silently rewrites sites	possible x high	CM2 done-rule is byte-identity on a no-op edit before any editing feature is admitted.
Extensions reintroduce the foreign-code surface WordPress suffers from	possible x high	CM3 admits organs only -- sovereign source, declared hooks, no foreign runtime.

Release history and work log

The worked plan, on the board. Every leg appends what it measured, landed, retracted, learned and left queued, per rung, so a crash leaves the next seat a ledger here rather than a transcript to mine. Kinds: measure, land, retract, lesson, queue. Newest last.

When	Rung	Kind	Entry
2026-09-21	CM4	land	gospelofwest.com production v3 = tier showcase; nx_site_build promoted 81f8b82b, nx_nishi_editorial 0d5ffe7c; andelinwest.site byte-identical 7 of 7; page_verify GREEN; fonts served from the release with LICENSE
2026-09-21	CM5	measure	nx_ui_audit live gospelofwest.com hygiene 12 of 12; file mode craft 15 of 15 on the showcase index vs 12 of 15 on the kit index (font_face 1 vs 0, uppercase 26 vs 1, letter_spacing 34 vs 3, aspect_ratio 1 vs 0, full_bleed 2 vs 0, token_defs 73 vs 28); nike.com mirror pending the NAS promote
2026-09-21	CM6	queue	beauty is UNMEASURED by any estate referee today; nx_browser is a window organ and the seat may not pop a browser; the pixel referee is the contract
2026-09-21	CM5	measure	Awwwards September 2026 listing read from the mirror: 25 Site of the Day entries, WebGL named 16 times, GSAP 21 times; the ranker refused LADDER-PARTIAL NO-FRONTIER until this frontier target was declared
2026-09-21	CM4	land	gospelofwest.com production v4 (footer headings h3, a11y heading-skip closed); nx_site_build promoted 69326cf0, _offc installed; apex and www both serve 27,156 B; www lifecycle row relabelled v4
2026-09-21	CM5	measure	nx_ui_audit promoted edb606b0 with file mode and the craft block; same ruler on three pages: gospelofwest v4 index craft 15 of 15 (img 0, svg 5, script 0, external src 0, font_face 1, reduced_motion_guard 2) -- nike.com home craft 6 of 15 (img 31, svg 115, alt 31, loading 31, aspect_ratio 40, script 53, external src 21, font_face 0 inline, skip_link 0, landmark_main 0) -- awwwards listing craft 9 of 15 (img 63, svg 104, script 11, external src 39); the honest gap the counts name is photography (CM7), the exceed the counts name is zero script and zero external assets
2026-09-21	CM6	lesson	A hygiene ruler that reads 12 of 12 on both the kit page and the showcase page discriminates nothing; the craft axis and the rival's raw counts beside ours are what separate them, and beauty still waits on the pixel referee
2026-09-21	CM5	measure	EVIDENCE LOOP LIVE on the NAS: nx_ui_audit derive over four mirrors (nike, awwwards listing, apple, stripe) wrote gospelofwest.tokens -- body 18 px (n=2), consecutive-size ratio 1.066 (n=1), display 162 px (n=1, the listing's largest fixed size), spacing base 12 px and section 48 px (n=2, step 1.319), line-height 1.50 body / 1.20 display, breakpoints 640 and 1068 px (n=4), content width 1255 px (n=2), tracking -0.15 em (n=1), motion 800 ms, radius 24 px; apple and stripe contributed only breakpoints because their CSS is external (extract reads bytes on disk), so their stylesheets are the next sources
2026-09-21	CM4	lesson	Our own pages derive a FLAT scale (ratio 1.053, display 21 px): the ruler measured the kid work before any eye did; the emitter now refuses the showcase tier without a tokens file and every size on the page is a slot the derive filled, so a hand value has nowhere to live
2026-09-21	CM6	queue	A tokens file with n=1 rows is thin evidence and says so per row; widen the corpus with the winners' stylesheets and Site of the Day pages before the pixel referee grades the render
2026-09-21	CM5	measure	tokens re-derived from 8 sources after the measure_px band was split from text columns (900..1600 px page containers): body 14 px n=6, ratio 1.214 n=4, t6 36.9 px, display 162 px n=2, spacing 12 px base n=4 and 48 px sections n=2 (step 1.319), line-height 1.40 body n=5 and 1.25 display n=2, breakpoints 640 and 1068 px n=8, page 1200 px n=3, tracking -0.15 em n=4, motion 800 ms n=4, radius 8 px n=6; v5 re-emitted from these tokens into releases/v5 and previewed at https://gospelofwest.com/v5 through a symlink under the live v4 release (no golive, no browser popup)
2026-09-21	CM4	lesson	A band that blends two kinds (text columns and page containers) derives a 608 px page from column widths; the ruler names the source values per row, so the band was corrected as DATA and the derive re-run -- the fix lives in the conf, never in a hand value
2026-09-21	CM5	retract	the v5 preview at gospelofwest.com/v5 is RETRACTED and its symlink under the live v4 release removed (404 now): the operator's eye read it as squished and a mess, and a public path on the production host is not the alpha/beta/production methodology -- a candidate reviews on the alpha plane (nishifamily.com/alpha/SITE/), stabilises on /beta/SITE/ with a browser check, and reaches production only on the operator's eye
2026-09-21	CM5	measure	headless Edge captures of v5 and v4 at 1440 px (operator-permitted, no window): every v5 heading letter-crushed -- the trk token was the single MOST NEGATIVE letter-spacing any source set (-0.15 em, n=4) applied to every heading down to the 17 px wordmark and 21 px card titles
2026-09-21	CM4	lesson	An extreme is not evidence: the derive now takes each source's most-used negative letter-spacing and the median across sources (-0.03 em, n=4), and the emitter applies it to display sizes only (h1, section h2) because tracking is an optical-size property in every measured source; and a median on an even count is the mean of the two middles -- the first derive took the upper one and read a 162 px display from sources reading 56 and 162
2026-09-21	CM5	measure	ruler extended with two evidence rows: fam (display face class by majority of the sources naming a generic family: sans, n=3) and dch (median h1 length inside the h1_chars band 8..60, n=2); the emitter picks the self-hosted face by class, ships it with its OFL LICENSE into RELEASE/fonts, and fits the display size per page as td x dch / len so a longer headline holds the sources' line count
2026-09-21	CM5	retract	the phone-overflow reading of the first 390 px captures is RETRACTED: headless Edge and Chrome clamp a 390 px window to 504 and 512 px (innerWidth probe) and crop the shot, so the overflow was the capture's, and a 390 px iframe control is refused by the edge (frame-ancestors); phone rendering stays UNMEASURED until a capture driver sets device metrics
2026-09-21	CM6	queue	capture driver: a sovereign CDP client over the estate's websocket client that sets device metrics and captures a real browser at phone and desktop widths -- the oracle stays outside the build path and the pixel referee reads its output
2026-09-21	CM5	land	v6 emitted from gospelofwest_v6.tokens (8 mirrors, true median; nx_ui_audit promoted 62741fa8, nx_site_build 2ff7ba05): body 14 px n=6, ratio 1.178 n=4, display 109 px n=2 fitted to 93 px for the 35-char headline (dch 30 n=2), spacing 10 px base n=4 and 44 px sections n=2, line-height 1.40 n=5 and 1.22 n=2, breakpoints 640 and 1046 px n=8, page 1200 px n=3, tracking -0.02 em n=4 on display only, face sans n=3 self-hosted with its LICENSE; craft 14 of 15 (img 0), hygiene 12 of 12, page_verify GREEN a11y 0; served for review at https://nishifamily.com/alpha/gospelofwest (the release bytes copied into the alpha dir, sha 954bf52e identical to releases/v6/index.html; the ad strip below the footer is the nishifamily edge's own, not the release's); headless desktop capture read by the seat: no crushed glyphs, a two-line sans headline, coherent index -- an interpretation, not a grade; phone UNMEASURED; LIVE stays v4 until the operator's eye, then /beta/gospelofwest with the worklog on this board, then golive
2026-09-22	CM5	measure	research leg 2026-09-21: two arXiv listings (webdesign 98 results, aesthetics 440), 8 abs pages and 8 vendor and roundup pages mirrored through nx_research_fetch and read from the bytes; 20 refs pinned by sha256; Relume app, the Awwwards annual page and the Framer press release refused as mirrors (JS shells, status 0); envato and figma trend pages mirrored but not yet read, so not cited
2026-09-22	CM6	lesson	the estate already carries nx_ui_judge, an Awwwards-axis rubric judge over the ship manifest with an 800 permil competency floor, registered with no cap ever minted: a dark incumbent the referee rung extends rather than twins
2026-09-22	CM9	queue	the ladder to SOTA declared as data: M6 builder parity from evidence (CM9 to CM15) then M7 the research frontier as product (CM16 to CM18), every rung with its role toward a dated target; the ranker prices it after this beat
2026-09-22	CM9	lesson	a future contract pointed at an EXISTING organ reads DARK-CONTRACT on this board (the 09-06 mechanism cannot tell a capability nobody built from one built under another name), so every open contract now points at an absent lib file (nx_site_brand_lib, nx_ui_corpus_lib, nx_site_catalog_lib, nx_nishi_motion_lib, nx_nishi_layout_lib, nx_site_refine_lib, nx_ui_tokenref_lib, nx_site_agentready_lib, and the 08-19 rows nx_cms_restore_lib, nx_cms_canvas_lib); the refs gate's provenance check refused 10 rows whose mirror was fetched under a different url than the row cited (a paper row may not cite a listing as its mirror; an arXiv feed re-fetched under the row url keeps its byte count and changes its sha because the feed embeds a per-request id), all 10 repaired and the gate reads cms clean except the acquitted Framer benign hop
2026-09-22	CM9	land	SHIPPED, referee grade: gate GREEN 29 of 29 on the NAS, awaiting operator acceptance. nx_site_brand_lib (sb_brand_profile): a profile row (industry and personality) selects the corpus and recipe from knowledge/design_corpora.conf (award-product 8 mirrors, builder-marketing 5 mirrors, both recipe showcase) and forks the ONE ruler nx_ui_audit derive --corpus over it through nx_tool_run, writing profile.corpus and profile.tokens beside the pages; an explicit tokens row wins and says so. Promoted on the NAS: nx_ui_audit 523aab7e 642,528 B (derive --corpus, contentdiff GREEN lost 0), nx_site_build efd1bb76 226,488 B (contentdiff lost 1 = the reworded no-tokens refusal, named), nx_site_brand_gate a2c7cf83 88,360 B registered, rostered and declared in cms.gates; _offc installed for both organs; sources pushed through the content door with CAS
2026-09-22	CM9	measure	done-rule met: two personalities on two planted corpora derive body 1600 vs 1400 and display 9600 vs 16000 centi-px, the same corpus derives the same sizes twice, seven refusals named (unknown or empty personality, empty industry, absent mirror, unimplemented recipe, missing recipe row, unreadable conf); the incumbent path is untouched: andelinwest.site emitted by the promoted binary and by the banked previous binary hash identically on all 7 files (index 13,289 B d36748f2, sitemap 573 B, robots 68 B, four practice pages 12,105 B each), and the showcase tokens
2026-09-22	CM9	lesson	a corpus whose pages carry no content width derives a tokens file the emitter refuses (mpx is a required row): the first smoke refused on it, so the gate now asserts the derived file carries every emitter-required row, a derive that succeeds into a refusal being the vacuous half of the contract; and the live sites/andelinwest docroot is not this emitter's output (7,164 B index against 13,289 B emitted, hand-kept contact, legal, schedule and llms.txt), so the honest control was LIVE versus the banked .prev binary on the same blueprint, never the live folder
2026-09-22	CM10	queue	the corpora conf is the substrate CM10 refreshes: a beat that mirrors the month's Site of the Day and Sites of the Year pages with their stylesheets through nx_research_fetch, appends corpus rows, re-derives and journals every n that moved
2026-09-22	CM9	measure	REAL CORPUS ON THE NAS: gospelofwest_v7.site (the v6 blueprint with the profile row faith-community, award-product in place of its tokens row) emitted through /api/organ_run derives profile.tokens byte-identical to gospelofwest_v6.tokens (sha e289f005, 2,331 B, 8 sources, t1 1400 n=6, td 10900 n=2, mpx 120000 n=3) and an index.html byte-identical to the shipped releases/v6/index.html (sha 954bf52e, 24,192 B): the profile path reproduces the hand-listed derive exactly, from data
2026-09-22	CM9	queue	OWED: the scratch registration nx_site_build_prev (elf nx_site_build.elf.prev, registered for the A/B) has no unregister route (/api/tools/unregister answers 404), so it must be reaped before the next nx_site_build promote rotates .prev under it; a tools/unregister verb is an engineshift gap
2026-09-22	CM15	land	SHIPPED, referee grade: gate GREEN 35 of 35 on the NAS against the promoted ruler, awaiting operator acceptance. nx_site_agentready_lib (sb_agent_ready): an agentready row (a schema.org type and an updated date, data in the blueprint) arms the editorial emitters to write Microdata (a WebPage scope, itemprop name and description on the hero, a publisher of the declared type on the wordmark, a dateModified time in the colophon) and a data-action cue on every call to action and card (ed_sc_act_open), zero script by construction; nx_ui_audit grades an additive 4-point agent block (itemscope, itemtype schema.org, itemprop, datetime, action cues, key agent_points). Promoted: nx_ui_audit 3f489573 645,016 B (contentdiff GREEN lost 0), nx_site_build 2608b3a0 234,680 B (contentdiff lost 3, the three template fragments restructured into ed_sc_act_open, bytes proven identical), nx_site_agentready_gate 4bbec1a5 128,104 B registered, rostered and declared in cms.gates; _offc installed for both organs
2026-09-22	CM15	measure	done-rule met: the armed fixture page carries the scope, five itemprops, the dated time and six cues (13,461 B) while the plain page (13,034 B) carries none and is byte-identical before arming and after disarming; the ruler reads agent 4 of 4 on the armed page and 0 of 4 on the plain page and the zero-script craft point survives; the real gospelofwest blueprint with the row emits itemscope 2, itemprop 5, data-action 10, datetime 1, script 0 and grades agent 4 of 4, craft 14 of 15 (imagery, the CM7 gap); the tokens-only path re-emitted the smoke byte-identical 7 of 7 on the laptop
2026-09-22	CM15	lesson	JSON-LD is a script element to every byte ruler including ours, so the honest agent-ready emission is Microdata on the elements the page already writes (a script tag would have cost the zero-script exceed for a second syntax of the same facts); and the first NAS gate run read 33 of 35 because the LIVE ruler was still the pre-agent build: two ruler teeth failing on the promoted subject before its ruler was promoted is the gate reading the right subject, not a flake, and it read 35 of 35 the moment the ruler promote landed
2026-09-22	CM11	land	SHIPPED, referee grade: gate GREEN 64 of 64 on the NAS (65 of 65 on the laptop, where the template sources live), awaiting operator acceptance. The section catalog as DATA: knowledge/site_sections.conf declares each of the 12 crawled section tokens per tier (renderer or shell or none, required brief keys, token slots, done-rule); knowledge/site_catalog.conf holds the 28 crawled patterns as pickable entries, generated by nx_site_catalog build from the 30 template sources through the nx_tpl_ingest extractors, keys content-addressed as the corpus and the published v- pages, rows sorted, no timestamp, so a rebuild is byte-identical; nx_site_catalog_lib sa_catalog_pick composes a page from the brief by id, archetype or auto and refuses by name (a missing input names the section and the key, no renderer names the section, unknown id, malformed row, undeclared token), the last refusal kept as state a gate reads; nx_site_build reads the pattern and catalog rows and the showcase index follows the picked seq (hero leads even when a pattern lacks it, sub-pages keep their shape), byte-identical without a row. Promoted: nx_site_catalog 64ebc295 78,128 B registered and in _offc, nx_site_catalog_gate a8f29abb 103,144 B registered, rostered, in _offc and declared in cms.gates, nx_site_build c36b9fe4 257,744 B (contentdiff GREEN lost 0 gained 90) and in _offc
2026-09-22	CM11	measure	done-rule met: the live catalog carries 28 pattern rows from 30 sources (dups 2, skipped 0, partition sums) and on the NAS all 28 ids are the published sites/nishifamily/factory variation pages (28 of 28); a brief with a missing input is refused by name (section proof missing step; section ft missing footer; section price missing plan in a tier that renders pricing); the emitter proves order in bytes: the default index puts the services section at 21,034 before the band at 21,521, the proof-first pattern puts the band at 21,035 before the cards at 21,279; the real gospelofwest brief with pattern auto picks v-ecommerce-5595255323 (hd>nav>hero>cta>ft, 2 sections) after 28 considered, because 19 of 28 patterns carry a form, gallery, video, pricing or faq section the showcase tier cannot render yet and the covered renderable set ties at two sections; 9 of 28 patterns are fully renderable by the showcase tier today
2026-09-22	CM11	lesson	the corpus seg-store the 28 patterns were extracted into (knowledge/index/tplcorpus) survives on neither tree, but the 30 template sources on the laptop and the 28 published pages on the NAS do, and the keys are content-addressed, so the catalog was rebuilt from the sources and cross-checked against the published names instead of the store (a derived store can be lost while its inputs and its receipts remain); and a gate that resolves an organ from a root ladder must put the local build dir before the _offc twin on the laptop, where _offc is a fossil store: a July nx_site_build.elf there read the kit index and failed two teeth until the ladder was ordered serving root, then _build, then _offc
2026-09-22	CM11	queue	auto composition on a real brief is thin by construction until the showcase tier renders form, gallery, video, pricing and faq sections (19 of 28 patterns need one); the tie-break among covered patterns is catalog order, and an industry-to-archetype row on the brand profile is the next data row so a faith-community brief prefers a landing pattern over an ecommerce one; a tools unregister verb is still owed for the scratch nx_site_build_prev row, which now points at the CM15 binary after this promote rotated .prev
2026-09-22	CM7	land	SHIPPED, referee grade: gate GREEN 46 of 46 on the NAS and the laptop, awaiting operator acceptance. nx_site_media_lib (sm_admit_image): a hero or card image the blueprint names is admitted only with an image row beside it (image, path, licence id, source url, origin) and a SHIP_OK verdict from the estate's ONE provenance ledger (nx_asset_prov_lib over knowledge/provenance/assets.jrnl and the rights table knowledge/model_license.conf), the file hashed in the release, the ledger row written once per sha and licence (a relicence is a new row, never an edit), every refusal naming the path and the reason (no row, unreadable file, malformed row, unknown licence, non-redistributable licence, conditional or unverified licence which the ledger calls REVIEW and a person settles); nx_site_build admits the hero and every card image before emission and takes a medialedger row as the ledger door; the rights table gained two estate classes as data, client-owned and estate-generated, beside the existing cc0-1.0, each verified because the class is the estate's own design decision. Promoted: nx_site_media_gate fc6c235f 106,776 B registered, rostered, in _offc and declared in cms.gates; nx_site_build 16bedaaf 304,152 B (contentdiff GREEN lost 0 gained 113 over five added sources) and in _offc
2026-09-22	CM7	lesson	the incumbent ledger writes the claim row before the verdict judges it (the nx_asset_prov order), so a refused licence still leaves one journal row per sha and licence and the last row wins: a cc0 re-admit after three refused claims recovers the image, and the gate asserts that recovery rather than a clean journal; and a CC-BY image stays unadmitted by design until a credit renderer exists, because the rights table marks attribution CONDITIONAL and a conditional right is a person's decision, never the emitter's
2026-09-22	CM7	queue	the credit renderer for attribution licences (a rendered source and licence line in the colophon and a cc-by-4.0 row added only after its legal code is mirrored and read) is the next imagery data row; the imagery ladder the site program declared stands (public domain, then cc0, then the estate's own render, then attribution last), and a real photograph on gospelofwest waits on the client's own media under client-owned
2026-09-22	CM7	measure	done-rule met: an unlicensed path is refused by name (the forked promoted emitter on a showcase blueprint with heroimg media/hero.bin and no image row exits 1 printing MEDIA REFUSED image=media/hero.bin no licence row) and a licensed one renders in the hero (the same blueprint with image rows exits 0, the receipt admits the hero and the card, the index carries figure class heroimg with src media/hero.bin at offset 21,010 and the scratch ledger carries the cc0 row); the admission gate reads 46 of 46 on the laptop and on the NAS; the byte-identity control: gospelofwest_v7.site emitted by the promoted emitter and by the banked previous one (nx_site_build_prev, the CM11 build) gives index.html sha 954bf52e 24,192 B on both, the same bytes as the shipped v6 release, so a blueprint without images is untouched by this rung
2026-09-22	CM12	land	SHIPPED, referee grade: gate GREEN 51 of 51 on the NAS and the laptop, awaiting operator acceptance. knowledge/motion_presets.conf declares six presets as data (reveal, stagger, parallax, scrub, sticky, view-transition) and one selector per section; nx_nishi_motion_lib (ed_motion_preset) emits each preset over the derived tokens from the conf alone: a view preset is wrapped in a supports test for animation-timeline view() with its keyframes and a reduced-motion guard, a layout preset pins with no guard, the page preset opts the document into view transitions with its duration on the motion token; five refusals name the preset, the section, the unreadable conf, the malformed row and a template without a selector; nx_site_build reads motion rows (motion, section, preset) from the blueprint and emits one style block after the showcase CSS on the index and every sub-page, and a blueprint without motion rows emits no preset. Promoted: nx_nishi_motion_gate 079b27af 77,560 B registered, rostered, in _offc and declared in cms.gates; nx_site_build ffc54653 317,960 B (contentdiff GREEN lost 0 gained 37) and in _offc; nx_site_catalog_gate a8a6b1d9 64 of 64 and nx_site_media_gate afc81b8d 46 of 46 rebuilt on the shared fixture lib nx_site_fixture_lib and re-promoted
2026-09-22	CM12	measure	done-rule met: the fixture showcase page with five motion rows carries five distinct presets on the index and on the sub-page (index_distinct_presets 5 of live_presets_declared 6), the promoted craft ruler reads 15 of 15 with script_tags 0, and the reduced-motion guard covers every preset and the built-in motion (index_reduced_motion_guards 7); the plain control without motion rows carries no mv- keyframes and no view-transition rule; an unknown preset makes the emitter exit 1 naming the preset; the gate reads 51 of 51 on the laptop and on the NAS
2026-09-22	CM12	lesson	a feature block behind a switch makes the switch-off run a neg-control of the whole gate, not only of the new teeth: the first fixture read 14 of 15 because the imagery point wants an img and the fixture had none, so the motion fixture now carries a licensed hero through the CM7 door, and the three site gates share one fixture lib because a fixture spelled in three gates drifts three ways
2026-09-22	CM13	queue	next rung by rank: the capture driver, a sovereign CDP client over nx_ws_client that sets device metrics on an outside headless browser (the oracle stays outside the build path) and hands 390 px and 1440 px pixels to the referee; its grading half waits on CM6, and CM6 waits on grid layout in nx_browser_render (the sovereign renderer lays out box, flex and float over real TTF and has no grid, so it cannot lay the showcase page out yet), declared here as a browser-board dependency rather than solved as a cms rung; the ranker's near-tie between CM6, CM13 and CM12 was a missing signal, not an order
2026-09-22	CM12	measure	byte-identity control: gospelofwest_v7.site emitted by the promoted emitter (ffc54653) and by the banked previous one (nx_site_build_prev, the CM7 build) into /tmp/gw_cm12_live and /tmp/gw_cm12_prev gives index.html sha 954bf52e 24,192 B on both, the same bytes as the shipped v6 release, so a blueprint without motion rows is untouched by this rung; /compare/cms republished 290,851 B (regen GREEN, forkfresh 4 of 4), page_verify GREEN a11y 0; the scratch nx_site_build_prev registration now points at the CM7 binary after this promote rotated .prev
2026-09-22	CM1	land	Shipped 2026-09-22: nx_cms_restore_lib (snap_record, snap_list, snap_has, snap_restore_n over the NXSNAP1 archive lib, the site editor's own .v<id> revision naming and the store's atomic writer; refusals SR_RC_* by name) and nx_cms_restore_gate 38 of 38 GREEN on the NAS, promoted 154e0f1a, registered, rostered, in cms.gates and _offc; sources buildroot/runtime/_hdl_build on both trees; matrix symbol snap_restore_n flipped from its contract
2026-09-22	CM1	measure	nx_cms_restore_gate on the NAS 2026-09-22: first_id 1790056177 and second_id 1790056178 (two records inside one second, the bump path), pre-restore id 1790056179, first archive 76 bytes for three files, ledger rows 5 after two records and three restores, ids listed 7 with two planted malformed archives; restore-then-restore byte-identical on a text file, a binary file with NUL CR LF and a high byte, and a markup file; a file of 8388609 bytes refused TOO-LARGE-FOR-PACKER with nothing recorded
2026-09-22	CM1	lesson	two incumbents already held the halves of this rung (the NXSNAP1 archive one step deep, the site editor's N-deep .v<epoch> naming for its blueprint): composing them cost one lib and no second parser or naming scheme; the base packer reads each file into an 8 MiB reserve and would truncate a larger one in silence, so the new door measures every file with lseek END first and refuses by name rather than inheriting the silence; a refused restore is proven to write nothing by comparing the bytes after every refusal, not by reading the refusal code alone
2026-09-22	CM1	queue	owed on this rung: wire snap_record and snap_restore_n into nx_cms_admin (the daemon still exposes the one-step snapshot and restore only; a revisions browse page in the shape of the site editor's history page is the restore UI WordPress has) and retire the base packer's silent truncation at its root by sizing the per-file read from the file; both are daemon deploys through the health-checked door, so they wait for a deploy window rather than a promote
2026-09-22	CM12	measure	2026-09-22 CM12 and CM15 are on the gospelofwest alpha (v8 blueprint: motion page view-transition, feat stagger, proof parallax, cta reveal; agentready Organization): the index carries 3 mv- keyframe blocks and itemscope 2, page_verify GREEN a11y 0, published through the owner's plane plan-site-release-gospelofwest-; CM11 pattern auto deliberately not applied (it picks a five-section pattern that drops feat and proof until the industry to archetype map lands); CM7 waits on a licensed family photo
2026-09-22	CM12	lesson	a capability shipped on the board is not on the product until the release lane republishes it, and a lane a seat runs by hand has no owner and no recipe: the lane is now the publisher's plane with a RACI, rerunnable by any seat or agent through nx_plan_run site-release-gospelofwest or nx_orchestrate_raci site-release run
2026-09-22	CM20	queue	2026-09-22 the ux role is Accountable for design-architecture (RACI row, crew row ux design-architecture nx_ui_audit, binding design-architecture to nx_plan_run) and runs plan-design-architecture-cms- (measure the emitted board with the design ruler, derive board.tokens from six board-class mirrors, verify the page); the seat tutors by landing the design rows GD1 to GD6 with their evidence and by building the shell as the first landing under that ownership; nothing is deleted to structure the page
2026-09-22	CM20	lesson	a page can carry every fact and still read as a dump: structure is a layout and a join, not a cut; the three complaints (one endless column, records as prose, relationships scattered) are one architecture defect with three faces, and the remedy is measured widths, one table renderer and one join by symbol
2026-09-22	CM19	measure	The six mirrored documentation pages carried NOT ONE breakpoint, track or sticky rule between them (nx_ui_audit derive REFUSED: type ratio n=0) because every one links its CSS externally; the layout evidence is in the stylesheets, so fourteen were mirrored and pinned beside the pages: MDN [mdn-css-page-layout] [mdn-css-reference-layout] [mdn-css-left-sidebar] [mdn-css-reference-toc] [mdn-css-global], GitHub Docs [ghdocs-css-1] [ghdocs-css-2] [ghdocs-css-3] [ghdocs-css-4] [ghdocs-css-5], Tailwind [tailwind-css-1] [tailwind-css-2], NN/g [nng-css-1] [nng-css-2]. The corpus is data: buildroot/knowledge/compare/board_layout.corpus (20 sources).
2026-09-22	CM19	lesson	A page whose CSS is external contributes only what its bytes carry -- a derive over pages alone reads the field as carrying no layout at all, which is a false NOTHING, not a measurement. Mirror the stylesheets a page links and pass them as sources; the derive prints every source's value so the corpus can be checked row by row.
2026-09-22	CM19	queue	nx_ui_audit has no gate of its own: nx_organ_ship refused NO-GATE-FOUND, and the ship was re-run with gate=nx_ui_layout_gate (the in-process proof of the extracted measurement base class nx_ui_measure_lib and of ua_layout_derive) plus a behaviour probe of the untouched file verb on the live cms page. OWED: nx_ui_audit_gate -- the file, extract, derive and layout verbs end to end over fixture pages, either by forking the freshly built binary (gk_run_capture) or, the OO direction, by moving the verb bodies into nx_ui_audit_lib so an in-process gate can drive them; declare the pair in knowledge/organ_gate.conf when it lands.
2026-09-22	CM19	land	nx_ui_measure_lib.nx (the audit's measurement base class, extracted verbatim: value parser, bands + ua_bands_use seam, distinct-value table, um_median_field, um_bp_scan legacy+range syntax, the ONE token-row writer) + nx_ui_layout_lib.nx (ul_track_scan, ul_sticky_scan, ul_toc_count, ul_measure, ua_layout_derive) + nx_ui_audit layout verb over ua_verb_sources (derive and layout share ONE source reader); nx_ui_audit shipped sha 943dccb7 with gate=nx_ui_layout_gate 37/37 (sha f108708f) and a behaviour probe of the file verb IDENTICAL 2184 B; bands rail_px 160..360 px and sticky_top 0..120 px in knowledge/design_evidence_ranges.conf; the design-architecture plane (ux-owned) ran 4 of 4: audit file, layout --corpus, derive --corpus, page_verify GREEN.
2026-09-22	CM19	measure	board.tokens DERIVED over 20 sources (6 pages + 14 stylesheets): bp2 1012 px n=5, rail 306 px n=6, stickytop 0 px n=5, mpx 1280 px n=5, bp1 768 px n=10, toc present in 4 of 20 sources (the pages carry the marker, the stylesheets cannot). board_type.tokens now derives too (t1 16 px n=7, ratio 1.16 n=5, s1 10 px n=12, s6 60 px n=3, mot 150 ms n=9, rad 6 px n=9, fam sans n=5) where the pages alone REFUSED at ratio n=0.
2026-09-22	CM20	measure	nx_organ_ship nx_swcompare_matrix: the content ruler read lost_from_live=2 (SS-COMMIT REFUSED: segid= and (seq1730). NOTHING -- seg-store lib messages from a sibling's shared-lib change, SOURCE-ADDED nx_scorecard_receipt_lib.nx + nx_store_seed_checked.nx, not the shell) and GAINED 55 (every shell string among them); approved by the exact evidence record allow_loss_v1=38dc0c2b6833582c1c7b1b45069b6a31a54cd32b09c5f89f3c12edb899b22980 and re-run. The behaviour probe on cms html is EXPECTED to differ by exactly the wrappers and the shell CSS: that difference is the rung.
2026-09-22	CM20	land	garden shell LIVE: nx_swcompare_matrix 04c61be8 (1121032 B) on the serving root, _offc and buildroot/_offc (all three hashed); nx_compare_garden_lib + nx_compare_garden_gate 32/32 (9a5cbe18); /compare/cms republished 333965 B via nx_compare_regen_scoped --domain cms (the EC67 per-call organ, no shared scope file); page carries data-shell='two-column' x1 and data-bp2='1012' x1; page_verify GREEN a11y-issues=0; GD1 adopted
2026-09-22	CM20	measure	ship loop: rebuild is byte-stable (jobs 1790095253 and 1790095572 both built 04c61be8; job 1790094509 built ba342a09 before a sibling shared-lib change landed); behaviour probe live-vs-staged cms html DIFFERS +761 B at offset 9053 = the shell; review record knowledge/status/harness/review-nx_swcompare_matrix-cm20-shell.txt sha 6eafcc80 over observation 56363174 (bound to ba342a09, superseded by the rebuild); the promote that landed 04c61be8 is journaled in organ_ship.jrnl at 1790095687 SHIPPED GREEN by a concurrent organ_ship run, not by this seat's reviewed run (which read artifact-hash-unavailable once the stage was consumed)
2026-09-22	CM20	lesson	an exact-artifact review binding cannot land while a sibling ships the same target: each run rebuilds the candidate, the review binds to one candidate, and a concurrent promote consumes the stage; the loss token is deterministic across byte-stable rebuilds so approvals converge only in a quiet window -- coordinate a shared generator through its RACI lane, never by racing; a cap's second field is its expiry (the lane cap expired at 1790092533 and read as scope denials all afternoon); regen.scope is the EC67 race in the wild (re-written to lang under this seat), so the scoped regen is called with --domain on argv from now on
2026-09-22	CM20	queue	owed: nx_ui_audit_gate declared in organ_gate.conf; CM21 sc_record_table (records as tables); CM22 sc_rung_dossier (join related rows by rung id); CM23 dm_mode_load (garden vs wabi-sabi as data); operator ACCEPT row for the shell in the accept- plane is a separate signature from this land
2026-09-22	CM20	measure	closure ran as the ux-owned plane nx_plan_run design-architecture-cms extended to 8 rows (5 nx_eco_graph_build, 6 nx_importcensus, 7 nx_compare_regen_scoped --domain cms, 8 page_verify): 8/8 COMPLETED in 41 s, graph 21707 nodes 61605 edges, census asof 1790096972, /compare/cms republished 340919 B with exactly one nx-shell wrapper (tag-anchored count 1; the other 3 mentions are board text), page_verify GREEN a11y 0; nx_compare_rank cms reads rungs=24 done=10 open=14 with CM20 out of the FINISH list; RACI design-architecture ux=A engineer=R migrated 243 rows 8/8 GREEN; debt 1790096923 filed for the TSV->store migrate that cannot express a removed row
2026-09-22	CM21	measure	DARK-CONTRACT cleared before the build: cms.matrix rows 54 and 55 repointed from runtime/nx_compare_garden_lib.nx (an existing organ, so the ruler read the missing symbol as DARK) to the absent lib files the rungs will land in, runtime/nx_compare_record_lib.nx (sc_record_table) and runtime/nx_compare_dossier_lib.nx (sc_rung_dossier); CM23 already pointed at absent runtime/nx_design_modes_lib.nx; cms republished 336568 B via nx_compare_regen_scoped --domain cms; nx_capsearch (record table render prose disclosure columns) names no incumbent renderer for record prose (nx_record_tsv_gate is a TSV bridge, nx_design_ledger_lib renders one ledger), so sc_record_table is novel and composes the ledger's table shape; baseline structure on cms: 23 tables, 12 meth prose blocks, 7 sections, 0 pre
2026-09-22	CM21	land	nx_compare_record_lib (rct_begin/rct_row/rct_row_word/rct_row_sum/rct_end: a capture renderer in the page's own plan-table shape with the row count DERIVED and published as data-rows, disclosure above the conf bar as a details summary; rct_census_list: the record-as-prose ruler that NAMES every offending block) + nx_compare_record_gate 60/60 promoted ab445398 registered offc-twin + nx_record_prose CLI (live d8cc7d64 count-only; the listing candidate c4d6b806 is staged behind the loop's zero-exit probe until the cms page reads CLEAN); the four bars are threshold rows in knowledge/design_evidence_ranges.conf (record_kv=3 record_pipes=2 record_named=5 disclose_rows=12, each with its calibration); adoption source-landed at all five prose-record sites: ladder+bars and treatment verdicts (nx_swcompare_lib plan_pass), scorecard tally, design ledger verdict (nx_design_ledger_lib), rival-claim provenance (nx_swcompare_matrix), every partition re-derived by that ruler's own sum and read RECONCILED or LEAK on the page
2026-09-22	CM21	measure	baseline census on the LIVE cms page before the matrix ships (nx_record_prose, thresholds SET): blocks=17 records_prose=5 kv=9 pipes=42 named=39 partition=3 tables=23
2026-09-22	CM21	lesson	a row count handed to a renderer beside the rows is a second copy of the rows' shape and the two drift, so the renderer captures the rows and derives the count; and a count without a worklist is not actionable, so the census prints RECORD-BLOCK id= kv= pipes= named= partition= text= for every offender before its verdict line (the first cut of this very instrument printed a count and nothing else)
2026-09-22	CM21	lesson	the ladder's target counters are not a partition (best in class and frontier count by code, undated, bad class and duplicate by state) so the table prints no sum row for targets; rungs, roles, bars, treatment rows and design rows each have a sum function in their own ruler (ld_partition_sum, ld_role_sum, bf_partition_sum, trt_partition_sum, dl_partition_sum) and the page re-derives every one beside its total
2026-09-22	CM21	queue	closes when nx_swcompare_matrix and nx_swcompare_sota are promoted, buildroot/_offc refreshed, cms regen'd and nx_record_prose reads records_prose=0 on every published board; the sota generator ship was refused NO-GATE-FOUND (declare its gate in organ_gate.conf first); owed: the regen prints records_prose= per domain as a receipt line; nine dated candidate copies of the emitter in buildroot/runtime still carry the old prose and are lifecycle EC71 work, not this rung
2026-09-22	CM21	measure	done-rule on the LIVE cms page after both generator ships: nx_record_prose sites/nishifamily/compare/cms/index.html reads blocks=17 records_prose=0 kv=0 pipes=0 named=7 partition=0 tables=28 thresholds=SET verdict=CLEAN, against the 1790100651 baseline of records_prose=5 kv=9 pipes=42 named=39 partition=3 tables=23 on the same page; the seven named counts left are sentences under the record_named bar, not records
2026-09-22	CM21	land	both generators SHIPPED through nx_organ_ship with the field discovery receipt as a table (fi_summary_table in nx_swcompare_lib copies the field
2026-09-22	CM21	land	THE CENSUS RIDES THE PUBLISHED LINE (the queue row's owed regen receipt): nx_compare_regen 9aa402ca and its scoped twin nx_compare_regen_scoped c88a8c42 both SHIPPED through nx_organ_ship (nx_forkfresh_gate 45/45, content GREEN one-way gain, behaviour reviews review-nx_compare_regen-cm21-records-receipt.txt 026feb7a and review-nx_compare_regen_scoped-cm21-records-receipt.txt eea93aca over the scoped cms publish as the representative probe): rg_records_receipt censuses the exact bytes each published .html page with rct_census, the same ruler nx_record_prose and nx_compare_record_gate run, prints records_prose= on that page's published line and records_pages= records_prose_total= records_uncal= before verdict= (additive, never fails); witnessed on the promoted scoped twin: published cms/index.html (349553 bytes) records_prose=0 and records_pages=1 records_prose_total=0 records_uncal=0 verdict=GREEN, forks 4/4 FRESH, agreeing with the CLI's independent reading; owed: a fixture-tree tooth for the receipt in a regen gate (today the proof is the probe observation plus the CLI agreeing), and the fleet total from one full regen run
2026-09-22	CM21	lesson	the contract was DARK while the capability shipped: matrix row 54 and this rung named sc_record_table, a symbol nx_compare_record_lib never declared (its entry is the capture, and rct_end is the function that emits the table), so the cell read DARK-CONTRACT (organ present, symbol absent) after four promotes; repointed BOTH the row (_ABSENT_:rct_end) and the rung symbol (rct_end) to the real emitting entry instead of minting a name-only alias, because the flip must reward the entry point that delivers the capability, never the name; the cell flips on the next regen of cms
2026-09-22	CM21	measure	fleet witness from the full nishi_compare_regen run 1790106496 with the receipt in the path: published=255 fails=0 forks 4/4 FRESH records_pages=129 records_prose_total=10 records_uncal=0 verdict=GREEN; 120 of 129 published pages read records_prose=0; the residue is 9 pages carrying 10 records in TWO classes the census names by block: the sota page's Measured bindings verdict paragraph (named=6, one record each on llm foodscience water livingdocs cleanserve lang smallos researchreach) and sharehub's two head-to-head measured blocks (Measured on both sides kv=5; Where to beat them named=7); so the done-rule holds on cms and on 120 boards and NOT yet on every board, and the worklist is those two emitters in the sota generator, the same rct_begin/rct_row/rct_row_sum/rct_end conversion the five matrix sites got
2026-09-22	CM21	lesson	the emitter of the Measured bindings block is not the organ its name suggests: the sota-class boards (llm foodscience water livingdocs cleanserve lang smallos researchreach) are rendered by nx_swcompare_matrix, which imports nx_swcompare_sota_lib, while nx_swcompare_sota never imports that lib and carries its own copy of the paragraph in main (spelled consistent= stale_under=, dormant: no published page carries that shape); converting the lib alone left the sota candidate byte-equal on that block (the staged elf kept the old fragment and none of the new caption, which is how the mistake was caught before approval); both copies now render the rct table with the partition sum, and the lib-vs-program duplicate is EC71 lifecycle work for the sota lane, named here rather than merged blind
2026-09-23	CM21	land	residue conversion SHIPPED through nx_organ_ship, both trees: nx_swcompare_sota 8b3d77e0 (loss set e3d83173 = 16 retired sentence fragments, review 02ab29c0) and nx_swcompare_matrix ac683187 (loss set 6011ff50 = 21 retired fragments, review 667a21a4; the cms probe's same-size diff adjudicated as clock-derived age fields since the field receipt is all-digit and fi_kv_rows is output-identical on it); the sota lib and the sota program both render the Measured bindings partition as a table with the sum beside bound, the shared lib renders Where to beat them as ppp-gap rows and the probe stamp as ppp-stamp rows through fi_kv_rows (the one key=value row emitter fi_summary_table composes); buildroot/_offc forks refreshed and hash-identical (sota 8b3d77e0, matrix ac683187), the scoped regen read forkfresh=4; witnesses of both residue classes: sharehub 313412 B records_prose=0 (CLI CLEAN named=5 tables=18, was 2) and foodscience 186456 B records_prose=0 (was 1); fleet witness owed from the full run
2026-09-23	CM21	measure	DONE-RULE HOLDS ON EVERY PUBLISHED BOARD: full nishi_compare_regen run 1790126244 with the shipped generators (sota 8b3d77e0, matrix ac683187) and all four buildroot/_offc forks FRESH: published=255 fails=0 records_pages=129 records_prose_total=0 records_uncal=0 verdict=GREEN; every one of the 129 published .html receipt lines reads records_prose=0 (count of ' records_prose=' lines 129, count of ' records_prose=0' lines 129, difference 0); the nine residue pages of run 1790106496 (llm foodscience water livingdocs cleanserve lang smallos researchreach sharehub, ten records) are all zero, and the publisher receipt and the nx_record_prose CLI agree on the two witness classes (sharehub, foodscience); the residue was the sota page's Measured bindings paragraph and sharehub's two head-to-head blocks, now nx_compare_record_lib tables
2026-09-23	CM21	land	THE OWED PUBLISHER REFEREE EXISTS: nx_compare_regen_gate (buildroot/runtime, both trees) drives the PROMOTED nx_compare_regen and nx_compare_regen_scoped on a /tmp fixture tree in --domain mode and proves the records receipt against an independent rct_census over the bytes each run published: planted meth-block record reads records_prose=1 == census, the clean page 0 == 0 (gv_bite both ways), the promoted nx_swcompare_matrix's real sota page (48010 B) reads 0 == 0 with forkfresh=1, the scoped twin reads 1 == 1, api.json lines carry no records field, summary records_pages=1 records_uncal=0; the gate doubles as the fixture generator through a symlink at the fixture's _offc/ (verdict 0, json and html above the publisher's floors, export 2) so no second organ was built; admission refusal (exit 4) reads SKIP. 22/22 GREEN on the NAS (job 1790129414), SHIPPED 6788e559 through nx_organ_ship (PROVE self 22/22, first promote, offc twin, registered), declared in knowledge/organ_gate.conf for nx_compare_regen and nx_compare_regen_scoped (the nx_forkfresh_lib row keeps its own proof) and on cms.gates. First cut read 19/22: the generator refused a sota fixture with no bound rows and my echo filter hid its words -- the fixture now mirrors nx_sota_drift_gate 1:1 and any nonzero publisher exit echoes its whole stdout
2026-09-23	CM21	lesson	A LINE FILTER OVER A PUBLISHER'S STDOUT HID THE GENERATOR'S REFUSAL AND SHOWED ITS DOWNSTREAM SYMPTOM: leg C first read ATOMIC-FAIL code=-107 syscall=-2 on the scope receipt, one hop after the real event (the promoted generator refused a sota fixture with no bound rows in its verdict mode, and GATE RED plus the generator's own words were not among the echoed needles). The gate now prints the publisher's whole stdout on any nonzero exit; a gate that reports a failure without the diagnostic the subject already captured is unactionable
2026-09-23	CM21	queue	OWED, off this seat's tool surface: the subject-level mutation bite of the publisher through the new gate -- nx_gate_bite nx_compare_regen_gate buildroot/runtime/nx_compare_regen.nx 2 subject nx_compare_regen aim rg_records_receipt (the aim keeps the mutants inside the receipt function; every mutant build deploys the publisher to the serving root, so run it in a quiet window with compare_regen.progress reading domain=DONE and the harness restore verified by hash). The fixture-level bite (planted 1 / clean 0) is proven inside the gate; this is the second, subject-level proof
2026-09-23	CM1	land	FINISH ROW CLOSED -- snap_restore_n IS WIRED INTO THE SHIPPING PROGRAM: nx_cms_admin (buildroot/runtime/_hdl_build, both trees) imports nx_cms_restore_lib; ca_snap_names is the ONE list of the four site files the one-step archive and every revision pack; POST /admin/snapshot records revision <rid> beside snapshot.nxsnap (the 303 body names it, a refused revision is named and never fails the snapshot); GET /admin/revisions lists every recorded revision newest first with a guarded restore form each; POST /admin/restore/rev rev=<id> (session + CSRF, tested before the prefix-matching W4 route) restores through snap_restore_n, answers 303 with the prerestore id or a 409 naming the SR_RC refusal, and re-renders only after a successful unpack; the editor's Backup fieldset links the revisions page. PROOF ON THE WIRE: nx_cms_snapshot_gate extended with K7-K12 (revision file exists, listed, restore by id byte-exact, the undo of the undo, unknown id 409 by name with the live page untouched, unauthenticated 403), builds the daemon --build-only (a gate never twin-refreshes the serving root), D001-migrated onto nx_gate_verdict (every row a gv_check tooth, refusal rows neg-control-, legacy CMSGATE line restated from the base counters in the cms_gate.log dialect, gv_kv revision_id/prerestore_id, gv_verdict last): 13/13 GREEN on the NAS, SHIPPED 52fbc9f0 (loss d31dd1cc + review 24a30bae), organ_gate.conf nx_cms_admin -> nx_cms_snapshot_gate (the convention gate is a laptop dev fixture), cms.gates row. DAEMON SHIPPED through nx_organ_ship: PROVE by the declared gate 13/13, content GREEN, promoted 6dd75952 (prev 516115 B banked in the .prev slot), offc twin, registered. RELEASE STAGE: the promoted binary is READY; restarting any live nx_cms_admin instance onto it is the release owner's step (EC68), not this seat's
2026-09-23	CM1	queue	the ranker's FINISH row (state=LIB-GATE-ONLY importers=1) lags the wiring: nx_compare_rank reads the importers census, which nx_importcensus derives from knowledge/store/ecograph_full, and the graph is rebuilt by nx_eco_graph_build on its daily beat (off this seat's tool surface); the new import edge nx_cms_admin -> nx_cms_restore_lib lands in the census on the next beat followed by nx_importcensus, and the row flips by itself -- no seat edit is owed, only the beat
2026-09-23	CM5	land	nx_ui_audit given its execution surface: UI-MAP rows now end with ts=<epoch> (both trees; SHIPPED dcb15fac via nx_organ_ship, prove GREEN by the organ_gate.conf declaration nx_ui_audit->nx_ui_layout_gate, contentdiff GREEN, behaviour IDENTICAL over file sites/gospelofwestcom/current/index.html); clock row uiauditshowcase 86400 declared over that release page (desired plane, folds live at the scheduler's next window) so knowledge/status/ui_debt_map.log becomes a time spine; the ranker's REGISTERED-DARK label clears once the live clock plane carries the row
2026-09-23	CM2	land	the visual block canvas as a library and its referee: runtime/_hdl_build/nx_cms_canvas_lib.nx (pb_canvas_emit + pbc_*: the validated 12-grid row model on nx_editstack_lib, gestures RESIZE/SWAP/MOVE conserve pb_row_valid by construction, canonical layout text round-trips byte-identical, page fragment carries {{key}} placeholders for crd_render, no-script gesture canvas, layout.txt + layout.ops persistence with head cursor, commit flattens) and nx_cms_canvas_gate 52/52 GREEN on the NAS, shipped ed6a8082 through nx_organ_ship (registered, cms.gates + tool_schemas rows); composed the incumbents the spend gate named (nx_editor_canvas's stack adoption, nx_cms_pagebuilder's ruler, nx_cms_store's atomic writer) rather than a client-side JS canvas the estate rejects; daemon wiring (GET /admin/canvas, POST /admin/canvas/op, {{layout}} splice before the one renderer, layout files in every snapshot) in flight on nx_cms_admin with K13-K18 wire teeth on nx_cms_snapshot_gate
2026-09-23	CM2	lesson	the wire gate read 19/19 on its own ship and 17/19 one minute later on the daemon's ship: its /tmp fixture site persists between runs and the first run's commit had left layout.txt and layout.ops there, so the two teeth that begin from a site with NO layout met one (a gate that inherits its predecessor's fixture measures the predecessor); fixed at setup by moving both files aside exactly as the stale draft and snapshot already are -- the estate's idempotent-fixture law, paid for again; the magic ratchet also refused the first cut of the six teeth (typed reserves, the csrf length, row ids, a file mode) and every bound was derived instead: the capture reserve named once and read by the buffer line too, form bodies sized from their parts, the marker length read from the marker, MODE_0644 from the syscall layer, row ids from the base class counter
2026-09-23	CM2	land	WIRED INTO THE SHIPPING CMS: nx_cms_admin SHIPPED 70ea37d7 (GET /admin/canvas = the no-script editor over the committed layout with its gesture log replayed, POST /admin/canvas/op = init / resize / swap / move / undo / redo / commit with every refusal the lib's own name and 403 without session or csrf, ca_render_publish splices the COMMITTED layout's fragment at the template's {{layout}} marker before the one renderer so the columns' {{key}} placeholders fill in the same pass, layout.txt + layout.ops ride every snapshot and revision); nx_cms_snapshot_gate SHIPPED a4954f17 as the daemon's declared referee, 19/19 GREEN over real HTTP (K13-K18: init form, init 303, canvas with bar and zero script, lone-column resize refused 409 by name, 403 without session, commit reaches the live page through crd_render); every bound derived, fixture idempotent; the live admin instance restart onto 70ea37d7 is the release owner's step (EC68), reported READY; the ranker's FINISH row for CM2 will read LIB-GATE-ONLY until the eco-graph beat and nx_importcensus see the nx_cms_admin import edge, as CM1's did
2026-09-23	CM2	queue	the subject-level mutation bite of the canvas referee is owed and off this seat's tool window (nx_gate_bite registered, not surfaced): `nx_gate_bite nx_cms_canvas_gate buildroot/runtime/_hdl_build/nx_cms_canvas_lib.nx 2 aim pbc_apply` (in-process gate, no subject keyword needed) -- until it runs, the 52/52 is a gate that has fired on correct data plus its own planted-width bite, not a mutation-proven one; also owed: the live nx_cms_admin restart onto 70ea37d7 by the release owner (EC68), and the FINISH label flip after the eco-graph beat and nx_importcensus see the nx_cms_canvas_lib import edge
2026-09-23	CM3	land	nx_cms_plugins (runtime/nx_cms_plugins.nx 17862 B sha 11d90d05, both trees byte-identical) + knowledge/cms_extensions.conf (timeout_ms row with its calibration; fail-closed with no ext rows yet) + nx_cms_plugins_gate 35/35 GREEN shipped a4beecb0 (promoted, _offc twin, registered): in-process teeth over a /tmp fixture whose extension binary is the gate itself (header csv, declared vs undeclared, registry allowlist, timeout row, fork only after both say yes, script and nonzero-exit refusals after the fork, splice inside main / before body / append / byte-for-byte with none), then nx_site_build built --build-only and forked on fixture blueprints (declared extension accepted and spliced inside its main landmark; undeclared hook and unregistered organ refused by name with no index written); nx_site_build wired (extroot and extension rows, cp_ext_load before any byte is written, cp_splice_write on all three tiers) and SHIPPED 922e5289 through nx_organ_ship: prove=GREEN by nx_cms_plugins_gate (organ_gate.conf declaration, closure bank CURRENT), adopt content=GREEN behave=IDENTICAL (WEAK: a usage-line probe; the behavioural evidence is the gate's end-to-end teeth), promoted + _offc twin, pre-change live ffc54653 317960 B banked by hash; contract symbol REPOINTED cp_plugin_load -> cp_ext_load on this rung and the matrix watch row (the lib's vocabulary is ext throughout, cp_ext_new/load/emit; a second name for the same loader would be a stub) -- the watch flips on the next cms regen; cms.gates + tool_schemas rows landed
2026-09-23	CM3	lesson	the row-counter tooth passed n=51 for a 57-byte fixture on the gate's first ship (34/35): a hand-counted length beside a literal is a second copy of the literal's shape and drifts silently (rule 11's own law, paid for again inside a gate) -- the fixture is now one const and the tooth derives its length with pg_slen, so the count cannot disagree with the bytes; and nx_fs_write append refuses a row without its trailing newline (four rows bounced in one batch, file unchanged each time) -- end every row
2026-09-23	CM3	queue	(1) the gate re-ship with the derived fixture length is job 1790143327 -- read prove= on its receipt; (2) nx_gate_bite nx_cms_plugins_gate buildroot/runtime/nx_cms_plugins.nx 2 subject nx_site_build is off this seat's surface, run it from a seat that has it; (3) the first REAL extension organ + its ext row in knowledge/cms_extensions.conf (the registry ships fail-closed and empty: no site can load an extension until an owner registers one, and the owner column names who); (4) a live site blueprint carrying extension rows goes through the release owner (EC68), never by hand; (5) FINISH labels clear on the eco-graph beat + nx_importcensus (LIB-UNCENSUSED is a lag, not a gap)
2026-09-23	CM6	queue	dependency edge DECLARED on the rung (deps CM5,CM13): the referee reads the capture driver's real-browser pixels (the oracle outside the build path) -- the sovereign renderer nx_browser_render is a 99 KB lib with one mention of grid, so a sovereign render of the showcase tier would grade a picture no browser paints; the ranker's DEP-PULL now prices CM13 ahead of CM6 instead of the seat overriding the order
2026-09-23	CM13	measure	laptop gate nx_cdp_capture_gate 32/32 GREEN (in-process over nx_cdp_lib, then the subject end to end on the fixture page: phone innerWidth=390 ihdr=780x1688 png=16281 B, desktop innerWidth=1440 ihdr=1440x900; the absent-browser neg-control refuses rc=3 with no receipt); live https://gospelofwest.com/ on the final binary: phone door innerWidth=390 ihdr=780x1688 811532 B, desktop innerWidth=1440 ihdr=1440x900 906323 B, devtools answered in 1023 ms, settled in 353 ms, verdict=OK -- the 2026-09-22 retraction (headless clamps a 390 px WINDOW to 504) is closed because Emulation.setDeviceMetricsOverride sets the VIEWPORT; the phone render is captured and UNMEASURED until the CM6 pixel referee grades it
2026-09-23	CM13	land	nx_cdp_capture (the wire half: bounded loopback connect via nx_connect, the INCUMBENT upgrade nx_websocket_client_upgrade, frames via nx_websocket_frame, launch and reap of the host browser, one navigation per capture, per door Emulation.setDeviceMetricsOverride + double-rAF innerWidth + Page.captureScreenshot, receipt rows) + nx_cdp_lib (conf rows, every DevTools reply through the estate's JSON DOM nx_value_parse_json, headers via nx_http_header_find, image dims via nx_img_dims, RFC 6455 length arithmetic, the metrics rule, the receipt row) + nx_cdp_capture_gate + knowledge/cdp_capture.conf (the oracle candidates, the doors and every budget with its calibration, as data); laptop build sha b2a06615 (112334 B), gate 9e448a5e (96471 B); pushed to the NAS through the content door: lib cbdd2008 20413 B, driver 52e59b7b 35561 B, gate dc632324 22979 B, conf 6309cb84 2617 B at the root knowledge/; organ_gate.conf declares nx_cdp_capture -> nx_cdp_capture_gate; the DOM's string-loop bound derived from the source length on both hosts (nx_value_parse_json sha de5b0c97 16655 B)
2026-09-23	CM13	lesson	A FIXTURE THAT HAND-PLACES TWO BYTES OF A THREE-BYTE VALUE TESTS THE FIXTURE, NOT THE LIB: the 64-bit frame-length tooth read RED because 70000 (0x011170) was written into two bytes and the parser honestly read 4464; every length byte is now derived from the value and a tooth asserts the fixtures genuinely need the 16- and 64-bit forms. AN INCUMBENT'S OWN CAP BIT THE PAYLOAD CLASS: nx_value_parse_json's string loop stopped after 1,048,576 iterations and returned the truncated span with a clean verdict, while a phone screenshot reply carries 1,123,404 base64 bytes -- the bound is now the source length (a string cannot outrun its source), nx_json_complete_gate 10/10 on the new source, and this gate carries a 1,123,404-byte reply as a tooth so the cap cannot return in silence (nx_value_parse_json_test cannot run here: its GBIF fixture file is absent on the laptop, exit 20 before and after the change). Incumbents were found only by asking nx_spendgate PER HELPER (json, image dims, header find); the organ-level query named none. NishiLang out-params are mmap'd cells handed by pointer; the laptop build lane roots at nxc2/runtime
2026-09-23	CM13	queue	NAS ship through nx_organ_ship (its PROVE reads SKIP on the NAS by name: no browser row of the conf resolves there, so the laptop is the proving host and its 32/32 is the evidence); the cms.gates row for the gate; CM6 pj_referee over the capture receipts (the phone render of gospelofwest is UNMEASURED until the pixel referee grades it); a NAS-side browser row when a headless oracle exists there; nx_cdp_eval (graphics GR53, the incumbent CDP witness this driver composed its upgrade lane from) is laptop-only and ABSENT on the NAS -- a flow-back row for its owner; the gate's bite run (nx_gate_bite is off this seat's surface)
2026-09-23	CM13	lesson	A GATE WHOSE END-TO-END WITNESS IS STRUCTURALLY ABSENT ON THE PUBLISHING HOST MUST ANNOUNCE THAT LEG, NEVER gv_need IT: the first NAS ship refused exit 21 because the gate's verdict read SKIP there (no browser row of the conf resolves on the NAS) and the ship loop reads any nonzero gate exit as a refusal; the witness is now optional and announced (E2E-NOT-RUN-HERE reason=no-browser-row-of-the-conf-resolves-on-this-host, e2e_ran=0 among the values), the in-process teeth carry the NAS verdict and the laptop run carries the end-to-end leg (33/33 GREEN, e2e_ran=1) -- the 2026-09-02 law applied; gate source now sha 5daef5be 23696 B on both hosts
2026-09-23	CM13	land	NAS SHIPPED via nx_organ_ship: nx_cdp_capture sha 798f2e25 (build src_path buildroot/runtime/nx_cdp_capture.nx), PROVE GREEN 26/26 on the NAS through the organ_gate.conf declaration (the end-to-end leg announced E2E-NOT-RUN-HERE, e2e_ran=0; the laptop run carries it at 33/33), ADOPT-GATE no-live-elf = first promote (announced), harness token h0423172feb601dfe rows=11, PROMOTE live_sha == staged_sha, OFFC-TWIN installed from the promoted binary, REGISTERED callable (schema rows appended to knowledge/tool_schemas.conf for the organ and the gate; nx_toolreg_reconcile owed for tools/list discovery), DOC DOCUMENTED hits=3
2026-09-23	CM13	measure	closing measures: nx_cdp_capture_gate SHIPPED sha 4b5aad68 (prove GREEN self, promoted, offc twin, registered); nx_toolreg_reconcile registered=2 (both rows on tools/list discovery, the 47 no-schema residue is not this lane's); scoped regen cms published index.html 374713 B (prev 369396) api.json 90811 B forkfresh=4 verdict=GREEN; nx_page_verify https://nishifamily.com/compare/cms VERDICT=GREEN status=200 a11y-issues=0; nx_compare_rank cms rungs=24 done=14 open=10 DARK-CONTRACT=0, CM13 done via cd_capture with FINISH REGISTERED-UNAUTHORISED at rank time -- cap minted the same minute (allow nx_cdp_capture,nx_cdp_capture_gate exp 1792778435 nonce 1790186500, laptop file ~/.nishi/cdp0923.cap), so the label clears on the next rank; cms.gates row appended (total 11612 B)
2026-09-23	CM6	measure	referee on the live gospelofwest.com index captured by the motion-settled driver: phone door, the eyebrow p (11 px, white over the hero photograph) paints 3.00:1 against the 4.5 bar; desktop door 3.64:1; the h1 paints 5.07 and 5.78 against the large bar 3.0; every other text box passes and unmeasured=0 on both doors; the ONE defect on both doors is the eyebrow, and no bytes ruler can see it: a photograph is not a token pair
2026-09-23	CM6	measure	nx_uiq_contrast (the bytes ruler, run on the NAS through /api/organ_run) reads the same page FAIL fails=6 on --ink x --ink-soft 2.06, --ink-faint x --ink-soft 1.67, --accent x --ink-soft 1.28 in both themes: pairs a name heuristic makes (a muted-text token read as a surface) that never paint as text on a surface; the two rulers disagree in both directions and only the pixels adjudicate
2026-09-23	CM6	lesson	A SHOT TAKEN BEFORE THE PAGE'S TIME-DRIVEN MOTION SETTLES READS AS UNPAINTED TEXT: the first phone capture caught the hero mid-reveal (ink=0 on the paragraph, 158 px on the h1, the CTA painted entirely in its own bg) and the referee reported two contrast defects that were the instrument's; the driver now waits for finite document-timeline animations (scroll-driven timelines and infinite iterations excluded, -1 at settle_ms), the receipt carries motion_running, and the referee refuses an UNSETTLED capture by name rather than grade it
2026-09-23	CM6	lesson	the second method class for a contrast kernel is the WCAG definition worked by hand beside it (white on rgb 200 = 167, on 153 = 285, on 17 = 1888): the kernel truncates (284 for 2.849) so the gate holds those to +-1 and 167 exactly; a KAT computed by the kernel under test would have proved transcription
2026-09-23	CM6	land	nx_page_pixjudge SHIPPED sha c38adac8ed4442e0ededec6fca4c4d5942af8c14c11a1ba00865808f1cfa0ad1 (nx_organ_ship: PROVE nx_page_pixjudge_gate 30/30 GREEN on the NAS through the organ_gate.conf declaration, the five browser legs announced E2E-NOT-RUN-HERE; first promote; offc twin; REGISTERED) -- sources on both trees byte-identical: nx_page_pixjudge_lib.nx 6303c29c 22471 B, nx_page_pixjudge.nx e978e1c4 9007 B, nx_page_pixjudge_gate.nx 73078451 29860 B, knowledge/pixjudge.conf (bars as data with WCAG 2.2 citations, ink tolerance and floor calibrated on the fixture, the e2e fixture rows); the laptop gate reads 35/35 with the real browser (phone and desktop captures of the fixture page graded DEFECTS contrast_fails=2 unmeasured=0); the matrix row repointed to the lib that declares pj_referee
2026-09-23	CM6	land	CM13 extended for the referee: nx_cdp_capture writes <label>.boxes (the DOM box map, CD-BOXES line, nboxes on the receipt) and waits for the page's finite time-driven motion before each shot (scroll-driven timelines and infinite iterations excluded, -1 UNSETTLED at settle_ms, motion_running as the receipt's 15th field); nx_cdp_lib cdv_boxes_rows + the receipt row; nx_cdp_capture_gate 35/35 on the laptop (15-field round trip, every-door-settled reads exactly 0 on the still fixture); the re-ship carries one enumerated loss, the old usage sentence superseded by the one naming the three appended fields
2026-09-23	CM6	queue	EMITTER: the showcase eyebrow (11 px white over the hero photograph) paints 3.00:1 phone / 3.64:1 desktop against the 4.5 bar on the live gospelofwest.com index -- the referee's one live finding; the remedy is in nx_site_build's hero (a scrim behind the eyebrow, or a heavier eyebrow with its own paper), proven by re-running the referee on a fresh capture, never by loosening the bar
2026-09-23	CM6	queue	BYTES RULER: nx_uiq_color's uiq_is_surface reads any token whose name carries soft as a SURFACE, so --ink-soft (a muted text colour) is paired as paper and the page reads FAIL 6 on pairs that never paint; the referee is the adjudicator today; the heuristic wants a declared role per token (data), not a longer name list
2026-09-23	CM6	queue	nx_gate_bite over nx_page_pixjudge_gate with subject nx_page_pixjudge (and the CM13 gate with subject nx_cdp_capture) is off this seat's surface; the planted controls bite in-process (grey-on-grey ruler control, hero contrast, containment, 2 px brush) and the mutation run is owed from a seat that carries nx_gate_bite
2026-09-23	CM13	land	nx_cdp_capture re-promoted for the referee: candidate 07310c9c61ff2216e8854e04b6d1d40113a253899d7662f8b24026dcbc3332dc (140232 B, prev banked) -- nx_organ_ship BUILT it, PROVED nx_cdp_capture_gate 26/26 GREEN on the NAS (35/35 with the browser on the laptop), the content ruler named ONE lost run (the usage sentence superseded by the one naming boxes
2026-09-23	EC-SHIP	queue	nx_organ_ship cannot close a browser-bound or stdin-bound organ on a host without the oracle: its behaviour stage demands a complete successful representative probe and its observe verb retains no observation record to review, so the review lane is unreachable by construction for exactly the organs that need it; owed on the ship loop: a declared-oracle-absent behaviour path (the organ_kind or a conf row names the host precondition, the loop records the refusal-path IDENTICAL probe plus the proving host's gate receipt as the observation, and the review cites those)
2026-09-23	CM8	land	scene section chain shipped on the NAS: nx_wasm_vm memory.size and memory.grow with nx_wasm_vm_memgrow_gate 094d3f53 21 of 21; nx_nishi_scene_gate 0da88eb5 96 of 96; nx_site_build c692a0f9 then the mount build; nx_ui_audit 5972a972 (external script, never a self-contained first-party one, costs the 15th point; ui_layout_gate 37 of 37); module knowledge/scene/nx_scene_hero.wasm d71d6e68 57281 B on both trees; catalog knowledge/scene_catalog.conf 52ec6401
2026-09-23	CM8	measure	done-rule on production bytes: gospelofwest v8 re-emitted by the promoted emitter is byte-identical to its release, 7 of 7 generated files; the same blueprint plus one scene row grades 14 of 15 like the control 14 of 15, every craft count equal or higher, script_ext 0; the missing point on both is imagery because knowledge/provenance/assets.jrnl holds 7 rows and none is an image
2026-09-23	CM8	measure	the full-score clause is now a gate tooth: both fixture pages carry the engine's own native frame as an estate-generated hero under a scratch media ledger, base 15 of 15 and scene 15 of 15, each external script shape 14; bite: the scene page without its hero row reads 14 and the gate goes RED 93 of 96 while the base tooth holds
2026-09-23	CM8	measure	first alpha through plan-site-release-cmsscene- (24 rows: 9 registry rows by nx_store_put put, draft or asset, build, audit, 3 prepare-path, 9 publishes index last, page_verify), 24 of 24 ok, page_verify GREEN; but the host router injected an estate ad aside and a sendBeacon view script (served 35363 B for a 34368 B release) because the page lacked the router's nx-ad-optout marker; gospelofwest.com production carries no ad, its alpha preview carries both
2026-09-23	CM8	land	the ads row (value optout): nx_site_ads_lib writes the router's own marker as a body comment right after ed_doc_open (that opener closes the head itself, so a meta there would land in the body); refused by name for another value or a tier that does not write it; nx_site_ads_gate 945a412c 21 of 21 reads the page with the router's own hr_has_optout, bite killed 1 of 2 (the survivor only silences the lib's stdout); after the re-run the served index equals the release, 0 ad slots, 1 script
2026-09-23	CM8	measure	in a real browser (CM13, Edge, phone 390 and desktop 1440) the door read css-only refused sovereign-cpu:failed:error: http 404; the capture profile's cache held /alpha/scene/orbit.wasm and /alpha/fonts/LiberationSans-Bold.ttf for a page at /alpha/cms-scene/: the edge's clean-URL form (hr_canon strips a directory's trailing slash) serves the index at /alpha/cms-scene, so every relative href on it resolves against /alpha/; the module, the display face and the nav were broken on every alpha and beta preview index, gospelofwest's included; nx_page_verify stayed GREEN (no browser, and its relative-asset law reads src attributes only)
2026-09-23	CM8	land	a MOUNTED release: the emitter derives the mount from the path part of its domain argument (nishifamily.com/alpha/cms-scene gives /alpha/cms-scene/, sb_mount) and switches its one prefix pair to it, so every href, the face url() and the module data-href are root-relative from the mount; a bare domain keeps every byte; nx_site_mount_gate 20 of 20 on the laptop resolves every reference against the landing URL the router's own hr_canon computes: mounted index 17 refs 0 escapes, mounted card page 18 refs 0 escapes, the bare release served under the mount 17 of 17 escape (the neg-control is the defect itself)
2026-09-23	CM8	measure	CM6 referee on the capture: desktop DEFECTS contrast_fails=1 overlap 0 hierarchy 0 (the 12 px eyebrow at 3.22 to 1, the showcase token defect CM6 named on gospelofwest); phone contrast_fails=2 overlap_pairs=6, where the 6 overlaps and one contrast fail are the collapsed menu's hidden links, whose boxes the capture records and the referee measures over the heading
2026-09-23	CM8	lesson	a done-rule that names a score must name the base it starts from: production showcase pages read 14 of 15 because no image can earn the imagery point, so the clause was untestable there until the gate built its own full-score base
2026-09-23	CM8	lesson	a preview is the release only if the edge serves it the way the release was written: relative hrefs on a directory index and a canonical form that strips the directory's slash cannot both hold, and a static verifier with no browser passes a page whose face and module both 404
2026-09-23	CM8	lesson	a behaviour probe over a gate that prints its per-run fixture directory reads DIFFERS between two runs of the same binary: run the live binary twice, mask the run token, then locate the probe's offset inside it before writing the review
2026-09-23	CM8	queue	gospelofwest's alpha recipe builds a mounted release (domain nishifamily.com/alpha/gospelofwest into its own release dir) and carries the ads row (value optout), so the preview stops loading no face and showing an estate ad its production never serves; the same for every alpha or beta built by nx_site_build
2026-09-23	CM8	queue	nx_page_verify's relative-asset law also reads CSS url() and script-fetched data-href on directory pages, so it stops passing a page whose face and module resolve against the parent
2026-09-23	CM8	queue	the capture box map carries rendered visibility and the pixel referee skips boxes that were not painted, so a collapsed menu's links stop reading as overlaps
2026-09-23	CM8	queue	the scene loader's refusal names the URL it tried, so a 404 names its path instead of costing an investigation
2026-09-23	CM8	queue	poster frame for the css-only door: the scene's first frame rendered natively in the site's palette for both themes through nx_png_write_rgb, admitted as estate-generated, so the door shows a picture and the page earns the imagery point
2026-09-23	CM8	queue	WebGPU and WebGL2 passes in the module; frame time measured by CM13 in a browser (frame_ms 40 is provisional); the browser's digest compared with native 753114568; module memory sized from its framebuffer (365 pages today)
2026-09-23	CM8	queue	workstreams.conf binds site-release to one site's plan (site-release-gospelofwest); a second site needs a per-site plan id rather than a second row with the same key
2026-09-23	CM8	measure	in a real browser after the mounted release (CM13, Edge, one 1440 by 1700 frame): the band takes its canvas column and the sovereign CPU pass draws the core and three moons; the door line reads door sovereign-cpu refused webgpu-compute:failed:pass-absent-in-module, webgl2:failed:pass-absent-in-module, sovereign-cpu:taken threads=1 module-single-thread; the served index equals the release, 1 script, 0 ad slots, 17 root-relative references
2026-09-23	CM8	measure	CM6 referee on that frame: DEFECTS contrast_fails=2 overlap_pairs=0 text_over_image=0 hierarchy_violations=0; both fails are the small uppercase eyebrows (3.22 and 3.80 to 1 against 4.5), the showcase token defect CM6 already named, not the scene; the scene band itself draws no finding
2026-09-23	CM8	land	shipped to alpha at https://nishifamily.com/alpha/cms-scene/ (release sites/cmsscene/releases/v1, recipe plan-site-release-cmsscene-), referee grade DEFECTS on the eyebrow contrast only, awaiting acceptance; gates rostered: nx_nishi_scene_gate, nx_wasm_vm_memgrow_gate, nx_site_ads_gate, nx_site_mount_gate (nx_site_mount_gate b57e93b9 20 of 20 on the NAS)
2026-09-23	CM8	queue	the scene reads nearly monochrome because all four bodies take neutral roles (ground, accent-wash, rule, card), and the CPU pass draws the core's rim without anti-aliasing; both are visible at 1440 and both are the next quality rungs for the scene, to be judged by the referee panel and the operator, not by the seat
2026-09-25	CM8	measure	the gospelofwest alpha at nishifamily.com/alpha/gospelofwest was an indexable duplicate of the live client site: nishifamily.com robots.txt disallows /preview/ and not /alpha/, and no alpha page carried a robots meta; its release recipe also built for the production domain, so its relative links resolved against /alpha/ under the edge's slash strip, and the router injected an estate ad and a view beacon into a page whose own copy says no ads and no trackers
2026-09-25	CM8	land	nx_site_stage_lib (51a757e0): a release STAGE is declared on the build line by its recipe; stage=alpha or stage=beta writes a robots noindex meta inside every page head on every tier (ed_doc_open for editorial and showcase, sk_head for the kit, the CM15 static-switch pattern, no syscalls so the kit keeps its contract); stage=production and no token are the bytes the emitter always wrote; a misspelt value, an empty one or two tokens refuse the build before any page exists. nx_site_build d62bb0fc SHIPPED with one enumerated loss (the kit's head-close literal, split in two so the meta can sit between); nx_site_stage_gate 7d341970 47 of 47 on the NAS, registered
2026-09-25	CM8	measure	controls for the stage change: andelinwest (kit tier) re-emitted 7 of 7 byte-identical without a stage on the laptop; the pre-ads gospelofwest v8 blueprint (showcase) re-emitted 7 of 7 byte-identical to releases/v8 on the NAS by the promoted emitter; the ship probe on the kit workload read IDENTICAL; every armed head and page in the gate is exactly the 38-byte meta longer than its plain twin
2026-09-25	CM8	land	gospelofwest alpha rebuilt through its own recipe (plan-site-release-gospelofwest-, putmany of rows 1 to 11): gospelofwest_v8.site gains the ads optout row, the build goes to sites/gospelofwestcom/alpha (production serves a different release, current is not releases/v8), domain nishifamily.com/alpha/gospelofwest, stage=alpha; 12 of 12 steps ok, page_verify GREEN; served: the noindex meta inside the head, the opt-out marker, zero beacons, 19 links mounted under /alpha/gospelofwest/. The cms-scene recipe row 10 declares stage=alpha too: 24 of 24 steps ok, the published index carries the meta once
2026-09-25	CM6	measure	referee on the live gospelofwest alpha (CM13 capture, phone and desktop): desktop DEFECTS contrast_fails=2, both the 12 px mono eyebrows (hero eyebrow and section kicker) painting 3.21 and 3.94 to 1 against 4.5; phone DEFECTS contrast_fails=3 overlap_pairs=5, where two fails are the same eyebrows (3.76 and 4.11) and one fail plus the 5 overlaps are hidden navigation links the referee still measures (declared 14.74 to 1, painted 1.00, OWN-PAPER); by hand the ink-faint on ground pair is about 4.26 to 1, so the eyebrow fails before anti-aliasing
2026-09-25	CM8	lesson	a release built for one host and published under another's path is two defects wearing one symptom: its links resolve against the wrong directory, and its content is indexed as a duplicate of the host it was built for. An alpha is a preview, and the recipe that builds it must SAY so, because nothing downstream can tell a preview from a mounted production site by its path
2026-09-25	CM8	queue	EDGE: rung E6 filed on edge.plan with its matrix row (directory indexes keep their trailing slash, the root remedy; debt 1790351077 is its recurrence row beside the still-open 1785865143); the router-side X-Robots-Tag for preview prefixes the emitter does not build (today /alpha/beach.html) is debt 1790351101 for the sites daemon's owner
2026-09-25	CM6	queue	REFEREE: skip boxes that are not painted -- hidden navigation links read as contrast fails and overlaps on the phone door on both alphas; the capture's box map should carry rendered visibility and the referee should grade only painted boxes, with a neg-control that a visible link over its own paper is still measured
2026-09-25	CM6	queue	EMITTER: the showcase eyebrow and kicker ink (ink-faint on ground, about 4.26 to 1 declared, 3.2 to 4.1 painted) fails the 4.5 bar on both alphas, the no-photograph case of the eyebrow row above; the remedy is a derived eyebrow ink (the lightest palette ink that meets the bar against its ground in each theme), proven by a fresh referee capture, never by loosening the bar
2026-09-25	CM6	land	the referee now grades only PAINTED boxes: nx_cdp_capture records the browser's own Element.checkVisibility() as a 14th box field (1 painted, 0 not, -1 cannot say; nx_cdp_lib CDV_BOX_FIELDS 14), and nx_page_pixjudge_lib gives an unpainted box its own role so contrast, overlap, over-image and hierarchy all skip it, counted as unpainted with boxes = text + img + unpainted in PJ-GRADE; a zeroed slot reads UNKNOWN and is graded as before, so an older 13-field map and every hand-built gate table stay judged, never silently hidden. nx_page_pixjudge_gate 44 of 44 on the laptop (38 of 38 in-process on the NAS, SHIPPED 51ea8b1e) with bites both ways and an e2e closed-menu link reading unpainted on both doors; nx_cdp_capture_gate 35 of 35
2026-09-25	CM6	measure	the live gospelofwest alpha re-captured with the fresh driver and re-graded: phone boxes=18 = text 13 + img 0 + unpainted 5 (the closed menu sheet's four links and its call to action, painted=0 in the map), contrast_fails 3 to 2 and overlap_pairs 5 to 0; desktop unchanged at contrast_fails=2 (the menu is not rendered there). The two fails left on both doors are the 12 px eyebrow and kicker -- the real defect, queued above as the derived eyebrow ink
2026-09-25	CM6	lesson	asking a skipped subtree for its geometry FORCES its layout: a closed details sheet answers getBoundingClientRect with real boxes while painting nothing, so a referee that trusts geometry grades invisible text. The browser's own checkVisibility is the witness; the capture records it and the referee decides, and the value that means unknown is the one a zeroed slot holds
2026-09-25	CM8	measure	nx_site_stage_gate bite (nx_gate_bite over nx_site_stage_lib, 2 sites): valid_mutants=1 killed=1, restore byte-identical, both twins restored; rostered in knowledge/gateroster.conf. Gates re-shipped because this leg moved their closures: nx_cms_plugins_gate d645f836, nx_nishi_scene_gate b40b55b9 (behaviour review: the live gate differs from itself, its fixture directory is named from the clock)
2026-09-25	CM8	measure	DECLARED, NOT PROMOTED: the stage lib moved the import closure of every nx_sclass_kit and nx_nishi_editorial importer, so these served binaries read behind their source until their own lanes rebuild them -- nx_connect_page, nx_connect_lds_page, nx_connect_reconnect_page, nx_connect_youth_page, nx_fin_stack_page, nx_dataeng_compare_page, nx_site_archetype and the meal pages. Their emitted bytes are unchanged (the unarmed switch writes nothing: 7 of 7 byte-identical on the kit tier, 7 of 7 on the showcase tier), so a rebuild is a no-op on output. Re-shipped this leg because they are this lane's: nx_site_build, nx_site_stage_gate, nx_cms_plugins_gate, nx_nishi_scene_gate, nx_site_agentready_gate, nx_page_pixjudge, nx_page_pixjudge_gate, nx_cdp_capture (through /api/promote with the loss named: the ship loop needs a successful representative probe and this organ needs a browser the NAS does not have), nx_cdp_capture_gate
2026-09-25	CM6	land	the editorial palette's faint ink is DERIVED against the referee's own bar (source on the NAS, promote held for a sibling site seat's window): nx_ink_derive_lib walks the declared ink along its hue until its WORST ratio over ground, ground-sunk and card meets knowledge/pixjudge.conf contrast_min_ratio100; nx_site_build arms it for the editorial and showcase tiers and refuses by name when the row cannot be read or met; unarmed callers write the old bytes. Light #6E7269 -> #656860 in 79 permil steps (worst 4.52 on ground-sunk, 4.91 on ground, 5.28 on card, each worked by hand; one step back reads 4.47), dark #8A8C84 unchanged (4.85). nx_ink_derive_gate 39 of 39 on the laptop incl. fixture roots for every refusal and an AAA row that derives a darker ink
2026-09-25	CM6	measure	A/B on one emitted page where only the faint ink differs, captured in the browser and graded: old #6E7269 (declared 4.26 on ground) paints 4.11 on the phone door and 3.82 on desktop, both fails; new #656860 (declared 4.91) passes the phone door (worst 4.75, OK) and paints 4.34 on desktop, a fail at 4.5. The painted estimate runs 0.88 to 0.90 of the declared ratio for 12 px mono at 1x and about 0.97 at 2x
2026-09-25	CM6	lesson	a pixel referee that averages every pixel within its ink tolerance measures anti-aliasing along with colour: on thin strokes at 1x most ink pixels are partial-coverage blends, so a WCAG-compliant ink reads as a fail. The ink_tolerance row was calibrated on a 16 px sans face whose cores sit within 12 of the declared colour; 12 px mono at 1x is a second corpus it was never tested on. Fix the estimator, never darken the palette to fit the instrument's bias
2026-09-25	CM6	queue	REFEREE ESTIMATOR: estimate the painted ink from the pixels nearest the declared colour (the glyph cores) and the paper from its dominant colour rather than both means, calibrated on the A/B pair above (declared 4.26 and 4.91 must read within tolerance of declared) while the true positive, the production eyebrow over the hero photograph, still fails; read how the field's pixel-based checkers pick foreground and background before choosing
2026-09-25	CM9	land	a third personality in knowledge/design_corpora.conf, added by the connect SITE seat and used by the Connect alpha front door (nishifamily.com/alpha/connect-faith): social-trust = signal.org, hinge.co and meetup.com home pages plus the stylesheets they link (8 corpus rows, recipe showcase, additive only, mirrored 2026-09-25 by nx_research_fetch); derived n=4 over the stylesheets: t1 1600 ratio 1101 td 6200 s1 1200 s6 4400 lhb 150 bp1 67250; the lesson it carries: for a rival-class corpus the home pages alone measure almost nothing (every modern home ships its type in linked CSS), so each corpus row set must include the linked stylesheets or the derive reads n=0 on type and spacing
2026-09-25	CM6	land	the referee reads the PAPER the way the field reads pixels and the INK the way WCAG evaluates contrast: the Android Accessibility Test Framework's ContrastSwatch (exact-colour histogram, split at the unweighted mean luminance of the distinct colours, the modal colour each side) finds the paper mode under the text [atf-contrastswatch] [atf-contrastutils], and the declared text colour is held against it, because WCAG's Understanding note on 1.4.3 says to evaluate the colours from the user agent rather than the anti-aliased text as presented [wcag-understanding-1-4-3]. The exact method runs only where the pixels prove its picture -- some pixel is exactly the declared colour, and every pixel is that colour, the paper mode or a per-channel blend of the two, no threshold -- and everything else keeps the band means, whose bias only lowers a ratio; the field's own two-mode ratio is printed as atf100 and never judged, as the framework itself reports it only as a warning [atf-textcontrastcheck]. nx_page_pixjudge SHIPPED 65b3f864 (lib 02c668fd), gate SHIPPED ee72d68c at 50 of 50 on the NAS and 57 of 57 on the laptop with the browser leg, where every box of the solid-backgrounded fixture is judged exact on both doors
2026-09-25	CM6	measure	eight captures graded with the production bars, measured = exact + band summing on every run: the ink fixture with the derived #656860 reads OK on both doors, worst 4.91, 9 of 9 and 8 of 8 boxes exact; the same page with the old #6E7269 fails both doors at exactly 4.25, its declared ratio; the production gospelofwest home keeps its one true defect, the eyebrow over the hero photograph, 3.64 desktop and 3.00 phone by the band means (paper mixed: 3,333 and 6,241 unexplained pixels); the alpha still carrying the old ink reads its two faint lines at 4.25 on both doors, and its twelve pre-reveal card texts stay UNMEASURED (no pixel of their declared colour, the reveals had not run). Over every solid-paper box of three pages the band means had read 0.67 to 0.98 of the declared ratio and the field's two-mode ratio as low as 0.31, because on thin text its foreground mode is an edge blend
2026-09-25	CM6	retract	row 1790354901 read the derived ink as failing the desktop door at 4.34 against 4.5 and queued an estimator change: the number was the band means' anti-aliasing bias, not the page -- the same capture paints its glyph cores exactly #656860 (29 pixels) on exactly #EFEFE9, and the exact method reads 4.91 on both doors. The estimator change it queued is the land row above
2026-09-25	CM6	lesson	a fixture whose foreign colour sat outside the ink-paper interval on green and blue only let a mutant that disabled the red-channel check survive (nx_gate_bite aim pj_box_swatch site 82): three stripes, each outside on exactly one channel, killed it, and the mutants at the unexplained-count site were killed too. A per-channel rule needs a per-channel fixture, asserted as the fixture's own precondition before any outcome
2026-09-25	CM21	measure	BENCHMARK (operator, 2026-09-25): the Claude Code changelog [claude-code-changelog-2026-09] against this board, both captured in a real browser at both doors and graded by the same instruments. Pixel referee: ours DEFECTS on desktop by HIERARCHY -- the OVERVIEW h2 paints 40 px over a 37 px h1 -- theirs clean; contrast clean on both (worst 5.20 ours, 4.85 theirs). Box map, first viewport: longest text block 759 characters on ours against 147 to 189 on theirs, list items 0 against 3 to 7, text boxes on desktop 27 against 89 (their section rail, tabs and a one-link-per-release contents rail); on phone both titles land at 192 css px, their first release record at 577, our first section heading at 652 with the 759-character block beneath it. Bytes ruler: ours ABOVE theirs (hygiene 12 against 11, craft 10 against 7) -- it counts ingredients, and none of the qualities that separate the two pages is a count it takes
2026-09-25	CM21	lesson	a ruler that grades our page above a benchmark it visibly trails is measuring presence, not the reader's task: the gap is scannability and wayfinding -- block length, list share, a contents rail, the current item marked, identifiers as code -- and every one of them is measurable from the box map the capture already writes. Grow the ruler; never tune a verdict to agree with an eye, including this seat's (its first phone reading, that our navigation ate the first screen, was refuted by the box map: both titles land at 192 css px)
2026-09-25	CM20	queue	the benchmark is this rung's done-rule made concrete: a section rail that marks the current section, a contents rail with one link per record, the rails folding into one breadcrumb line on phone; and the shared emitter's section heading must never outrank the page title -- the referee's hierarchy tooth fires on this board today (h2 40 px over h1 37 px), so the fix is in the one generator every board uses, shipped through its lane
2026-09-25	CM21	queue	every ledger record renders the way the benchmark renders a release: an anchored chip (rung and date), the text as short bullets, identifiers as inline code; done when the box-map census reads no block in the first viewport longer than the benchmark's longest (189 characters, measured 2026-09-25) and the first record on phone no lower than the benchmark's (577 css px)
2026-09-25	CM19	queue	add the benchmark's page and stylesheets to the board-class derive corpus beside MDN, GitHub Docs and Tailwind, so board tokens are derived with it in the population, never copied from it
2026-09-25	CM6	queue	a SCANNABILITY and WAYFINDING axis for the referee, read from the box map it already loads (longest block, list share, first-record offset, rail presence, current-item marking), reported beside contrast and graded against measured benchmark pages -- the axis the bytes ruler cannot have
2026-09-25	CM24	land	THE PAPER BEHIND THE GLYPHS, SHIPPED: the capture driver shoots every door a second time with the text painted transparent (a style the page is given and loses; the page restored or every later door refused by name) and writes <label>.paper.png plus two receipt fields; the referee reads it and grades the declared ink against the measured paper by WCAG's failure technique F83 [wcag-understanding-1-4-3] -- the paper most like the text behind any glyph pixel, and when that fails, letter by letter (a letter is an 8-connected component of the pixels the two shots disagree on). nx_cdp_capture 63b815b8, nx_cdp_capture_gate 41024282 (27 of 27 on the NAS, 39 of 39 on the laptop with the browser leg), nx_page_pixjudge 211a29cc, nx_page_pixjudge_gate 38dd9194 (60 of 60 on the NAS, 67 of 67 on the laptop); mutants at the blend-check, unexplained-count and core-counter sites KILLED; both gates rostered. Done-rule met on its fixtures: a gradient running from the ground toward the ink passes at 4.91 by the exact method alone and fails on its own paper
2026-09-25	CM24	measure	production gospelofwest home, captured with the text-free shot and graded: the h1 over the hero glow FAILS by the letter it sits over -- worst letter 2.71 desktop and 2.67 phone against the 3.0 large-text bar -- where the band means had read 5.78 and 5.07 (a defect the averaging hid); the eyebrow PASSES on desktop (quick check 5.18, worst letter 5.77, where the band means read 3.64 -- the anti-aliasing bias) and FAILS on phone (quick check 2.48, worst letter 2.60); every solid-paper box reads its declared ratio exactly, the paper method agreeing with the exact method on all 17 boxes of the ink fixture
2026-09-25	CM24	lesson	two capture-driver defects found by using the driver, both fixed at the root: (1) a browser error page is complete and loaded like any page, so a certificate interstitial was captured and graded as the site with verdict OK -- the driver now reads Page.navigate's own errorText and refuses by name, and readiness requires an http, https or file document; (2) a WARM browser answers /json/version before its first tab exists, so a capture launched right after another read an empty target list -- the list is now polled on the boot budget. The first was a vantage artefact (the laptop's network was in transition; the edge's certificates verify), found only because the capture checked its own pixels
2026-09-25	CM6	land	THE DERIVED FAINT INK IS LIVE: nx_site_build e6a81423 SHIPPED through the ship loop. The 4 lost runs are the static palette literals, now assembled around the derived token (allow_loss efc73062); the behaviour change was reviewed from the outputs, not the counts (review 6a557794, evidence knowledge/status/harness/evidence-nx_site_build-1790372800.txt): on the gospelofwest showcase workload the candidate's stdout is the live stdout plus the INK receipt line (nx_srcdiff B-SUPERSET, 55 lines kept, 1 added), and each of the 5 emitted pages equals the live page with #6E7269 swapped for #656860, byte for byte. nx_ink_derive_gate f721b47f 39 of 39 on the NAS, closure banked, on the SLOW roster by measurement (2,999 ms; its trial at the fast roster's 1,800 ms bound was killed at 1,848 ms). Both alphas rebuilt through their recipes: gospelofwest 12 of 12, cms-scene 24 of 24, page_verify GREEN on both
2026-09-25	CM6	measure	both rebuilt alphas captured with the text-free shot and graded by the production bars: gospelofwest OK on both doors, contrast_fails 0, worst 4.91, 14 of 26 desktop and 10 of 18 phone boxes read by the paper method (the same alpha with the old ink read its two faint lines at 4.25 and failed, row 1790358501); cms-scene OK on both doors, worst 4.91, 11 of 13 desktop and 9 of 15 phone boxes measured. Still UNMEASURED on the gospelofwest desktop door: 12 boxes, the three feature cards at the fold (y 733 to 890 of 900, four text boxes each) caught mid-reveal by the stagger preset, no pixel of their declared ink
2026-09-25	CM6	lesson	an emitter's behaviour change is reviewed from the outputs, and the proof that nothing else moved is a substitution identity: capture live and candidate on the ship probe's own argv (the candidate through a scratch row on its stage file, reaped by nx_toolreg_reap once the promote consumed it), line-diff stdout with nx_srcdiff, and show each page equals the live page with the one derived token swapped -- a byte count says the sizes match, only the identity says the change is that token and nothing else. And a gate's roster is decided by its measured runtime (nx_gate_roster_run trial, then admit), never by where the last gate was hand-appended: the two gates hand-added earlier today measured 1,203 and 702 ms and stay on the fast roster, while this one would have read TIMEOUT on every fast beat
2026-09-25	CM6	queue	MEASURE THE RESTING STATE OF SCROLL-REVEALED TEXT: the capture shoots the first paint and settles only finite document-timeline motion, so text that a scroll-driven preset reveals at the fold is UNMEASURED on every run (12 boxes on the gospelofwest desktop door). Done when the capture declares a reduced-motion door (the page's own guard disables every preset under prefers-reduced-motion) or drives each box through its view range, and those boxes read a ratio
2026-09-26	CM13	queue	W11 connect visual-review seat, three measurements for the capture lane: (1) the shared profile C:\Users\elder\AppData\Local\Temp\nx_cdp_prof carries a SAVED credential for nishifamily.com -- the alpha login painted an autofilled handle and a masked passphrase while the served HTML carries no value (count 0 over 11278 B), and a clean profile of the seat's own painted the placeholders with the same referee grade; a capture that inherits another run's browser state is not a first visit, bakes an operator handle into evidence PNGs, and keeps a password in a TEMP profile every seat shares; debt 1790438914 (a fresh user-data-dir per capture, the password manager off in the launch argv). (2) a full-height door on a scratch conf (same oracle and widths, 390x6000 dsf 2 giving ihdr 780x12000 in 930537 B, and 1440x3600) captured and graded cleanly and measured the fold cards the standard doors leave UNMEASURED (the connect-faith about pages 48 of 48 and 54 of 54 boxes measured) -- a candidate door row for pages whose content lives below the fold, with the cover-range caveat that a view() reveal completes only in the upper part of a tall viewport. (3) the referee's text_over_image flag reads every text box on a page whose body carries a CSS gradient (the capture maps body as an img box): 19 of 19 and 29 of 29 on the connect alpha home, so the flag discriminates nothing there; the verdict is unaffected because the paper method grades.
2026-09-26	CM15	queue	W11 connect visual-review seat: ed_sc_foot (nx_nishi_editorial) appends the dated validity signal (a time element reading Updated <date>) straight after the colophon text with one space, so an armed release whose colophon ends without punctuation paints one run-on line -- measured on the alpha front door https://nishifamily.com/alpha/connect-faith: CONNECT ALPHA -- INDEPENDENT; NOT AFFILIATED WITH OR ENDORSED BY THE CHURCH OF JESUS CHRIST OF LATTER-DAY SAINTS OR ANY CHURCH UPDATED 2026-09-25. One fix in the emitter (a separator or its own line before the time element) covers every armed showcase site; nx_site_agentready_gate's tooth checks the time element's presence, not its separation, so a separation tooth rides the fix.
2026-09-26	CM13	queue	W11 connect visual-review seat, two more gaps the capture lane owes, both named from this review: (1) NO DARK-SCHEME DOOR -- both connect alphas ship a dark theme (the front door @media prefers-color-scheme dark plus :root[data-theme=dark]; the app's default palette is dark and light is the media query), and every capture this seat took is the light scheme because the driver never calls Emulation.setEmulatedMedia; the dark palettes stay UNMEASURED on the painted page (declared only: front door --ink-faint #8A8C84 on #14161A about 4.85, app --faint #6a7488 on #0b0e15 4.11); owed: a door column or conf row carrying the colour scheme. (2) THE RECEIPT NAMES THE REQUESTED URL, NOT THE DOCUMENT THE PIXELS CAME FROM -- a capture of a local page that posts a form painted the POST response (https://nishifamily.com/alpha/connect/register, the register result) while capture.receipt and CD-SETTLED both say file:///...submit.html; the driver already reads location.href (CD_EVAL_HREF) but prints it only on a settle timeout. Owed: a document field on every receipt row, so a page that navigates after load is filed under the subject that was actually graded.
2026-09-26	CM6	queue	W11 connect visual-review seat, a referee FALSE POSITIVE on real data, classified (a first-draft hit is a bucket, not a finding): on the Connect alpha account page (desktop door) nx_page_pixjudge graded box div 190,99,1060,676 -- the page wrap, a text box only because a markup defect leaves a 54-character clause as a direct text node in it -- DEFECTS at painted 2.84 with letters=225 and exact100=1518: the F83 letter search ran over every glyph pixel inside the wrap's box, children included, and graded the wrap's declared ink #182031 against a child's paper, the Sign-out button's #2f5fd0 (2.85 by hand; the button's own white text on that blue paints 5.33 and passes). A container text box must be graded on its own text runs only, the pixels of descendant boxes the map already grades excluded; bucket name for the census: container-over-descendants. Fixture for the tooth: a div holding one bare text node beside a child button with a declared ink and a solid dark background. Until then a DEFECTS on a box whose letters far exceed its textlen is a question, not a verdict.
2026-09-26	CM14	land	DESIGN DECISIONS VERIFIED IN THE BYTES, SHIPPED: nx_ui_tokenref_lib (ua_token_ref) reads every custom property a page defines inside its style blocks and every var() reference on the page, refuses a defined-but-unreferenced property by name unless a marker row exempts it, judges every blueprint row the claims table (knowledge/design_claims.conf: fields, field:i, needle, prefixed, narrowed by key=value and key@i=value, per surface) declares and refuses a claim that does not land, and reports every derived token row; nx_ui_audit tokenref is the verb; nx_ui_audit_gate 1531b98c 38 of 38 on the NAS with the live leg over the rebuilt alpha (35 of 35 in-process on the laptop, the live leg SKIPs there by name), declared in organ_gate.conf for nx_ui_audit and the lib
2026-09-26	CM14	measure	the ruler's first real run REFUSED yesterday's showcase alpha on three findings the eye had passed: the type scale defined --t4 and nothing referenced it, the blueprint's heronote row never landed on the index, pagenote never landed on a sub-page, and the section_trust kicker, heading, lead and note landed nowhere (the band carried the sigs under the steps' heading); nx_site_build 96c3feb3 fixes all four (the unreferenced token gone, the note under the hero actions, a trust group heading the sigs), both alphas rebuilt (gospelofwest 12 of 12, cms-scene 24 of 24), and the live leg reads CLEAN on the index and a sub-page
2026-09-26	CM14	lesson	a blueprint is a set of claims and the emitted bytes either carry them or do not: a row the emitter silently drops is design theater with the operator's own words, and a token the scale defines that no rule reads is the same defect in CSS; both were invisible to every ruler that counted ingredients, because a count of definitions rises when theater is added. The claims table is data so a new key is a row, never a rule, and a key no row names is counted unclaimed rather than acquitted
2026-09-26	CM24	land	the dark contract is lit: the text-free shot is its own entry point, cd_textfree_shot in nx_cdp_capture (the style on, the paper shot through the same writer as the page shot, the style off, the restore reported), extracted with no change to a byte of output; nx_cdp_capture 3cb33eb3 promoted with lost_from_live=0, _offc twin installed, nx_cdp_capture_gate 27 of 27 on the NAS and 39 of 39 on the laptop with the browser leg
2026-09-26	CM19	land	the watch row named a file that never existed (runtime/_hdl_build/nx_ui_layout_lib.nx) while the derivation lived in runtime/nx_ui_layout_lib.nx and board.tokens carried its six rows with n; the row is repointed, nothing rebuilt
2026-09-26	CM25	queue	the research the operator handed the board is landed as data, not prose: fourteen refs pinned from mirrored primary pages and arXiv API listings read this month, knowledge/design_rubrics.conf (Awwwards D40 U30 C20 Cn10, the Webbys' seven, the FWA's bands, NN/g's ten, Baymard's topics), the bars SB-RUBRICS, SB-RSI-LOOP and SB-UIOPT with their targets T-BIC-RUBRIC and T-FRONT-LOOP, milestone M10 with CM25 to CM29; CSS Design Awards is UNFETCHABLE by the estate client today (TLS 1.3-only host) and its rubric is not landed; the paste's FWA axis labels were not in the primary bytes and are not landed either
2026-09-26	CM17	queue	the operator's standing direction 2026-09-26: no navel gazing -- the referees exist so the loop can improve what is emitted; the spine is CM17 with CM26 as its browser-side receipt and CM27 as its memory, then CM28 and CM29; every rung's proof is a planted defect repaired and a refusal by name, never a score that only rises
2026-09-26	CM10	queue	the listings the corpus refresh will read beside the Awwwards Site of the Day page: the Ecommerce Design Awards log [eda-home-2026-09], Siteinspire's daily showcase [siteinspire-2026-09] and Land-book's landing-page gallery [landbook-2026-09], each mirrored this month; a listing row names the page and the beat mirrors the winners it names with their stylesheets, so the three design philosophies the field converges on (immersive 3D storytelling, precision SaaS minimalism, accessible consumer clarity) become personalities in design_corpora.conf derived from named evidence, never typed
2026-09-26	CM26	land	THE JOURNEY DRIVER, SHIPPED: nx_cdp_capture journey <name> <out> <journeys.conf> walks a site's declared path at every door of the capture conf -- land, click <selector>, fill <selector>, expect <needle> -- recording per step whether the target was found, whether it sat in the viewport at the moment of the click, how many viewport heights down it sat, the settle time after it (a click that navigates settles the new document too) and whether the expectation held in the href or the visible text, then the actions a user pays (clicks + fields + scrolls) against the journey's declared bar; INCOMPLETE, WORSE-THAN-BAR or OK per door, the worst door decides, every row in journey.receipt. One session opener now serves the capture and the journey (cd_session_open), the decisions live in nx_cdp_lib (cdv_journey_verdict, cdv_pipe_field, cdv_atoi_z, cdv_step_kind) so the gate proves them in-process. nx_cdp_capture f01a4ada promoted with lost_from_live=0, gate 43 of 43 on the NAS and 59 of 59 on the laptop where the e2e legs run: the fixture path OK within its bar, a planted extra step WORSE-THAN-BAR, a missing target INCOMPLETE naming the step
2026-09-26	CM26	measure	knowledge/journeys.conf declares two paths with the funnel floor's own promise as the bar (3 actions): gospelofwest (land, Start reading, first theme card) reads OK on both doors at 2 actions (2 clicks, 0 scrolls, settle 675 and 414 ms); cms-scene (land, See the scene, first note card) reads OK on both doors at exactly 3 actions -- the scene band pushes the first card one viewport below the CTA's landing, so the path costs a scroll on every door; the friction receipt names it, nothing is tuned
2026-09-26	CM26	lesson	a numeric field read as 0 on every step while the label beside it read the clicked text -- cdv_atoi returns the index after the digits, never CDV_OK, and a reader that compares to the wrong contract silently counts nothing (the direction that under-reports friction); the fix is the contract read from the lib, and the gate now holds atoi to its arithmetic. And an in-page action that navigates leaves the OLD document complete for a moment: a settle poll that returns at once on readyState measures nothing, so the driver re-reads the href after the click and settles the new document when it moved
2026-09-26	CM6	land	W13 SEAT ALPHA-UI-AND-PROD (connect lane, debt 1790438899) extended nx_ink_derive_lib ADDITIVELY (buildroot/runtime/_hdl_build, 6821 to 9358 B, sha256 aefcd17b): inkd_conf_bar reads the referee's bar row from conf text (the key followed by the field separator at a line start; the value an unsigned decimal ending at the next field, a CR or the line end; a comment line and a longer key sharing the prefix are not the row) and returns the value or a named refusal INKD_E_ROW or INKD_E_VALUE, both named by inkd_code_name; inkd_meet and every existing function are unchanged. First consumer: nx_connect_appd, which derives the Connect app's --faint in both themes over --bg --surface --surface-2 at startup (light 8b94a6 to 686F7C, dark 6a7488 to 80899A); proven by nx_connect_ui_gate T13 (six teeth incl. neg-controls, 39/39 on a fresh compile, SHIPPED 1734504731e0aded) and the live alpha pixel referee (the footer 471 where it read 284). HONEST LIMIT: the nx_ink_derive_gate re-run after the edit read 39/39 on a BUILD-CACHE HIT (key 0f93a9af, job 1790456373), so it is evidence about a cached artifact, not a fresh compile of the edited lib; two --no-cache re-runs were deferred by build admission (load1 17.15 and 16.90 over the 16.43 ceiling, jobs 1790456542 and 1790456894). The edited closure was compiled fresh by the connect ui gate and the connect daemon, which exercise inkd_meet inkd_worst inkd_walk inkd_hex inkd_hex6 inkd_ratio and inkd_conf_bar. QUEUE for the cms owner: nx_site_build sb_ink_arm still reads the same row through sb_cg sb_fld sb_uint, so two readers of one bar row now exist; migrating sb_ink_arm onto inkd_conf_bar makes it one (not done here: nx_site_build is the cms lane's file), then re-run nx_ink_derive_gate with --no-cache.
2026-09-26	CM17	land	THE LOOP IS BUILT AND PROVEN: runtime/_hdl_build/nx_site_refine_lib.nx carries sb_refine_loop (CM17) and sr_archive_select (CM27), nx_site_refine is the CLI (loop <site> <out_root> <url_base> <journey> [budget=N] [refconf=], archive <out_root>), nx_site_refine_gate reads 62 of 62 GREEN on the laptop with the browser leg. EVALUATE: every variant is emitted by nx_site_build and judged by the estate's referees forked as separate binaries (nx_page_pixjudge at every capture door, nx_ui_audit file and tokenref, nx_cdp_capture journey), their fields the score vector and their own vocabulary the defect names. REFINE: a named defect maps through knowledge/refine_remedies.conf to one knob move, a token row moved one step (the larger of the page's own scale step t6 over t5 and the referee's bar) inside a declared evidence band, or a blueprint row set from the catalogue; a content-owned class is named as residual, a class the table does not declare refuses by name. IMPROVE: every variant archived content-addressed with its vector and verdict, the next parent the best admitted by objective then novelty, a child admitted only when no defect or score axis worsens and one improves, an axis the parent measured that the child cannot a rejection never an acquittal; the budget is charged per move and derived from the remedy rows when the caller gives none
2026-09-26	CM17	measure	end to end on the laptop: a showcase fixture with a PLANTED flat type scale (t5 3022, step 1049 permil against the referee's 1178 bar) reads hierarchy_step_short=2 at the root (h1 35 px beside h2 32 px at the phone door), the loop moves td 10900 to 12840 by the referee's bar, the child reads 0 short with every other axis equal and is ADMITTED, and the run closes CONVERGED-RESIDUAL naming craft_missing:img (the imagery point is the client's photograph); budget=0 refuses REFUSED-BUDGET with the root judged and nothing moved (bite-proven against the converging run); a remedy table without the class refuses REFUSED-UNDECLARED:hierarchy_step_short. Archive rows 2, receipt rows iter
2026-09-26	CM17	lesson	the loop's first runs were refused three times by its own referees, and each refusal was a defect in a referee or a driver, never in the loop: a display sized at the heading read OK (the strict hierarchy axis), a moved child failed contrast on a card caught mid-fade at the fold (the band means judged blends), and the file-door journey died on a clean-URL link (chrome-error); each became an instrument change with its gate tooth before the loop was believed. A loop that repairs what a referee cannot see is theatre; the referees were grown until the loop's convergence meant something
2026-09-26	CM6	land	THE SCALE-STEP AXIS: pj_hierarchy convicts only a text box LARGER than the h1, so a display sized at the section heading read OK (measured: a planted variant with td at t6 painted h1 32 px beside h2 32 px, hierarchy_violations=0). nx_page_pixjudge_lib pj_hierarchy_step counts the non-h1 text boxes the h1 does not lead by one step of the type scale (h1 x 1000 below font x hierarchy_min_ratio_permil), disjoint from the strict axis, UNMEASURED without an h1 or a bar; the bar 1178 permil is knowledge/pixjudge.conf's new row, calibrated from the derived profile's own ratio row (n=4); the grade line carries hierarchy_step_short, a short box flags HIER-STEP and counts toward DEFECTS. Two abstentions beside it: TEXTFREE-MISS (glyph cores painted while the text-free shot removed nothing under them) and NO-CORE (ink inside the tolerance band with no pixel at the declared colour: a scroll-driven reveal mid-fade), each UNMEASURED, flagged and counted on its own axis, never a manufactured verdict. Gate nx_page_pixjudge_gate 88 of 88 on the laptop, 81 of 81 on the NAS (0286aafd): planted h2 at the h1 size flips the verdict on the new axis alone, an h2 one step below is the positive control, agreeing shots and a faded run abstain and a lone abstaining box reads UNPROVEN
2026-09-26	CM6	measure	the new axis on real pages before the emitter fix: the loop fixture at the phone door read h1 35 px beside h2 32 px (one short, 35000 below 32 x 1178) and the LIVE gospelofwest alpha at the phone door read h1 32 beside h2 32 (one short); both desktops read 0 short (109 and 93 px displays). The showcase clamp bottomed at t6, so every phone-door display sat within a step of its section headings by construction, and the fitter sent every long headline there on purpose
2026-09-26	CM6	land	THE DISPLAY FLOOR: nx_nishi_editorial ed_td_floor puts the clamp's low end and the fitter's floor one step of the scale above t6 (t6 over t5, read off the derived tokens), in whole pixels and whole centirem rounded up (a floor carried at centipixel precision and printed through the two-decimal rem writer lost a pixel at each end and still read one short: 37 px against 37.7); nx_site_build 3569418a SHIPPED (nx_cms_plugins_gate 35 of 35, nx_ink_derive_gate 39 of 39 rebuilt and shipped 79348f42, stdout IDENTICAL on the gospelofwest workload), the gospelofwest alpha rebuilt in place (DISPLAY index td derived 10900 fitted 9342); the loop fixture now reads 0 short at both doors (phone h1 38 px)
2026-09-26	CM26	land	the file door walks the links a served root walks: nx_cdp_capture's click script returns the anchor's own resolved target beside the document href, and a target the browser could not load (chrome-error) is resolved to <x>.html then <x>/index.html in that order, settled and announced (CD-JOURNEY-FILE-FORM ... landed=1), so a file:// journey completes on a clean-URL sub-page link exactly as https does (edge E6 names the served mapping); the walker carries the board's contract name cd_journey_run; the journey verb's usage line said conf=... while the driver reads a bare fifth argument (usage fixed). nx_cdp_capture_gate 62 of 62 on the laptop (a page linking `sub` beside sub.html completes, the receipt names sub.html, a link to a .html file never triggers the resolution), 43 of 43 on the NAS (8c2bdf08)
2026-09-26	CM14	land	two rulers grown for the loop: nx_ui_tokenref_lib gains the suffixed:<literal>:<i> claim kind (field i then the literal) and design_claims.conf's pages row moves from prefixed:/:1 to suffixed:/:1, because a root-relative mount writes /mount/notes/x and a relative site writes notes/x and both carry notes/ (the prefixed kind read the relative form MISSING on every file-door fixture); nx_ui_audit's craft block now publishes its absent points by name ("missing":"img ...") through ua_pt, the same expression that adds each point, so the refine loop acts on a point BY NAME and an unnamed point refuses. nx_ui_audit_gate 41 of 41 on the NAS (9bdd7b58, promoted with its 6 renamed teeth and fixture pages enumerated as the loss): a mount href and a relative href both land, a bare /notes anchor does not, and exactly the pages row is missing
2026-09-26	CM27	land	the archive and the selection rule (sr_archive_select): variant
2026-09-26	CM17	queue	OWED after this leg: nx_site_refine and its gate shipped to the NAS with a cap and a registered row (the browser leg SKIPs there by name, the in-process teeth run); CM28 sr_knob_search (the factorial search over the token bands and catalogue choices, offline by the referee panel then online through nx_cms_abtest) and CM29 sr_policy_pin (the hash-pinned policy the loop cannot edit); the cms-scene alpha rebuild through its release recipe (nx_plan_run now requires an owned workflow definition: managed-owned-definition-required); the referee's remaining blind spot named by its own counts: the phone-door card straddling the fold abstains NO-CORE on every capture, so the pixel contrast of scroll-revealed text at the fold is UNMEASURED until the capture settles reveals (row 1790373803)
2026-09-26	CM6	measure	the served gospelofwest alpha after the display floor, captured from the laptop and graded by the shipped referee 28dcc838: desktop h1 93 px, phone h1 38 px, hierarchy_step_short 0 at both doors, contrast_fails 0, worst ratio 4.91, textfree_miss 0, no_core 0, verdict OK at both doors; the declared journey (land, Start reading, first theme card) OK at both doors at 2 actions. Two lane facts measured on the way: the alpha is SERVED from the published copies the release recipe's nx_site_publish rows write under nishifamily/alpha/gospelofwest, so a rebuild of sites/gospelofwestcom/alpha changes nothing served until its five pages are published (done by hand as an alpha-stage direct rollout, the recipe's own rows 3 to 11; nx_plan_run answers managed-owned-definition-required for the recipe plane); and the nx_site_build MCP row emitted the pre-floor clamp (1.98rem) twice while the promoted root binary emitted the floor (2.38rem) into a /tmp probe -- the MCP row forks a copy the ship's promote and offc install did not refresh (the third-twin class); the served pages were rebuilt through /api/organ_run of the promoted root and published
2026-09-26	CM17	land	SHIPPED to the NAS: nx_site_refine_gate e76f2914 (56 of 56 in-process on the NAS, the browser leg announced SKIPPED-BY-NAME there, 62 of 62 on the laptop) and nx_site_refine 30fabd5b (first promote, registered, offc twin installed); the referees it forks are all live at their shipped shas: nx_page_pixjudge 28dcc838 (scale-step axis, TEXTFREE-MISS and NO-CORE abstentions), nx_cdp_capture 1bab135a (file-door clean-URL resolution, cd_journey_run), nx_ui_audit dde9e060 (craft missing list, suffixed claims), nx_site_build 3569418a (display floor); confs on the NAS: refine_remedies.conf, refine_referees.conf (conf rows carry candidate paths for the two knowledge trees), pixjudge.conf hierarchy_min_ratio_permil, design_claims.conf pages suffixed:/:1, organ_gate.conf rows for the loop and its lib
2026-09-27	CM29	land	THE POLICY PIN, SHIPPED: the refine loop's evaluator (the four referees) and mutation operator (the remedy table) are pinned by sha256 in knowledge/refine_policy.conf (pin rows per kind, a gain row naming the held-out receipt and its root and best vectors, a seal over every non-comment row); the loop hashes what it resolved and refuses REFUSED-TAMPER:policy-seal (a pin row edited without the pin verb), REFUSED-TAMPER:<kind> (a referee or the remedy table moved, a forged seal included), policy-unreadable or policy-missing-pin:<kind> BEFORE any variant is judged, and stamps referee rows (kind, path, sha256) into every receipt; nx_site_refine pin <policy> gain=<receipt> mints a policy ONLY from a receipt that converged under exactly the live referees with an admitted child beating the root (refusals receipt-not-converged, receipt-no-gain, receipt-referee-mismatch:<kind>), banking the previous policy to <policy>.history; a referee conf without a policy row announces POLICY unpinned and proceeds (opt-in, never a fleet-wide red). Entry points sb_policy_pin and sr_policy_check in nx_site_refine_lib; nx_site_refine PROMOTED d193d81c on the NAS (contentdiff GREEN lost 0 gained 216, offc twin), nx_site_refine_gate SHIPPED 6f8d2fee (93 of 93 on the NAS, 105 of 105 on the laptop with the browser leg: pin from the real converging receipt, a moved pixjudge referee refuses before any variant is judged with no archive written)
2026-09-27	CM29	measure	the laptop lane is PINNED in production: refine_policy.conf minted from the gate's converging receipt over the _build referees (seal c97b5496), declared in the laptop referee conf, and the pinned loop on the fixture read POLICY pinned ok then CONVERGED-RESIDUAL with one admitted move; the NAS referee conf stays UNPINNED and announces it, because a pin needs a converging receipt under the NAS referees and the loop's browser leg runs only where a browser row resolves; the rung's board symbol sr_policy_pin was repointed to the real entry point sb_policy_pin on both the rung and the matrix row rather than minting a name-only alias (the CM21 rule)
2026-09-27	CM28	land	THE KNOB SEARCH, SHIPPED (offline): knowledge/refine_knobs.conf declares token knobs (levels inside a named evidence band) and row knobs (catalogue choices) plus a calibrated evaluation budget; sb_knob_search sizes the full factorial before the first evaluation and refuses REFUSED-BAND:<knob>=<level> outside <band>, REFUSED-KNOB (a token the page lacks) and REFUSED-BUDGET plan+incumbent=N budget=B by name, then emits and judges every plan point with the loop's referees, prints variant rows (levels, id, Pareto admission against the incumbent with the axis, the vector) and archives them beside the loop's variants, choosing the best admitted by the objective or letting the incumbent stand (a completed search); a plan point identical to the incumbent reads INCUMBENT-SAME and is never re-evaluated. THE ONLINE HALF: nx_abtest_bayes_lib computes P(challenger beats incumbent) for two Beta posteriors by the exact closed form in the log domain over a log-factorial table on the bits-up f64 family (bit-identical on every host), and nx_site_refine abtest <experiment> <ledger> tallies the visitor ledger (hit rows per arm) against knowledge/refine_abtest.conf (incumbent, challenger, confidence permil, Beta prior) and journals CHALLENGER-WINS, INCUMBENT-HOLDS or UNDECIDED beside the ledger; nx_abtest_bayes_gate 26 of 26 SHIPPED 516e2621 (registered, offc twin, organ_gate row)
2026-09-27	CM28	measure	four independent methods agree on the probability: exact rationals (Beta(2,1) against the uniform reads one third, Beta(2,1) against Beta(3,1) reads three fifths, delta 0 at 1e-12), the symmetry P(B>A)+P(A>B)=1 at a hundred visits per arm (delta 0), a log-free product evaluation of the same sum (delta 0 at 1e-10) and a 4000-cell grid integration sharing no arithmetic with the lib (delta 0 cells); 30 of 100 against 40 of 100 reads 930 permil, UNDECIDED at 950 and CHALLENGER-WINS at 900, 300 of 1000 against 400 of 1000 reads 1000 permil; an empty ledger reads 500 and stays UNDECIDED under every honest bar, a bar at or below 500 is REFUSED-BAR because an empty ledger would pass it. The e2e search on the laptop: a two-level plan over the display token read plan=2 evaluated=1 admitted=1 (the planted level is the incumbent's own point, the level one referee step up cleared the scale-step axis) and CHOSE it; a budget of one refused REFUSED-BUDGET plan+incumbent=3 budget=1 before any evaluation
2026-09-27	CM28	lesson	a set of a token to the value it already holds must be a byte no-op, or the plan re-evaluates the incumbent under a note and calls it a variant (the first e2e read evaluated=2 same=0 with a REJECTED-NOGAIN twin of the root; fixed at the token setter, not by comparing vectors); the loop resumes from an existing archive by design, so a gate that inherits its last run's archive converges at iteration 0 on the old child and measures the predecessor (77 of 81 before the loop dirs were reset at setup); the magic ratchet named the two fixture literals the new teeth had spelled (50000 and 12840, kind FIXTURE) and both were derived instead (the level read back from the fixture row, the bar move from its declared const); a cap transcribed from a summary is a forged cap (the cms lane cap's nonce differed on disk) and nx_filecopy takes _cap as a sibling like every row
2026-09-27	CM28	queue	the online half is arithmetic, reader and verb proven on fixture ledgers; the LEDGER PRODUCER is not built: a served site must assign arms through nx_cms_abtest (deterministic, no PII; today it has no importer but its gate) and write hit rows on journey-step completion, which is a serving-daemon change through its RACI lane; until it lands the done-rule's declared confidence is met on fixtures, not on live traffic. Also owed: a search over row knobs across tiers (editorial beside showcase), knob levels derived per site from its own tokens rather than the fixture's absolute values
2026-09-27	CM29	queue	owed: a NAS-side converging receipt so the NAS referee conf can declare its pin (needs the browser leg on a host with a browser row, or a served-root capture door); re-pin the laptop policy after any referee rebuild (the loop refuses by name until then, which is the point); the proposal policy evolved under the pin (the rung's second clause) is the knob search over refine_remedies.conf rows themselves, not yet built
2026-09-27	CM10	land	THE CORPUS REFRESH BEAT, SHIPPED: nx_corpus_refresh (entry crf_refresh; decisions in nx_corpus_refresh_lib, prefix crf_) re-mirrors every listing
2026-09-27	CM10	measure	DONE-RULE MET ON THE NAS, two live runs: run 1 (epoch 1790541180) mirrored both Awwwards listings (604733 and 509177 B, zero static stylesheet links on either), derived over 10 sources and read BASELINE (no previous derive); run 2 (epoch 1790541232) re-mirrored them, derived over 12 and read REFRESHED moved=26 -- every moved row names the two dated listing mirrors as its sources (t1 1500 to 1700 n 8 to 10, ratio 1103 to 1064, t0 1359 to 1597, bp1 62000 to 60000, trk -9 to -15, and every n up by two) -- which is the mechanism working as declared and ALSO a defect: the second copies were byte-identical to the first (52 s apart) and doubled every n; the beat now refuses unchanged bytes by sha, the four run-2 rows were retracted in the conf with the reason beside them and the token state restored to run 1's derive (the run-2 mirror files stay in knowledge/fetched as fetched). Awwwards listings are script-rendered (zero /sites/ anchors on the served page, measured 2026-09-27), so no winners rule is declared for award-product; the listing pages' own inline styles are what the derive reads from them
2026-09-27	CM10	lesson	a beat that re-mirrors the same page every run must dedupe by CONTENT, not by url: the url is the same by design (the month's listing) and only the bytes tell whether new evidence arrived; the laptop's _offc fetcher is an older generation that prints SAVED without a PROVENANCE row, so the beat reads the provenance row first and the SAVED line second and hashes the mirror itself when the fetcher did not; a while-read loop that feeds the door client eats the loop's own stdin (one file of three pushed, caught by the push log), so the client runs with stdin closed; the board contract symbol must be a declared function (main is not one): the driver became crf_refresh before the matrix cell could light
2026-09-27	CM10	queue	owed: winners rules for listings that carry static winner links (Siteinspire, Land-book and the Ecommerce Design Awards log, once each is inspected and a prefix declared), a listing whose stylesheets are external (the winners' own pages) exercising the stylesheet and per-page-budget legs live, the beat's first clock-driven run in the journal (the row is in the desired plane; the scheduler folds it at its window init), and the derived-token consumers (the brand emitter derives on demand from the conf rows, so a refreshed corpus reaches every emitted site at its next build without a further step -- to be witnessed on the next alpha build)
2026-09-27	CM10	measure	run 3 (epoch 1790541786, the beat's first MCP invocation under its own cap, on the promoted dedupe build a003223d): both listings had CHANGED since run 1 (604980 against 604733 B and 509037 against 509177 B, new shas), so same=0, both admitted as new evidence, derive over 12 sources, REFRESHED moved=26 naming the two dated mirrors; the SAME-BYTES refusal is therefore proven on the gate's fixtures and NOT yet witnessed live (an unchanged listing between two beats is the case it exists for, and Awwwards moved between every pair of runs today); the four run-2 rows stay retracted, so the corpus holds 12 sources: 8 hand-mirrored plus the two dated listing copies of run 1 and run 3
2026-09-27	CM23	land	design modes as DATA: knowledge/design_modes.conf (mode composed/organic, 8 facets nav framed summary band cta footer related disclosure, default home/notes) read by nx_design_modes_lib (dmo_mode_load refuses no-such-mode / facet-undeclared / facet-not-0-or-1 by name; dmo_mode_for = blueprint mode
2026-09-27	CM23	measure	byte-identity control: the composed defaults rebuild the incumbent alpha pages IDENTICAL on all 3 pages (stage=alpha; the only earlier DIFFERS was the noindex meta the loop adds at stage=alpha, not the modes); gate 22/22 laptop and NAS; the modes conf lands in knowledge/ on both hosts and in buildroot/knowledge
2026-09-27	CM23	lesson	a conf read by cwd alone is a gate-host dependency: nx_ink_derive_gate runs the emitter inside fixture roots where design_modes.conf is absent and read 34/39 until the conf was resolved on a ladder derived from the executable (cwd, exe/../knowledge, exe/knowledge, exe/../../knowledge) -- the same law as the trust-store ladder (dataio DI20), arriving through the emitter
2026-09-27	CM23	queue	the disclosure facet is declared 0 on both modes and OWED (no emitted page carries an expandable section yet); no blueprint opts into organic yet, so the alphas are unchanged by construction; an organic page goes under the operator's eye and the pixel referee before any client site declares mode
2026-09-27	CM25	measure	the served-page judge on the LIVE alpha https://nishifamily.com/alpha/gospelofwest/ (journey gospelofwest, doors desktop+phone, laptop browser host; referees nx_cdp_capture, nx_page_pixjudge, nx_ui_audit over the wire, nx_page_verify): signals measured 9/10 -- pixel.contrast 1000, hierarchy 1000, overlap 1000, overimage 1000, bytes.craft 933, hygiene 1000, journey.complete 1000, friction 1000, a11y 1000; tokenref UNMEASURED by name (a served page carries no token declaration). awwwards 992 over covered weight 70/100 (design 986, usability 1000; creativity and content UNMEASURED no-map-row), webby 995 over 4/7 axes, nngroup 983 over 2/10, baymard 966 over 1/6. Receipt knowledge/status/rubric_alpha_gospelofwest_2026-09-27.receipt
2026-09-27	CM25	measure	the SAME judge on the live rival https://www.nike.com/ (journey rival-nike: one land step whose expectation is the page's own name; a rival capture conf with a 45 s settle because the estate's 15 s timed out on the cold first visit, CD-SETTLE-TIMEOUT, and the warm capture settled in 3.1 s): 9/10 measured -- pixel.contrast 894 (one failing box per door), hierarchy 1000 (the desktop door carries no h1 and reads hierarchy UNMEASURED, so it is out of that denominator), overlap 1000, overimage 789 (four text boxes over photographs on desktop), bytes.craft 333, hygiene 750, journey 1000/1000, a11y 250 (one image without alt, seven heading-level skips, no main landmark). awwwards 795 over 70/100 (design 803, usability 786), webby 826, nngroup 833, baymard 666. Receipt knowledge/status/rubric_rival_nike_live_2026-09-27.receipt. The mirrored bytes alone (rf_20699706.raw through nx_ui_audit file) measure 2/10 signals and read awwwards 400 over 40/100 -- a mirror without a browser names eight signals unmeasured, the live judge names one (receipts rubric_mirror_nike_home_2026-09-21 and rubric_mirror_awwwards_sotd_listing_2026-09-21)
2026-09-27	CM25	lesson	three laws from the referee: (1) a rubric axis is the MEAN of the estate signals that measure it and prints UNMEASURED by name when none can, never 0 -- creativity, innovation, content and overall experience have no estate referee and the map says so; a platform number carries its covered weight beside it (992 over 70/100 is not 992 over 100). (2) every pixel signal carries its own denominator, summed only over the doors that measured it: a door reading hierarchy UNMEASURED contributes neither its violations nor its text boxes, or the share is a distance measured from a region that does not exist (the 2026-09-06 flood law in referee form; the fixture's phone door is exactly that case, 913 not 937). (3) a served-page judge must not end at the first refused stage: the first cut returned on capture:refused and the bytes ruler and verifier never ran, so a rival whose cold load outlived the settle budget measured 0/10 -- now every stage runs, each banks as it finishes, and why names the refused stages comma-joined; the scorer names the unmeasured, an early return manufactures them
2026-09-27	CM25	lesson	the refine gate went 104/105 the moment CM23 rebuilt nx_site_build: the laptop policy pin (CM29) named the OLD evaluator sha and the loop refused REFUSED-TAMPER:build exactly as designed; the re-pin came from the gate's own fresh converging receipt under the new referees (PIN ok seal=2e7b6e42, the previous policy banked to refine_policy.conf.history) and the gate read 105/105 -- a pin that outlives its referee is the alarm, not the defect. The rival landing journey first read INCOMPLETE because its step label was prose and the driver's land expectation IS the label text (the needle in the visible text); the page's own name is the honest expectation, and the journeys conf carries it as data
2026-09-27	CM25	land	nx_page_rubric_lib (rub_signals_from_dir, rub_axis_score, rub_rubric_score: the ten signals, an axis = the mean of its measured mapped signals or UNMEASURED by name, platform roll-ups with covered_weight and axes_measured) + knowledge/design_rubric_map.conf (map
2026-09-27	CM25	queue	(1) a fixture-referee tooth for sr_judge_url in the rubric gate (today proven live on two URLs, not in a gate); (2) the NAS has no browser leg, so the served-page judge there reads capture:refused PARTIAL (audit + verify only) -- a browser host for the NAS is the standing gap behind CM13, CM6 and this rung; (3) creativity, innovation, content and overall experience stay UNMEASURED until a referee exists: CM18's pairwise referee with a human-judged set is the road to the first of them; (4) the rival journey is one land step; a rival's 1-2-3 path needs its own step rows before journey.friction means the same thing on both sides; (5) tokenref reads UNMEASURED on every served page: the token declaration could ride the page (a meta row the emitter already knows) so the served-page judge can measure it
2026-09-27	CM22	lesson	the NAS copies of nx_swcompare_matrix (+93 B, a sibling's pm plan surface), nx_swcompare_evidence (390 NAS-only lines: a TjBuffer stamp writer and a sealed gate binding) and four of their imports were AHEAD of the laptop and were pulled and adopted before a byte was edited; the evidence organ does not build on the laptop at all (stale sibling libs), so its edit was applied to the NAS generation by unique-line insertion and proven by the NAS build (cb4a37e7); a python heredoc through the harness mangles backslash-n anchors -- edit scripts live in files; the matrix generator SEGFAULTS on a fixture root where the estate root is absent (nx_catalog_lib cl_open_estate -> cl_estate_path with a -1 path pointer, laptop-only: the NAS run of the same fixture emits 92 KB), so the dossier gate's end-to-end leg is need-guarded and proves itself where the generator resolves beside it
2026-09-27	CM22	queue	(1) the per-gate verdict record (nx_swcompare_evidence writing knowledge/status/evgates_<dom>.tsv, evg_path/evg_resolve in nx_evprofile_lib) is BUILT on the NAS (candidate cb4a37e7) and REFUSED by its release scope gate (COMPARE-RELEASE-REFUSED reason=changed-closure: knowledge/gates/nx_swcompare_evidence.release.scope pins every closure file with a decision and an explicit review reference) -- needs-intervention for the compare referee's release owner: re-adjudicate the changed closure with a review record, then ship; until then every dossier reads gate UNRECORDED and evgates UNSTAMPED by name, which is the honest state; (2) the laptop-only generator crash on a fixture root without the estate root deserves a null guard in nx_catalog_lib; (3) the sota generator renders no dossiers yet (matrix-class boards only); (4) a symbol two rungs share links its cell to the first rung only
2026-09-28	CM16	measure	the section layout solver on the estate's fixture tokens (14 px body, 74 ch measure, 1200 px container, spacing s1..s7 4..64 px): two cards solve to two columns (objective 125: rows 125, every other term 0), three steps to three columns (83: measure 0, rows 83), four cards to four columns at 1200 px (110: measure 48 + rows 62) and to two columns at a 600 px door (four columns would hold nine characters a line, three would orphan the fourth); a planted orphan (three steps forced into two columns) scores 1316 (orphan 1000 + empties 150 + rows 166) against the solved 83 on the printed objective; the pixel referee over browser-visible twins of both builds (desktop door) counts 0 defects on each -- it cannot see an orphan, and it never rates the solved layout worse. Gate nx_nishi_layout_gate 31/31 on the laptop (the pixel leg runs where cdp_capture.conf declares a browser-visible fixture root; elsewhere it is announced absent, never a precondition)
2026-09-28	CM16	lesson	the objective is built from shares the tokens and the brief already carry, so only five weights and three bands had to be typed and each is a conf row with its calibration; the line-length band is derived (a third of the measure to the measure), the cell width from the container and the gap token, chars per line by the emitter's own ch convention -- and the gate re-derives the argmin by brute force over the whole band (a second method class for the solver's one claim) rather than asserting the column count a seat predicted: the first draft asserted two columns for four cards and the solver was right to choose four at 1200 px on those tokens
2026-09-28	CM22	land	nx_swcompare_matrix SHIPPED 603de9cd through nx_organ_ship (build, 7 of 7 declared gates proven on fresh closure banks including nx_compare_dossier_gate 24/24, content ruler RED adjudicated: 8 lost runs = 2 literal splits of the capability row and its cell close by the new id and dossier link, 6 supersessions by sibling lanes, loss token 7c7cff3a; behaviour probe on cms html DIFFERS STRONG 514186 to 705311 B, first diff at the dossier stylesheet rule, reviewed under NISHI-SHIP-BEHAVIOR-REVIEW/1 record b05d4039; promoted, offc twin installed, registered); nx_compare_dossier_gate SHIPPED d43f180e on its own lane (first promote, 24/24, registered) and declared on cms.gates; organ_gate.conf rows nx_compare_dossier_lib and nx_swcompare_matrix -> nx_compare_dossier_gate. The per-gate verdict record (nx_swcompare_evidence writing knowledge/status/evgates_<dom>.tsv beside its stamp) is BUILT-UNPROMOTED: the NAS build cb4a37e7 is refused by nx_swcompare_evidence_scope_gate as changed-closure until the release owner re-adjudicates knowledge/gates/nx_swcompare_evidence.release.scope, so every dossier reads its gate verdicts UNSTAMPED until that ships (needs-intervention, release owner; the seat does not forge the review)
2026-09-28	CM16	land	nx_site_build SHIPPED 74132c46 through nx_organ_ship (build; four declared gates proven on fresh closure banks: nx_cms_plugins_gate 35/35, nx_ink_derive_gate 39/39, nx_design_modes_gate 22/22, nx_nishi_layout_gate 30/30 on the NAS where no browser resolves; content ruler RED adjudicated: the 3 lost runs are the three stylesheet rules the solver replaces, each superseded by its custom-property twin, loss token 2a7810b5; the bare probe was refused as incomplete once a loss is approved, so the representative probe drives both binaries on the layout gate's showcase fixture site, DIFFERS STRONG 976 to 1773 bytes at the LAYOUT conf line, reviewed under NISHI-SHIP-BEHAVIOR-REVIEW/1 record a706f4eb; promoted, offc twin installed, registered); nx_nishi_layout_gate SHIPPED 08776c47 on its own lane (first promote, 30/30, registered; its first ship read 26/30 because the emitter beside it was still the pre-solver build -- the emitter ships first, the gate proves against the fresh build); knowledge/layout_solve.conf on the NAS; organ_gate.conf rows nx_nishi_layout_lib and nx_site_build -> nx_nishi_layout_gate; import census refreshed so the finish state reads from the graph. Scope stated: only showcase-tier blueprints (tokens row) are solved; the kit-tier client sites and the gospelofwest alpha are byte-untouched; a showcase client re-emit carries the solved grid through the release lane, never by this seat
2026-09-28	CM22	lesson	the first live publish rendered all 30 dossiers NO-GATE-DECLARED for organs whose gates the ship loop proves every day: buildroot/knowledge/organ_gate.conf is a 354-byte stub carrying one sibling row, the regen runs from buildroot, and a CWD-first ladder returned the stub and never looked at the 44 KB declaration file at the estate root -- the two-tree trap, arriving through a file the compare emitters had never read before. A declaration is a fact and two lists of facts add, so the dossier lib now reads the UNION of the CWD copy and the root copy (rows deduplicated byte for byte, identical copies counted as one) and prints which copy answered with how many rows in the section itself; the evgates verdict record keeps its stamp's ambiguity rule because a verdict is a measurement and two different measurements refuse. Gate teeth: a stub beside the publisher cannot hide the root, no row twice, a copy without a trailing newline, comments are not declarations -- every expected count derived from the fixture texts after the NAS magic ratchet refused the first draft's hand-counted lengths (the estate's own law, paid for again)
2026-09-28	CM22	queue	the organ_gate.conf census (nx_shelltool over buildroot/runtime, 26959 files, coverage complete): the readers are nx_organ_ship_lib (OSL_CONF_DEFAULT, read from the ship's working directory, the estate root), nx_rungclose (RP_GATE_CONF, a clock-stage organ whose working directory decides which copy it reads) and the dossier lib (now the union of both roots). The 354-byte buildroot/knowledge/organ_gate.conf stub is a sibling's declaration (nx_hostctl_supervise_gate) and must not be deleted; any reader that runs from buildroot and reads the CWD copy alone inherits the same blindness -- for nx_rungclose's owner to adjudicate (read both roots as the dossier lib does, or run from the estate root). The evidence organ's per-gate record stays queued behind its release scope (release owner)
2026-09-28	CM22	measure	second ship: nx_swcompare_matrix SHIPPED a892f53b (content ruler GREEN, one-way gain of 11 runs, 7 of 7 declared gates including nx_compare_dossier_gate rebuilt 31/31 in the ship; behaviour on cms html 715960 to 717263 bytes reviewed under record 6039d4c9), nx_compare_dossier_gate SHIPPED e6f0ce71 on its own lane (31/31); scoped regen published cms index.html 720283 bytes (prev 709183), api.json 117188, matrix generator fork FRESH at a892f53b, nx_page_verify GREEN with 0 a11y issues. The dossier section's own provenance line: organ_gate.conf read from this working directory (1 rows) and from the estate root (154 rows), rows in both counted once 0, declarations read 155. Foot partition: dossiers=30 cells_joined=30 no_matrix_row=0, gates_recorded=0 gates_unrecorded=14 no_gate_declared=16 (sum 30), evgates=UNSTAMPED -- the 14 name their declared gate and wait for the evidence organ's per-gate record; the 16 are organs with no organ_gate.conf row (some because the declaration names the program while the matrix row names its lib, nx_cms_snapshot against nx_cms_admin's nx_cms_snapshot_gate, for their owners). nx_compare_rank cms: 29 of 30 done, open CM18, finish states 4 (REGISTERED-DARK: CM10, CM13, CM24, CM26, all laptop-executed browser or fetch organs the NAS actlog window cannot see)
2026-09-28	CM18	measure	first real run of the pairwise referee (laptop, the browser host, the two variants the refine loop archived under nx_refine_gate/loop/variants): `nx_ui_judge pairwise <cands> -` on the awwwards platform -- candidates=2 scored=2 pairs=1 referee_judged=1 ties=0; RANK 1 v24fd5f76 elo=1516 score=992, RANK 2 v5ac366ee elo=1484 score=985 (Elo start 1500, K 32, the estate's nx_dr_elo, bit-exact); HUMAN set=- rows=0 comparable=0 agreement UNMEASURED; VLM oracle=none UNMEASURED (no image route in nx_extllm); PAIRWISE-VERDICT floor=UNCALIBRATED verdict=UNCALIBRATED-FLOOR. The rung's done-rule (agreement above a pre-declared floor on this board) is NOT met and cannot be met by a seat: it needs the operator's judged page pairs (pair
2026-09-28	CM18	lesson	extend, never twin: the referee is a lib (nx_page_pairwise_lib, prefix pj_) composed from three incumbents found by nx_capsearch before a line was written -- the one rubric referee (nx_page_rubric_lib, which grew a SILENT per-platform score by delegating the printed platform row and the compared number to ONE function, rub_rubric_score_into, so the number on the receipt and the number in the comparison cannot drift), the estate's deterministic integer Elo (nx_dr_elo, start 1500, K 32, bit-exact) and the gate base class's Wilson bounds (gv_wilson_permil at z95) -- and it rides nx_ui_judge as a verb, the incumbent the matrix row names, whose no-argument path stays byte for byte (the gate proves it). Three shapes worth naming: an unmeasured candidate must rank LAST, never on its untouched start rating above a rated loser (the first draft would have ranked it second); the floor is a conf row whose calibration is DERIVED by the referee itself (the inter-judge agreement printed from any judged set with two judges) because no seat may type the bar a referee is held to; and a mechanical rewrite of a helper's call sites rewrote the helper's own body into a self-call -- the crash guard named the unbounded recursion and nx_addr2line named the line in one call, a defect the estate's tooling turned into a two-minute fix
2026-09-28	CM18	land	the pairwise-preference referee SHIPPED on the NAS: nx_ui_judge 55105965 through nx_organ_ship (the incumbent competency judge EXTENDED with the `pairwise` verb, never twinned; content ruler GREEN; the bare probe compares the no-argument path, byte-identical; gate proven 41/41 from the promoted nx_page_pairwise_gate once its sibling ladder preferred the staged candidate), nx_page_pairwise_gate SHIPPED 588446bd (first promote, registered, declared on cms.gates), runtime/nx_page_pairwise_lib.nx (pj_pairwise, the symbol the matrix row watches), nx_page_rubric_lib grew rub_platform_score (the silent per-platform score by the one arithmetic; its consumers nx_page_rubric and nx_site_refine read BEHIND until their lanes re-ship, behaviour unchanged), knowledge/pairwise_referee.conf on the NAS (platform awwwards; floor UNCALIBRATED with its calibration rule), organ_gate.conf rows nx_page_pairwise_lib, nx_ui_judge and nx_page_rubric_lib -> nx_page_pairwise_gate. Status: BUILT and PROVEN, the matrix cell flips to PRESENT on the next publish, the rung stays OPEN by its done-rule -- agreement with a human-judged set above a pre-declared floor is on this board only when the operator's judged pairs exist (two judges on the same pairs calibrate the floor); the VLM oracle arm reads UNMEASURED until nx_extllm carries an image route
2026-09-28	CM22	lesson	a gate promoted at the serving root proves the INCUMBENT: the judge's ship ran its declared gate from the root (the closure bank said the root copy IS the fresh build, so no rebuild) and the gate's sibling ladder (<selfdir>/<name>.sov.elf, then <name>.elf) found the LIVE judge beside it, not the candidate the ship had staged as <name>.sov.elf.new -- 39/41 with exactly the two verb teeth failing on a judge that did not carry the verb yet. The dossier and layout gates carried the same ladder and had passed by luck (their subjects' ships happened to rebuild the gate into buildroot/_build, where the fresh subject sits). All three ladders now try the staged candidate (<name>.sov.elf.new beside the gate and at the root above buildroot/_build) before any promoted twin, re-shipped: nx_page_pairwise_gate 588446bd, nx_compare_dossier_gate 02798af7 (31/31), nx_nishi_layout_gate 9e6bdad0 (30/30). The general law: an end-to-end gate that forks its subject must prefer the artifact under test over the artifact in service, or a green proves the wrong binary
2026-09-28	CM18	queue	two arms measured for feasibility on 2026-09-28 and declared as rungs rather than half-built: (1) the VLM oracle arm needs an image route on the ONE egress door (nx_extllm_call): a capture PNG must be a DECLARED public file (a release row with its sha256, the C-by-default classification refuses an undeclared capture), the provider's terms row must admit the JUDGE use, and the request gains a base64 image content part -- every conjunct of the egress law, so it is an egress-board rung, not a patch; until then the oracle set is any pairs file in the referee's grammar written by any outside judge. (2) A NAS capture door for the browser-dark rungs (CM13, CM24, CM26, and the served-page judge on a beat): the sovereign render core (nx_browser_render: br_layout to br_shot_png over the Liberation faces the NAS already holds) can paint a page to PNG without Chrome, but the pixel referee also reads the .boxes text-box record the DevTools driver emits, so the door is the render core emitting the capture grammar (png + boxes + paper) behind a renderer=sovereign row in cdp_capture.conf with its parity to Chrome measured by nx_parity_judge_gate before any verdict rides on it -- a browser-board rung. Neither blocks the human-judged set: the judging surface runs on the laptop, the browser host
2026-09-28	CM18	land	THE JUDGING SURFACE (the input only a human can give, made into minutes of clicking): runtime/nx_pairjudge_lib.nx (the pure core: every unordered pair of the referee's candidates in the referee's own order, what a judge already ruled on in either orientation, a no-script page with one form per open pair showing both captures side by side at a chosen door, the routes GET / and GET /img/<k>/<door>.png and POST /verdict with every refusal named: 403 csrf, 400 pair or verdict, 409 already judged, 404 elsewhere) + runtime/nx_pairjudge_serve.nx (a thin loopback daemon: bind 127.0.0.1:<port>, read, route, write, exit after a request budget) + nx_pairjudge_gate 30/30 on the laptop (real loopback end to end on a port proven free in a window; the budget exits the daemon with 0, no orphan). Every verdict lands through the referee lib's ONE row writer pj_pair_row (a pipe becomes a slash, a control byte a space), so the surface and the referee's reader share one grammar by construction. ALPHA LIVE on the laptop, the browser host: nx_pairjudge_serve 8098 knowledge/judged/cms.cands knowledge/judged/cms.pairs elder 0 desktop -> http://127.0.0.1:8098/ with four captured pages (alpha-gospelofwest, rival-nike-live, the two refine variants) = six pairs; a second judge runs the same command with another judge name; then nx_ui_judge pairwise knowledge/judged/cms.cands knowledge/judged/cms.pairs prints the agreement and the inter-judge floor candidate. NAS: nx_pairjudge_gate promoted f4de3003 DIRECTLY through /api/promote with the one lost run named (a renamed tooth) because the ship loop's probe cannot qualify against an incumbent that crashes by construction (the first NAS build's fixed fixture port 8097 was taken there and the gate walked into a null row; fixed: a free port found in a window, the wire teeth skipped by name when none is free, no row dereferenced when none landed); nx_ui_judge re-shipped 09eb7c3d against the new writer; nx_pairjudge_serve BUILT and STAGED 2290d307 on the NAS, declared `daemon` in organ_kind.conf and deliberately NOT deployed there -- it serves loopback on the host that holds the captures and the browser, so a NAS instance would have nothing to show; the stage slot is named here so the drift board's new row has an owner and a reason
2026-09-28	CM18	measure	THE OPERATOR'S FIRST JUDGED SET, 2026-09-28: six pairs over four candidates (the gospelofwest alpha, the live nike.com home, two refine-loop variants), judge elder on the alpha surface at 127.0.0.1:8098; the referee (nx_ui_judge pairwise, awwwards platform score, Elo start 1500 K 32) agrees on 2 of 6 (333 permil, Wilson 97 to 700) and ranks alpha 1531, refine-v24fd5f76 1530, refine-v5ac366ee 1485, nike 1454, where the operator ranks nike first (3 wins of 3), refine-v24fd5f76 over refine-v5ac366ee, and the alpha tied with both variants; floor UNCALIBRATED (one judge, inter-judge UNMEASURED); verdict UNCALIBRATED-FLOOR, exit 4
2026-09-28	CM18	lesson	the referee inverts the operator's top and bottom because its platform score is the bytes rulers' hygiene and craft counts (the nike home carries 53 scripts and 21 external assets and reads craft 6 of 15, so it ranks last) while the operator's eye judges photography, composition and confidence; a 2-of-6 agreement is the measurement CM18 exists to publish and it says the estate's judge does not yet see what the operator sees: the pixel referee's beauty axes and licensed imagery (CM6, CM7) are the missing signal, never a reason to move the floor; the operator's own verdict on the surface was that it is a good start and not the field's state of the art, which this milestone declares as data
2026-09-28	CM30	measure	the bar for the judging surface read from this month: two arXiv listings (30 and 13 entries, digested through nx_arxiv_digest) and the Design Arena home whose served bytes carry its methodology strings (four random models per session, two randomly paired models presented anonymously side by side, a choice is one vote, winner and loser brackets with a tie breaker to a final ranking, Bradley-Terry denoted Elo from 1200 with a standard-error margin, win rate and placement); the LMArena WebDev arena is Code Arena now and its methodology post is a script shell (URL-witnessed, not cited); eleven rungs CM30 to CM40 under M11 with their roles toward T-BIC-PAIRJUDGE and T-FRONT-PAIRJUDGE
2026-10-01	CM5	measure	October attestation of the board's eight bars (the NAS clock crossed into October and nx_compare_rank refused BAR-UNATTESTED): nx_frontier_scan attest cms read the five arXiv listings and appended five barscan rows (fetches 136, unattestable 3: the Awwwards listing, the unite.ai roundup and the Awwwards evaluation page carry no dated source row); the seat re-fetched and read those three (sha 8a0318d6, a831a117, 85ae70e6) and adjudicated the two listings the attester flagged NEWER: SB-WEBCRAFT moves to 2026-10 (Colonia Zacamil, Oct 01, 2026 Site of the Day), SB-RUBRICS to 2026-10 (living page unchanged), SB-UIOPT to 2026-03 (fatigue-aware VR interface layout optimisation 2603.26031, missed by the September reading), SB-IA stays 2026-08 (every September entry a homonym, named on the barscan), SB-BUILDERS attested at 2026-09 (the roundup is still September's); the attester reported CLASSES UNDECLARED for cms, so a cls row for the board is the mechanism that stops a homonym setting its bar month next time
2026-10-01	CM5	lesson	a bar month is a measurement with a clock on it: the ruler derives the current month from the NAS clock in UTC, so a board attested on the last day of a month is unattested the next morning, and the attester cannot judge subject without the board's class rows (cls
2026-10-01	CM30	measure	nx_compare_rank cms after the October attestation: BARFRESH bars=8 fresh=2 attested=6 stale=0 unattested=0 verdict=FRESH; LADDER targets=9 rungs=41 placed=41 verdict=LADDERED; rungs=41 done=34 open=7 finish=4 (the four REGISTERED-DARK finish states are the laptop browser and capture organs CM10 CM13 CM24 CM26); critical path on target version 10.0 (M11): #1 CM34 pj_kripp_alpha (DEP-PULL as the prerequisite of CM36), #2 CM35 pjj_live_route, #3 CM38 pj_loo_stability, #4 CM40 pjj_tournament, #5 CM37 pj_axis_verdicts, #6 CM39 pj_judge_profile, #7 CM36 pj_cojudge_anchor; CM30 to CM33 read done because their symbols are declared in buildroot/runtime/nx_pairjudge_lib.nx (a758912a), proven 53 of 53 on the laptop and NOT yet promoted on the NAS (the ship is queued behind build admission at load1 28, runqueue 27 of 8), so their cells flip on this regen while the gate verdict record stays the laptop's until the NAS promote lands; the five open rungs the ranker prices first are being built in parallel by a five-agent workflow (operator: fan out on workflows), each a new lib and gate, the integration into the referee receipt and the surface owed after
2026-10-01	CM30	land	CM30 to CM33 are proven on the NAS: the referee lib gained pj_rows_ratings, the one Elo ruler over judged rows (source 3adc3c8d, 31491 B), the surface's pjj_ratings is a one-line delegation to it (source bb9a51bd, 37846 B), nx_pairjudge_gate built on the NAS as 39c12c6b (217320 B), promoted, and /api/gate_run on the promoted binary reads GREEN exit 0 with 54 of 54 teeth against the staged daemon cefa7556; the daemon itself stays staged and undeployed
2026-10-01	CM30	lesson	a gate run once green is not a deterministic gate: nx_pairjudge_gate read 53 of 53 on single runs and GREEN on only 11 of 40 consecutive runs, because three teeth asserted facts that depended on the random session token (a two-character fixture id matching inside the hex token in the pair section, in-process and over the wire, and a count of left-side markers taken over every judge's meta rows while the other judge's coin was random); the ship loop's PROVE stage caught it as 51 of 53 once build admission cleared; fixed at the root with a reader that ignores an occurrence lying wholly inside the token, a count over one judge's rows, and a planted-leak control, then 40 of 40 runs green; run a gate twenty times before shipping it
2026-10-01	CM30	queue	the ship loop cannot observe a coupled gate and daemon redesign: the incumbent gate and the candidate resolve one daemon through the same ladder, the surface changed from every pair on one page to one blind pair, so the incumbent reads RED against the new daemon and the behaviour stage refuses complete-successful-representative-probe-required before any observation can be reviewed; this gate went through /api/promote with the exact sha after the loop's own BUILD and PROVE were green and its 31 lost runs were enumerated (every one superseded by the blind design: the old style block, the A and B section and buttons, hand-typed fixture rows and paths, renamed teeth, the old header and note); owed on the loop: an observation path for a superseded incumbent that fails for a named subject change
2026-10-01	CM30	measure	triangulation of SB-PAIRJUDGE read from the mirrored listing's own comment fields (cmp_cms_arxiv_202609_arena_human_preference.xml, 80638 B): peer-reviewed witnesses K-Sort Eval 2602.09411 (ICLR 2026), Transitivity Meets Cyclicity 2605.17342 (ICML 2026), Personalized Benchmarking 2604.18943 (Findings of ACL 2026) and DR-Arena 2601.10504 (ACL 2026); September preprints from two separate groups ANCHOR 2609.19599 and WebCraftBench 2609.15387; Design Arena's served methodology as a vendor claim; three independent witness classes with a peer-reviewed paper among them and the month's listing read

On these two registers. Rows are declared in the domain's plan file and carry the debt id, which is the join key back to the sovereign debt plane — that plane, not this page, is the authority on state. Reconciling them automatically (the regen reading the plane and refreshing these rows) is a named, owed rung; until it lands, treat an id here as a pointer to look up, not a status to trust.

Honest verdict. The coverage above records source-symbol presence. Stronger comparisons are authored claims, not measured runtime superiority or accepted outcomes. A watch contract names an expected source symbol; finding it changes source discovery only. Qualification, deployment and runtime acceptance require their own evidence.

Person · product · place — not yet measured for this domain

Every compare carries this layer. Declare knowledge/compare/cms.ppp (rows surface|nishi or c1..c4|label|url|connect naming OUR live surface and each rival's front door), run nx_ppp_probe domain cms, and this section fills itself on the next beat: the same ruler on both sides — privacy and CX (third-party hosts, tracker classes, cookies, security headers), design and longevity (design hygiene, computed WCAG contrast, render-blocking resources, unsized media, script weight, theme and motion queries), findability (landmarks, skip link, on-site search, breadcrumb, headings, internal links).

The field — discovered, not chosen

Rows written by nx_field_discover from cms.seeds: the industry's own lists (Wikipedia wikitext, GitHub topics, awesome lists) read mechanically, every candidate counted across seeds. The matrix columns above are a SEAT'S pick; this band is the population they were picked from, and the stats line measures one against the other. A rival here is a lead, never a verdict — it earns a column when its capabilities are read and pinned.

Discovery receipt. One row per counter the run wrote, in the receipt's own order; the rival rows follow in their own table.

Field discovery, counted
Class	Count	Note
seeds	4
scanned	4
fetched	4
reused	0
failed	0
named	4
candidates	4
mentions	4
capped	0

Rank	Rival	Seeds	Mentions	First seed	Kind	Link
1	WordPress	1	1	col0-wordpress	named	https://developer.wordpress.org/rest-api/
2	Wix	1	1	col1-wix	named	https://dev.wix.com/docs
3	Webflow	1	1	col2-webflow	named	https://developers.webflow.com/data/docs/getting-started-apps
4	Drupal	1	1	col3-drupal	named	https://www.drupal.org/docs/administering-a-drupal-site/multilingual-guide

Exclusive and shared capabilities

One capability is graded on one board; every other board references it. The fleet duplicate census (nx_rivalcensus dupes, on the fleetdupes clock row) reads every board's matrix and names each capability graded as a row on two or more boards, with a home derived from evidence: rung when exactly one board's plan contracts the symbol (that board is starred), gates when no plan contracts it and exactly one board's executed gate map runs the organ's gate (that board is plussed), ambiguous when several boards carry the deciding evidence (a decision for their seats), and none when no evidence decides it, with the reason beside it: unmapped (no board of the duplicate declares a gates map; declare the maps) or unproven (a map exists and runs none of the organ's gates; gate the organ on its proving board, or declare it a shared library by policy). A home elsewhere means this board should carry an @xrow reference to the home board's row rather than a graded row of its own; a home here means the other boards should. The table is this board's slice of one estate-wide census: the shape is shared, the rows are the board's.

This board grades 68 capabilities: 64 exclusive to it and 4 shared with other boards. Shared rows on this page: 4 = home here 1 + home elsewhere 0 + ambiguous 0 + undecided 3 + refused 0 (partition 4); undecided = unmapped 0 + unproven 3 + unevidenced 0; homonyms 0; since the previous census: new 0, resolved 0. Census FRESH age_s=6333.

Capability	Organ	Boards grading it	Home	Rule	Evidence	Standing on this board
sb_build_site	`nx_site_build.nx`	cms+, sharehub	cms	gates	gates	home here
ma_do_promote_content	`nx_mgmt_api.nx`	cms, deploy	—	none	unproven	undecided
cmt_score	`nx_cms_comments.nx`	cms, sharehub	—	none	unproven	undecided
ma_do_deploy	`nx_mgmt_api.nx`	cms, containers	—	none	unproven	undecided

Gaps from the record — what the estate does that no board carries

The record census (nx_goalmap record) reads the invoked-tool population and every plan queue row and files each organ or directive that NO matrix, plan or gates row names. A row here is a callout the boards missed: adjudicate it onto a board or declare it infrastructure. Census state STALE (age 1395438 s), sources read 4 of 7 declared — a BLIND census is a FLOOR: unread sources can only add rows.

kind	name	board	source	evidence
directive	`nx_site_build_prev`	cms	plan-queue	rung=CM9
directive	`nx_ws_client`	cms	plan-queue	rung=CM13
directive	`nx_cms_admin`	cms	plan-queue	rung=CM1

rows shown 3|this board's directives 3|estate-wide un-boarded organs 452 (listed in full on /compare/ecosystem)|census rows 639|malformed 0 (counted, never rendered)

References

Beyond a link list. Every reference below resolves twice — the publisher's copy and, where banked, the estate's own non-rottable library mirror with a content pin — and carries its evidence class plus the exact claim on this page it grounds. Keyed marks like [key] in the matrix notes jump here. A dash means honestly absent, never assumed.

[wp-rest-api] WordPress. REST API Handbook -- the WordPress REST API reference (routes, endpoints, authentication, field filtering). Vendor documentation. publisher · read in our library knowledge/fetched/cmp_cms_wp_restapi.html · pin heb2d3561e17ea11d2ec5b254f419742722f293d637abdcaf0f1abf239ad4d4a5 · accessed 2026-08-18 · vendor-docGrounds: The "REST plus GraphQL-style projection content API" row (the note: WP REST API is the reference) and the WordPress column's Best code there -- our cms_api_render_obj is graded against this documented API.
[wp-plugins] WordPress.org. WordPress Plugin Directory -- browse over 69,000 free plugins (page text 2026-08-18); the largest directory of free and open source WordPress plugins. publisher · read in our library knowledge/fetched/cmp_cms_wp_plugins.html · pin h0d9e772ddceb25404eaed2a4e5ddbdd4a346bb564563556f41067f22df252e6d · accessed 2026-08-18 · vendor-docGrounds: The "Plugin and extension marketplace" row (_ABSENT_ nx_cms_plugins; the note says WP 60k plugins -- the live directory page reads over 69,000): the honest structural gap and deliberate sovereignty tradeoff is measured against this directory, and the WordPress Best code on "Comments", "Forms handling" (plugins) and "SEO" (Yoast-class) leans on the same ecosystem.
[wix-dev-docs] Wix. Wix Developer Documentation, APIs and Guidelines (Wix Docs). Vendor documentation. publisher · read in our library knowledge/fetched/cmp_cms_wix_dev.html · pin h094a345d258b0a27562f0576ff12e4df2a16b5a932824d8731035c7abb61d423 · accessed 2026-08-18 · vendor-docGrounds: The Wix column codes (Yes on the content store, admin API, scheduling and forms rows; Part on tokens, revisions, API projection and multilingual): Wix is the proprietary-cloud site builder those codes describe, and the foil for the "Structured content store, portable text bundle, no SQL" row note (rivals sit on MySQL or a proprietary cloud).
[webflow-dev-docs] Webflow. Getting started with Webflow Apps -- Webflow Developer Documentation (Data API, Designer API, apps). Vendor documentation. publisher · read in our library knowledge/fetched/cmp_cms_webflow_dev.html · pin h1288a2889d0516a28b7a660ab955619207c332bd6a845f6475aef8eb2980a026 · accessed 2026-08-18 · vendor-docGrounds: The Webflow column: Best on "Block rows validated responsive" (the Webflow designer is a reference) and Yes on "Design-token SSOT" (Webflow variables are the comparator) -- the developer documentation is the vendor surface those codes are graded against.
[drupal-multilingual] Drupal.org. Multilingual guide -- Administering a Drupal site (Drupal core multilingual: language, interface, content and configuration translation). Vendor documentation, updated 2023-11-22. publisher · read in our library knowledge/fetched/cmp_cms_drupal_multilingual.html · pin h18ca599b3e4d1b8c503907784e93955a1078c7d9e2bdcdda75382e3196c9e7f3 · accessed 2026-08-18 · vendor-docGrounds: The "Multilingual i18n" row (the note: Drupal core multilingual is the reference) and Drupal's Best code there -- our i18n_resolve is graded against Drupal core's four multilingual modules.
[wcag22] W3C. Web Content Accessibility Guidelines (WCAG) 2.2, W3C Recommendation (12 December 2024 edition): success criterion 1.4.6 Contrast (Enhanced), AAA, contrast ratio at least 7 to 1. publisher · accessed 2026-08-18 · published-standardGrounds: The "Accessibility enforced on the emitted fleet" row: WCAG-AAA contrast measured on every generated page (nx_cms_a11y a11_find) is this Recommendation's 1.4.6 bar; rivals reach a11y via tools and plugins, so the row grades the enforcement, not the intent.
[sitemaps-protocol] sitemaps.org. Sitemaps XML format -- the Sitemap protocol (urlset, url, loc, lastmod, changefreq, priority; sitemap index files). Published protocol. publisher · read in our library knowledge/fetched/cmp_cms_sitemaps.html · pin h5e7570590fcb92d473d7015c88940a830383b6a61bc6c16be4df51e97f471654 · accessed 2026-08-18 · published-standardGrounds: The "SEO sitemap and robots emitted by construction" row: every generated site ships sitemap.xml (nx_cms_seo_gate) in this format; the WordPress Best code (Yoast-class depth) is graded on the same protocol.
[graphql-oct2021] GraphQL Foundation. GraphQL Specification, October 2021 edition (selection sets, fields and field projection by the caller). publisher · read in our library knowledge/fetched/cmp_cms_graphql_spec.html · pin h4f556e7bc74ffcdf6ba3c6a4d1f541bd736492db47e97349200a8dfcb80e0ad1 · accessed 2026-08-18 · published-standardGrounds: The "REST plus GraphQL-style projection content API" row: caller-chosen field projection (cms_api_render_obj emits injection-safe JSON with the caller naming the fields) is the GraphQL selection-set idea this specification defines -- ours is GraphQL-style, not a GraphQL endpoint, and the row says so.
[rfc9309] Koster, M., Illyes, G., Zeller, H., Sassman, L. RFC 9309: Robots Exclusion Protocol. IETF, September 2022. publisher · read in our library knowledge/fetched/cmp_cms_rfc9309.html · pin hbbaa280636d3e38f0a1a130a5ce40e9753db1ed18c860945b0bf795ff8dc1a87 · accessed 2026-08-18 · published-standardGrounds: The "SEO sitemap and robots emitted by construction" row: the robots.txt every generated site ships is this standard's file, so the emitted-by-construction claim has a published grammar to be checked against.
[nike-home-2026-09] Nike. nike.com home page as served 2026-09-21 (987,082 B): the reference for an image-led full-bleed hero, uppercase display type over photography, minimal sticky navigation and a dark multi-column footer. Source read. publisher · read in our library knowledge/fetched/rf_20699706.raw · pin h884505999a7ec9f04f635e7c66864d02d10be2cdfa9741b668d09237f788894a · accessed 2026-09-21 · source-readGrounds: The Showcase presentation tier row: the tier's hero, display face, nav and footer are graded against this page read from its mirrored bytes, and the craft ruler publishes the same counts for both.
[awwwards-sep-2026] Awwwards. Winning websites -- web design inspiration, the Site of the Day listing as served 2026-09-21 (605,913 B). Mirrored index. publisher · read in our library knowledge/fetched/rf_1a3bc144.raw · pin h7643673fbd010d930d67a5005b16726a93f62bb902e76f7138487db2b138b990 · accessed 2026-09-21 · mirrored-indexGrounds: The Per-page design-craft ruler row and the SB-WEBCRAFT bar: the September 2026 listing is the month's reading of award-grade web presentation that the ruler's presence points and the pixel referee are measured toward.
[arxiv-webdesign-2026-09] arXiv API listing, query all:web design OR webpage generation OR website generation OR UI generation, sorted by submission date, 60 of 98 entries as served 2026-09-21 (147,074 B): newest 2609.08105 (2026-09-08) is a cultural-heritage homonym, newest on-subject 2609.04184 (2026-09-03). Mirrored index. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_webdesign_202609.xml · pin he6e715987441e46fe389ef92a430cead593f97415e35f74ae407d2ea8768cc1b · accessed 2026-09-21 · mirrored-indexGrounds: The SB-WEBGEN-RESEARCH bar and its barscan: the September 2026 reading of generated-web-design research is this listing, read from the mirrored bytes.
[arxiv-aesthetics-2026-09] arXiv API listing, query all:web aesthetics OR UI aesthetics OR design quality OR layout generation, sorted by submission date, 60 of 440 entries as served 2026-09-21 (148,399 B): newest 2609.23183 (2026-09-19) is an RF-circuit homonym, newest on-subject 2609.11519 InterIL (2026-09-10) and 2609.03918 (2026-09-03). Mirrored index. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_aesthetics_202609.xml · pin he416f0e7fae4ab4e11bf328668c1168d0a4c26464d7aac578a10154e0d246193 · accessed 2026-09-21 · mirrored-indexGrounds: The SB-WEBGEN-RESEARCH barscan on the layout and aesthetics axis and the constraint-solved layout row: the September 2026 listing read from the mirrored bytes.
[arxiv-2609-04184-a2ui] Yang, Y., Xiao, W., Bullough, B., Deshpande, T., Agarwal, N. Toward Frontier-Quality Declarative UI Generation at Small-Model Cost. arXiv:2609.04184, 2026-09-03 (catalog-conditioned A2UI generation selects pre-built components and binds their props; a 4B fine-tuned student recovers about 98 percent of teacher semantic quality at more than an order of magnitude lower cost). Abstract read from the mirrored listing. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2609_04184_a2ui.html · pin h29ca329418d0e021c3826ee9c93dbaf270d788177e5fb7c20cec0ce120702daa · accessed 2026-09-21 · published-paperGrounds: The section-catalog row: generation that SELECTS from a declared component catalog and binds data is the September 2026 production shape, and the blueprint-to-archetype emitter is graded against it.
[arxiv-2609-03918-asp-gui] Wang, X., Romero Davila, J., Sourulahti, S., Schaub, T., Jokinen, J. P. P. Grounding GUI Design in Computational Psychology. arXiv:2609.03918, 2026-09-03 (answer set programming generates and optimises UI layouts under grid alignment, grouping, colour harmony, whitespace and designer preferences; two user studies rated the full model above random and heuristic layouts). Abstract read from the mirrored listing. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2609_03918_asp_gui.html · pin h3e9d451cd342fcd7aba781ec331edac0449db19a1721833dd985c9e5c8536efa · accessed 2026-09-21 · published-paperGrounds: The constraint-solved section layout row and the T-FRONT-WEBGEN target: layouts solved from declared constraints over measured tokens are the September 2026 frontier this board climbs toward.
[arxiv-2609-00194-redeck] Tian, M. et al. ReDeck: Step-Level Render-Grounded Refinement for Document-to-Slide Generation. arXiv:2609.00194 v2, 2026-09-05 (one edit, one renderer-derived observation; a submission-level gate for hard layout validation; DeckQuiz decouples content fidelity, spatial correctness and design quality). Abstract read from the mirrored listing. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2609_00194_redeck.html · pin h357c86ce19be65fea3bff3f7d49b9aa60379248fc8670914740759b4603d7965 · accessed 2026-09-21 · published-paperGrounds: The render-grounded refinement loop row: emit, render, judge on the render, re-emit, with a hard layout gate before publish.
[arxiv-2608-12518-webcompat] Guo, Z., Xiao, J., Sun, Y., Huo, Y. Does It Render Everywhere? A Study of Cross-Environment Compatibility in MLLM-Generated Webpages. arXiv:2608.12518, 2026-08-12 (WebCompat: 2,032 instances from 8 AI tools across 9 browser-and-device combinations; 68 percent of generated pages show at least one compatibility issue and 88.3 percent of symptoms are whole-layout failures; the XCompat detector reaches F1 0.903). Abstract read from the mirrored listing. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2608_12518_webcompat.html · pin h5092587340189456386a12b8be1eea193a2a3a42be070f5b426fc6e38d8f43ae · accessed 2026-09-21 · published-paperGrounds: The cross-device render verification row: phone rendering is UNMEASURED on this board today and the field measures a 68 percent failure rate for generated pages, so the capture driver is a contract, not a nicety.
[arxiv-2607-22928-design-theater] Imteyaz, K., Imteyaz, K., Rajpal, N., Shaikh, K., Muller, M., Savage, S. Design Theater: Evaluating the Gap Between User-Facing Design Reasoning and Implementation in Generative UI Tools. arXiv:2607.22928 v2, 2026-07-31, accepted at AAAI/AIES (over 25 percent of stated design rationales are not implemented, 34 percent for functional requirements, across 120 interfaces from five generative UI tools). Abstract read from the mirrored listing. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2607_22928_design_theater.html · pin he55d390d80240ebf96fc2febd69d5c297252c5bb22b94ee1a46db0ac8b732a30 · accessed 2026-09-21 · published-paperGrounds: The design-decisions-verified-in-bytes row: the field measures generators saying one thing and emitting another, and a token-referenced tooth is the mechanism that refuses that gap.
[arxiv-2607-12056-agent-ready] Elnaffar, S., Rashidi, F. Designing Agent-Ready Websites for AI Web Agents: A Framework for Machine Readability, Actionability, and Decision Reliability. arXiv:2607.12056, 2026-07-13 (an agent-ready version of one site took three browser agents from 74 to 134 PASS of 150 runs, strict success 49.3 to 89.3 percent, average steps 9.31 to 6.49). Abstract read from the mirrored listing. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2607_12056_agent_ready.html · pin hedf7240c9d32ee0da4e03d275c8c9c5f54a4968ab564ef780d97fc83d8d603b1 · accessed 2026-09-21 · published-paperGrounds: The agent-ready emission row: structured data, action cues and temporal validity signals emitted by construction are graded against this framework.
[arxiv-2608-25565-maru] Kim, E., Choi, D., Min, B., Yi, H., Jiang, Y., Kim, J. Maru: Information Architecture as a Shared Language for Generating Aligned and Persistent User Interfaces. arXiv:2608.25565, 2026-08-26 (four IA elements, partition, hierarchy, order and vocabulary, persisted as rules so generated UIs stay aligned across sessions where alignment without them degrades). Abstract read from the mirrored listing. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2608_25565_maru.html · pin hc3fd24cbb210c83c749d25b0d1188f3a1144b2bc7dea8ea540c3a91d0948607b · accessed 2026-09-21 · published-paperGrounds: The brand-and-IA profile row: a persistent blueprint carrying partition, hierarchy, order and vocabulary is what keeps regenerated pages aligned, and the .site file is graded as that persistence.
[arxiv-2603-01083-aeseval] Can Vision Language Models Assess Graphic Design Aesthetics? A Benchmark, Evaluation, and Dataset Perspective. arXiv:2603.01083 (AesEval-Bench: four dimensions, twelve indicators, three quantifiable tasks, aesthetic judgment, region selection and precise localisation; proprietary, open and reasoning VLMs show clear gaps against human aesthetic assessment). Abstract read from the mirrored abs page. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2603_01083_vlm_aesthetics.html · pin hf37d724cfb1af528e858061d3ab9ac68fc93078650e6ac6ccb4657eb5229e18b · accessed 2026-09-21 · published-paperGrounds: The pairwise-preference design referee row: a VLM judge is an outside oracle whose measured gap against humans is published here, so the estate referee is calibrated against expert judgement, never against a VLM alone.
[arxiv-2508-20410-uibench] UI-Bench: A Benchmark for Evaluating Design Capabilities of AI Text-to-App Tools. arXiv:2508.20410 v3 (10 tools, 30 prompts, 300 generated sites, over 4,000 expert pairwise judgments ranked by a TrueSkill-derived model with calibrated confidence intervals; public leaderboard at uibench.ai). Abstract read from the mirrored abs page. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2508_20410_uibench.html · pin h119da865b146cfd1bc3d01b88d83198cae080ec5b062c6146e94759fbadfd82d · accessed 2026-09-21 · published-paperGrounds: The pairwise-preference design referee row: expert pairwise comparison scored by a rating model is the published method for ranking generated sites on visual excellence, and it is the outside bar our referee is measured against.
[arxiv-2601-02430-webcoderbench] WebCoderBench: Benchmarking Web Application Generation with Comprehensive and Interpretable Evaluation Metrics. arXiv:2601.02430 v3 (1,572 real user requirements; 24 fine-grained metrics across 9 perspectives combining rule-based and LLM-as-a-judge scoring with human-preference-aligned weights; no dominant model across metrics). Abstract read from the mirrored abs page. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2601_02430_webcoderbench.html · pin h9ab02096ef7c984ec3884d8eefb364937766546e1740d4c9c153b9c00ee54540 · accessed 2026-09-21 · published-paperGrounds: The design-decisions-verified-in-bytes row: a per-perspective metric set with published weights is the shape a ruler over emitted pages takes, rule-based first.
[arxiv-2606-00154-webigbench] Benchmarking Multimodal LLMs on Code Generation for Complex Interactive Webpages (WebIGBench). arXiv:2606.00154 (103 real webpages, 5 interaction types, 871 distinct interactive actions, an automated pipeline that scores interaction consistency rather than visual fidelity alone). Abstract read from the mirrored abs page. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2606_00154_webigbench.html · pin h0eb2c31ef24bd783c82ff65f4d5a8e1f170a73e3fae2140de6f39a06bf7dda6f · accessed 2026-09-21 · published-paperGrounds: The render-grounded refinement loop row: the field grades interaction, not pixels alone, so a loop that judges only a still capture is a floor.
[arxiv-2606-00112-vlm-glicko] Evolving to the Aesthetics of a Vision-Language Model. arXiv:2606.00112 (CLIP-IQA scores versus VLM pairwise contests ranked by the Glicko rating system, compared with an artist's own aesthetic ranking). Abstract read from the mirrored abs page. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2606_00112_evolving_aesthetics.html · pin he69d402a1802c92112b405cc4f88c8e0dd7000637d55e4bbb1c24596f93286d7 · accessed 2026-09-21 · published-paperGrounds: The pairwise-preference design referee row: pairwise contests with a rating system are the published way to turn a judge into a population ranking, and the artist comparison is the reminder that the judge is calibrated, not trusted.
[arxiv-2506-06251-designbench] DesignBench: A Comprehensive Benchmark for MLLM-based Front-end Code Generation. arXiv:2506.06251 v3 (React, Vue, Angular and vanilla HTML/CSS; generation, edit and repair; 900 webpage samples over 11 topics, 9 edit types and 6 issue categories). Abstract read from the mirrored abs page. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2506_06251_designbench.html · pin h554eed5ce766a8663c35b8120a950b6c0950acbf0c40a4c6e152567795ec8b3d · accessed 2026-09-21 · published-paperGrounds: The render-grounded refinement loop row: edit and repair are graded tasks in the field, so a refinement loop is measured on how it repairs, not only on what it first emits.
[arxiv-2407-08850-uicrit] UICrit: Enhancing Automated Design Evaluation with a UICritique Dataset. arXiv:2407.08850 v3 (3,059 design critiques and quality ratings for 983 mobile UIs from seven experienced designers; a 55 percent gain in LLM-generated UI feedback; a reward model for generative UI proposed). Abstract read from the mirrored abs page. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2407_08850_uicrit.html · pin h566993d1582a69cb70ae01a29b208875954fd5448db864c26f91081ba599ca0f · accessed 2026-09-21 · published-paperGrounds: The pairwise-preference design referee row: designer critiques as a dataset are the calibration evidence a referee needs before its grade is believed.
[wix-harmony-2026-09] Wix. AI Website Builder (Wix Harmony), the page as served 2026-09-21 (1,534,095 B): conversational prompting with Aria, the built-in AI agent, beside precise drag-and-drop editing; the page FAQ calls it the successor of ADI; Wix Studio linked as the agency surface. Vendor documentation. publisher · read in our library knowledge/fetched/cmp_cms_wix_harmony_202609.html · pin h3914d3ec92f58a5424dbeb897620bcfbdc3bf5d35c5195ece82e55d5bf5b818e · accessed 2026-09-21 · vendor-docGrounds: The SB-BUILDERS bar, the evidence-derived design system row and the conversational edit row: what Wix ships in September 2026, read from the page bytes.
[wix-blog-aibuilder-2026-09] Wix Blog. What is the Wix AI website builder, as served 2026-09-21 (1,975,026 B): Harmony described as a complete evolution of ADI, replacing rigid questions with an agent that builds and edits from conversation. Vendor documentation. publisher · read in our library knowledge/fetched/cmp_cms_wix_blog_aibuilder_202609.html · pin h3908663d2b3fb0038aa93eeac535cf56c259af9860a765ff6e22f82e98418f99 · accessed 2026-09-21 · vendor-docGrounds: The conversational edit row: the vendor's own account of the ADI-to-Harmony transition, read from the page bytes.
[framer-updates-2026-09] Framer. Updates, as served 2026-09-21 (932,459 B): dated entries Agents: Reasoning and Fast Mode (2026-08-12), Agents: 3D Transforms (2026-09-01), CMS List Field (2026-09-08), Pinned Projects (2026-09-14). Vendor documentation. publisher · read in our library knowledge/fetched/cmp_cms_framer_updates_202609.html · pin h6bc777ccbd100bb04d0117e8942c8dc63f7cc5879024121553e13352a2cda175 · accessed 2026-09-21 · vendor-docGrounds: The SB-BUILDERS bar and the conversational edit row: Framer ships in-editor agents with reasoning and 3D transforms this month, read from the page bytes.
[webflow-gsap-presets] Webflow Blog. Webflow interactions with GSAP, as served 2026-09-21 (401,801 B): interactions built on GSAP inside the designer, custom actions saved as reusable presets for scroll, hover and other triggers. Vendor documentation. publisher · read in our library knowledge/fetched/cmp_cms_webflow_gsap_blog.html · pin h69faffe428716de9b75e8230cbd4e58bbc69098938fa6b94a80a033d564a6d8d · accessed 2026-09-21 · vendor-docGrounds: The motion-preset row: reusable motion presets are the Webflow bar, and ours must reach the same choreography with zero script.
[webflow-aisb-2026] Webflow AI Site Builder 2026: What It Builds and Falls Short, independent review as served 2026-09-21 (309,374 B; canonical slug webflow-ai-site-builder-in-2026-what-it-actually-builds-and-where-it-falls-short). Source read. publisher · read in our library knowledge/fetched/cmp_cms_webflow_aisb_2026.html · pin h0d4870f3280b41603432f33fbffdb4026920181f0c4f5e0f764e759ac28cc2c9 · accessed 2026-09-21 · source-readGrounds: The evidence-derived design system row: an independent review of what the Webflow AI Site Builder produces, cited beside the vendor page the unite.ai roundup links.
[squarespace-design-intelligence] Squarespace. Design Intelligence, as served 2026-09-21 (380,341 B): Blueprint AI builds a site from guided prompts, then Design Intelligence generates layouts, copy and imagery inside the editor. Vendor documentation. publisher · read in our library knowledge/fetched/cmp_cms_squarespace_design_intelligence.html · pin h0ee566d75ef2a40704ad38a666058670a968870d209f563aa28f246784b90301 · accessed 2026-09-21 · vendor-docGrounds: The evidence-derived design system and brand-profile rows: Squarespace derives a site's personality into palette and type from a guided brief, the shape the evidence-derived tokens are graded against.
[stitch-designmd-2026] Tech Insider (Marcus Chen). Google Stitch AI: vibe design and a 5-screen canvas, March 2026 update, as served 2026-09-21 (177,826 B): the DESIGN.md natural-language file format for portable design systems, noted as of April 2026, a design agent, instant prototyping and an infinite canvas. Source read. publisher · read in our library knowledge/fetched/cmp_cms_stitch_march2026.html · pin h64f8313058294e2942624b9b6936ae49603f5cdeb62aed840f8bb925146c456a · accessed 2026-09-21 · source-readGrounds: The evidence-derived design system row: a portable, human-readable design-system file is the 2026 shape of a design token SSOT, and the .tokens file with n per row is graded beside it.
[uniteai-builders-2026-09] Unite.AI. 10 Best AI Website Builders (September 2026), as served 2026-09-21 (619,327 B): the month's ranked roundup, Wix, Hostinger, 10Web, Divi, Webflow, Squarespace, B12, Dorik, Framer, GetResponse. Source read. publisher · read in our library knowledge/fetched/cmp_cms_uniteai_builders_202609.html · pin h09440fb814be2a6f4e1f83dcbd7fe74b85770701d57385cf878de7638c185f15 · accessed 2026-09-21 · source-readGrounds: The SB-BUILDERS bar: the September 2026 reading of the builder market, dated by its own title, with the ten builders read from its headings.
[awwwards-soty] Awwwards. Sites of the Year, the listing as served 2026-09-21 (508,995 B): 31 Site of the Year entries, Messenger (messenger.abeto.co) among them with a 2025 submission record. Mirrored index. publisher · read in our library knowledge/fetched/cmp_cms_awwwards_soty.html · pin h95bf45a3c9aac8df66df426fde79fb60a22deab09646314226c62f5df17f39c9 · accessed 2026-09-21 · mirrored-indexGrounds: The evidence-corpus refresh row: the Site of the Year class names the pages the derive should read next, with their stylesheets.
[ia-arxiv-2026-09] arXiv API listing for web page layout and information architecture sorted by submission date, read 2026-09-22 (September entries are homonyms; newest on-subject entries are EvoGenUI-Bench 2608.29387 v2 and Maru 2608.25565) publisher · read in our library knowledge/fetched/cmp_cms_ia-arxiv-2026-09.xml · pin h6761a27187b943a2fbab67ce5a22dc490bb6c5e169763caf527928e31dee7653 · accessed 2026-09-22 · mirrored-indexGrounds: Layout tokens derived for board pages
[nng-progressive-disclosure] Nielsen Norman Group, Progressive Disclosure (the pattern that keeps a dense page readable: show the essential first, the rest on request) publisher · read in our library knowledge/fetched/cmp_cms_nng-progressive-disclosure.html · pin hbbc4a694acb9ff09128f148aaf880c7991611a545c7726882c51b654ed398aad · accessed 2026-09-22 · source-readGrounds: Records as clean tables
[nng-layer-cake] Nielsen Norman Group, The Layer-Cake Pattern of Scanning Content on the Web (readers scan headings and subheadings; a page structured in framed layers is read, a dump is skimmed) publisher · read in our library knowledge/fetched/cmp_cms_nng-layer-cake.html · pin h02afdafbd4cc2a37075dbd9576aad1d1d55143d4532b6c8a59ace44ff63827c5 · accessed 2026-09-22 · source-readGrounds: Responsive shell with a section rail
[mdn-structuring-documents] MDN Learn web development, Structuring documents (header, nav, main, article, section, aside and footer as the page skeleton) publisher · read in our library knowledge/fetched/cmp_cms_mdn-structuring-documents.html · pin h87575f8f1a77cfea5c2728f7ff1440e82ce22baa389dfae9a93efe03b186a6a0 · accessed 2026-09-22 · published-courseGrounds: Responsive shell with a section rail
[mdn-grid-common-layouts] MDN CSS grid layout, Realizing common layouts using grids (the responsive two-column shell with a sidebar and the auto-fill card grid as the field's standard forms) publisher · read in our library knowledge/fetched/cmp_cms_mdn-grid-common-layouts-202609.html · pin h01be7b5ecfaaad44cff6d922f9c5d6bd46da860f49afc41ca652e0ee9cb33cf5 · accessed 2026-09-25 · published-courseGrounds: Responsive shell with a section rail
[ghdocs-hello-world] GitHub Docs, Hello World (a documentation page as the board-class reference: a fixed left rail, an in-page table of contents on the right, one measured reading column) publisher · read in our library knowledge/fetched/cmp_cms_ghdocs-hello-world-202609.html · pin h8ccc419b93ef8db02894765b9c9f8d0dda9bb790f9adc4c6cc9cfa7edb94405f · accessed 2026-09-25 · vendor-docGrounds: Layout tokens derived for board pages
[tailwind-responsive-design] Tailwind CSS docs, Responsive design (the mobile-first breakpoint ladder the field converged on: sm 640, md 768, lg 1024, xl 1280, 2xl 1536 px) publisher · read in our library knowledge/fetched/cmp_cms_tailwind-responsive-design.html · pin h58f434f800a671f9ff5a36b5c167538d6e789661a055cfc24a620131f23f8a66 · accessed 2026-09-22 · vendor-docGrounds: Layout tokens derived for board pages
[wiki-japanese-garden] Wikipedia, Japanese garden (the composed mode's vocabulary: a designed path, framed views, hide-and-reveal, asymmetry, borrowed scenery) publisher · read in our library knowledge/fetched/cmp_cms_wiki-japanese-garden.html · pin hb321dff33118883e977c87f8ede408df6bf1675198a45e65ebde90f6de73709d · accessed 2026-09-22 · source-readGrounds: Design modes as data
[wiki-wabi-sabi] Wikipedia, Wabi-sabi (the organic mode's vocabulary: imperfection, impermanence, incompleteness, restraint) publisher · read in our library knowledge/fetched/cmp_cms_wiki-wabi-sabi.html · pin h93d7d308ed83da418cbf43a2efaba54cebbeeeb5abb675f634cc52af6a9528d2 · accessed 2026-09-22 · source-readGrounds: Design modes as data
[mdn-css-page-layout] MDN Web Docs page-layout stylesheet (styles-page-layout.5a4354f33e894319.css) -- the two-column reference-page grid, read as layout evidence publisher · read in our library knowledge/fetched/cmp_cms_mdn-css-page-layout.css · pin h863877df9887d71125e35542dd987c01d04ed194d4461021653fdf62fdc562f6 · accessed 2026-09-22 · source-readGrounds: Layout tokens
[mdn-css-reference-layout] MDN Web Docs reference-layout stylesheet (styles-reference-layout.632d2f3884f10ca3.css) -- sidebar, article and toc placement publisher · read in our library knowledge/fetched/cmp_cms_mdn-css-reference-layout.css · pin h4845f6019dd9404ba7b542716532475ae7cfac8feb445071826a8f3738a1de70 · accessed 2026-09-22 · source-readGrounds: Layout tokens
[mdn-css-left-sidebar] MDN Web Docs left-sidebar stylesheet (styles-left-sidebar.46e37af4d0b5c883.css) -- the section rail and its sticky offset publisher · read in our library knowledge/fetched/cmp_cms_mdn-css-left-sidebar.css · pin he370f061ab3394b452187747882fbbca1dc4d2fc1489280eb3a4a71760edddd8 · accessed 2026-09-22 · source-readGrounds: Layout tokens
[mdn-css-reference-toc] MDN Web Docs reference-toc stylesheet (styles-reference-toc.9b9bd019efd16b86.css) -- the in-page navigation column publisher · read in our library knowledge/fetched/cmp_cms_mdn-css-reference-toc.css · pin h38fbce35790129128ccb75640b1a88fc29d96664fb7c6394a40fd7f2f12ce691 · accessed 2026-09-22 · source-readGrounds: Layout tokens
[mdn-css-global] MDN Web Docs global stylesheet (styles-global.13616ebb1fc8ee2f.css) -- breakpoints, type and spacing tokens publisher · read in our library knowledge/fetched/cmp_cms_mdn-css-global.css · pin h9d9f419d2f413474406dd6c2bdde92392b8dbcba998ed8fe3034fd0fd1e07a29 · accessed 2026-09-22 · source-readGrounds: Layout tokens
[ghdocs-css-1] GitHub Docs global stylesheet chunk 1 (1549dbeae3096028.css) -- the Primer layout rules the docs shell is built on publisher · read in our library knowledge/fetched/cmp_cms_ghdocs-css-1.css · pin ha92c63ceaee45d3efcd81bdc3270e7dd5cf99691e90ea3c6e0c6a9ea420a770c · accessed 2026-09-22 · source-readGrounds: Layout tokens
[ghdocs-css-2] GitHub Docs page stylesheet chunk 2 (bfca54961440e8ec.css) publisher · read in our library knowledge/fetched/cmp_cms_ghdocs-css-2.css · pin ha23cdfb80290d95456d90aec3b1fcf96351f5bf2a21bb0fdfca3447fc5dd14a1 · accessed 2026-09-22 · source-readGrounds: Layout tokens
[ghdocs-css-3] GitHub Docs page stylesheet chunk 3 (24ddeaf6a0c215c7.css) publisher · read in our library knowledge/fetched/cmp_cms_ghdocs-css-3.css · pin h84d88c8f431e281c806abb9e27e48a641b7552fd750d53106d8c204c7140ec5c · accessed 2026-09-22 · source-readGrounds: Layout tokens
[ghdocs-css-4] GitHub Docs page stylesheet chunk 4 (e003d3b09fc92fd0.css) publisher · read in our library knowledge/fetched/cmp_cms_ghdocs-css-4.css · pin h05ae0d1a352d02238901c3963f6a71538b4bff6e4dcfa4d9c6b4c94018bd0abf · accessed 2026-09-22 · source-readGrounds: Layout tokens
[ghdocs-css-5] GitHub Docs page stylesheet chunk 5 (7db80dea150d23a4.css) publisher · read in our library knowledge/fetched/cmp_cms_ghdocs-css-5.css · pin h1743dfdf1a5d1ec36f0e3d014161d6d0904c7cb83200930cd2c139fa2b74f575 · accessed 2026-09-22 · source-readGrounds: Layout tokens
[tailwind-css-1] Tailwind CSS docs stylesheet chunk 1 (017r-ibrf-le-.css) -- range-syntax breakpoints and the docs sidebar tracks publisher · read in our library knowledge/fetched/cmp_cms_tailwind-css-1.css · pin hc4870da5ca1efe37273ba6a706c47825413c72bc2c9c553623f6f7e56027a40e · accessed 2026-09-22 · source-readGrounds: Layout tokens
[tailwind-css-2] Tailwind CSS docs stylesheet chunk 2 (0n9_1tq74i09x.css) publisher · read in our library knowledge/fetched/cmp_cms_tailwind-css-2.css · pin h4381b1c970af2bb270f5e22caab62bb88f1c6fe88af3b7f3d8890afee6e21133 · accessed 2026-09-22 · source-readGrounds: Layout tokens
[nng-css-1] Nielsen Norman Group site stylesheet 1 (output.e9c610b5facf.css) publisher · read in our library knowledge/fetched/cmp_cms_nng-css-1.css · pin he9c610b5facfa1c4b9d5fa605e3ae7e68da5bb8e9331f0b8f917102c2f23dc06 · accessed 2026-09-22 · source-readGrounds: Layout tokens
[nng-css-2] Nielsen Norman Group site stylesheet 2 (output.abb342b5d0ed.css) -- the article layout, breakpoints and sticky rules publisher · read in our library knowledge/fetched/cmp_cms_nng-css-2.css · pin habb342b5d0ed23bd79fc6f6870e09200d430c4d54ebfe6d77f5df6e61aaded96 · accessed 2026-09-22 · source-readGrounds: Layout tokens
[atf-contrastswatch] Google. Accessibility Test Framework for Android, ContrastSwatch.java (master, Apache-2.0): the exact-colour histogram of a screenshot region, split at the unweighted mean luminance of its distinct colours, the most frequent colour on each side of the split, the more frequent of the two called the background (the default single-foreground method; a reduced-histogram multi-foreground method is opt-in). Source read. publisher · read in our library knowledge/fetched/cmp_cms_atf_contrastswatch.java · pin hd989f2f95993003728ee6ed56e678d6977c402153aa994d8fd8719546953ad6f · accessed 2026-09-25 · source-readGrounds: Rendered-pixel design referee on emitted pages
[atf-contrastutils] Google. Accessibility Test Framework for Android, ContrastUtils.java (master, Apache-2.0): the WCAG relative luminance and contrast ratio, and the CIE94 colour difference the reduced histogram merges near colours by. Source read. publisher · read in our library knowledge/fetched/cmp_cms_atf_contrastutils.java · pin h2c066165350c222d2f8b7f0cbc6a91d0f9ce55a753be229a5443132c0ebdfc5a · accessed 2026-09-25 · source-readGrounds: Rendered-pixel design referee on emitted pages
[atf-textcontrastcheck] Google. Accessibility Test Framework for Android, TextContrastCheck.java (master, Apache-2.0): the declared text and background colours are graded first (ERROR when the upper bound misses the bar, WARNING when only the lower bound does); the screenshot swatch, cropped to the text character bounds, runs only where the background is not a known opaque colour, and its heuristic ratio reports a WARNING, never an ERROR. Source read. publisher · read in our library knowledge/fetched/cmp_cms_atf_textcontrastcheck.java · pin hf06aea09114636ba55b186b262ad5322c08a56820ac785fe16a0b0527b330cb3 · accessed 2026-09-25 · source-readGrounds: Rendered-pixel design referee on emitted pages
[wcag-understanding-1-4-3] W3C. Understanding Success Criterion 1.4.3: Contrast (Minimum), the Understanding document's source in the w3c/wcag repository (main; the w3.org page refused the fetch with 403): because authors do not control font smoothing or anti-aliasing, evaluate the foreground and background colours obtained from the user agent or the stylesheets rather than the text as presented on screen; thin or unusual fonts can render much fainter than their CSS colour, and the best practice is a stronger font or a colour pair that exceeds the requirement. Published standard. publisher · read in our library knowledge/fetched/cmp_cms_wcag_understanding_contrast_minimum.html · pin h82da1aae840826f0f7d93ba3a75d9b68b86327763bdba2b6af15e5393aaa8813 · accessed 2026-09-25 · published-standardGrounds: Rendered-pixel design referee on emitted pages
[claude-code-changelog-2026-09] Anthropic. Claude Code changelog, as served 2026-09-25 (3,811,024 B): a three-column documentation shell (a section rail that marks the current page, a measured reading column, an on-this-page rail with one link per release), a serif display title under a small uppercase eyebrow, page actions (copy page, a feed), and every release a short record -- a version chip, its date, and bullets with identifiers set as inline code. Source read, captured in a real browser at both doors. publisher · read in our library knowledge/fetched/cmp_cms_claude_code_changelog_20260925.html · pin h95bdd83f054adbc473a9e8834945cb811e1b5736c9ba45c8856111e2da67d42a · accessed 2026-09-25 · source-readGrounds: Records as clean tables
[awwwards-evaluation-2026-09] Awwwards, Evaluation System (criteria, jury, voting), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_awwwards_evaluation_202609.html · pin h391de7d9903f3586df8638eb32c58638ea3fd3fc0b6f18f6266f5db8825f2005 · accessed 2026-09-26 · vendor-docGrounds: Award rubrics as referee axes
[webby-judging-2026-09] The Webby Awards, Judging Criteria (Websites and Mobile Sites: seven criteria), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_webby_judging_202609.html · pin h488e41c6ce48da8be74ad9519cee11d65d1800decc962d5d32792087c79a2792 · accessed 2026-09-26 · vendor-docGrounds: Award rubrics as referee axes
[fwa-about-2026-09] The FWA, About (100-point judging, score bands, FOTD cycle), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_fwa_about_202609.html · pin h7981808f0b5f13f06c9ef1f376b93de123ccc66123becf62613535bd48c1b262 · accessed 2026-09-26 · vendor-docGrounds: Award rubrics as referee axes
[baymard-research-2026-09] Baymard Institute, Research (200,000+ hours, 700+ guidelines, benchmark topics), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_baymard_research_202609.html · pin h4eff8d6e912e0d0ccd9baeabe7106476ba5fcbe8961d23121e5e7fe22def4275 · accessed 2026-09-26 · vendor-docGrounds: Journey driver measures the 1-2-3 path in the browser
[nngroup-heuristics-2026-09] Nielsen Norman Group, 10 Usability Heuristics for User Interface Design, read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_nngroup_heuristics_202609.html · pin hde4dbb2d9300080626507cd782a9908dbffd571b80f7f86a9fa345a81517f0ce · accessed 2026-09-26 · published-courseGrounds: Journey driver measures the 1-2-3 path in the browser
[eda-home-2026-09] Ecommerce Design Awards, home (the curated ecommerce log; Shopify, WooCommerce, BigCommerce winners), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_eda_home_202609.html · pin h41ca8250f4f18efc07c30e530cd7fe396934bb7762e3ca8fe56abac334ca7524 · accessed 2026-09-26 · mirrored-indexGrounds: Evidence corpus refreshed from the month's award listings (winners' pages and stylesheets), tokens re-derived with n per row
[builtformars-2026-09] Built for Mars, home (90+ UX case studies, 1,400+ UX examples), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_builtformars_home_202609.html · pin h066f0b31e111427543164f75532fc593f60eecf9d13deceea3ce76d27d603cd6 · accessed 2026-09-26 · mirrored-indexGrounds: Journey driver measures the 1-2-3 path in the browser
[pageflows-2026-09] Page Flows, home (user-flow screencast library), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_pageflows_home_202609.html · pin h45931ca399789d1be5b75dbf418aad923c26664cc8a3db82af24dcc6efc48874 · accessed 2026-09-26 · mirrored-indexGrounds: Journey driver measures the 1-2-3 path in the browser
[siteinspire-2026-09] Siteinspire, home (curated showcase updated daily, filterable by style, type, subject), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_siteinspire_home_202609.html · pin hf06aaf2342fe351ffc44c94d6423f266780dc77b27c3f1f73e01aff3272fe83b · accessed 2026-09-26 · mirrored-indexGrounds: Evidence corpus refreshed from the month's award listings (winners' pages and stylesheets), tokens re-derived with n per row
[landbook-2026-09] Land-book, home (landing-page and product-marketing gallery), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_landbook_home_202609.html · pin hf54bae0af5db8e5fed3acd61d6b877f6dc4fb443e7df83d28ea1d1a323d43859 · accessed 2026-09-26 · mirrored-indexGrounds: Evidence corpus refreshed from the month's award listings (winners' pages and stylesheets), tokens re-derived with n per row
[arxiv-godel-machine-2026-09] arXiv API listing, all:Godel machine sorted by submission date (Mendel GM 2608.07645, Red Queen GM 2606.26294, SGM 2510.10232, DGM 2505.22954), fetched 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_arxiv_godel_machine_202609.xml · pin h78e9cc6af5c020cbbab3f71bf96b8099a82808961af3e4f048d482b3e7dce383 · accessed 2026-09-26 · mirrored-indexGrounds: Variant archive and selection in the shape of the Darwin and Mendel Godel Machines
[arxiv-selfimproving-2026-09] arXiv API listing, all:self-improving AND all:agent sorted by submission date (492 results, newest 2026-09-24; RegenHarness 2609.27612, RRSI 2609.24972, GuardrailLoop 2609.12216, LLM-as-a-Judge Is Not an Oracle 2609.02246, harness tampering 2609.00069, trusting trust 2609.17817), fetched 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_arxiv_selfimproving_agent_202609.xml · pin hf5dcb5721b5df281549c4aa6b983c555e2fc141c5c611ed3b949c0a7884448cb · accessed 2026-09-26 · mirrored-indexGrounds: Guarded self-improvement of the improver
[arxiv-ui-optimization-2026-09] arXiv API listing, all:user interface AND (bayesian optimization OR bandit OR multivariate) sorted by submission date (19 results; newest on-subject 2508.16077, 2501.11829, 2006.11873), fetched 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_arxiv_ui_optimization_202609.xml · pin h1887c0ea82d0f21626a964c6d905393a286a174c1788f67cfdf80d7c7cd42881 · accessed 2026-09-26 · mirrored-indexGrounds: Multivariate optimisation of the design knobs
[arxiv-abtesting-web-2026-09] arXiv API listing, all:A/B testing AND (web OR website OR landing page) sorted by submission date (106 results; 2608.07517 calibrated abstention for A/B prediction, 2509.15576 stratified sampling, 2508.08077 Bayesian A/B with Dirichlet posteriors, 2507.12118 A/B decision support for web usability), fetched 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_arxiv_abtesting_web_202609.xml · pin h183155728b265103a844810296bf461723c75cb042994dc76f3275d3b890ba2f · accessed 2026-09-26 · mirrored-indexGrounds: Multivariate optimisation of the design knobs
[webflow-optimize-2026-09] Webflow Optimize product page (A/B testing, multi-variable testing, AI-powered optimization, personalization on live traffic), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_webflow-optimize-2026-09_canon.html · pin he1d723b0a6d275bccbea480b3d42000e136250d00afa3fb706b28229f37a0ee9 · accessed 2026-09-28 · vendor-docGrounds: Multivariate optimisation of the design knobs
[wp-plugins-abtesting-2026-09] WordPress plugin directory, search results for ab testing (57 plugins: Nelio A/B Testing, EasyTest, My WP A/B Testing, abtestkit, Mida), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_wp_plugins_abtesting_202609.html · pin h179fa325ae1ffc92628df483654440ddba37d507084a3ac804c31599a49f1622 · accessed 2026-09-26 · mirrored-indexGrounds: Multivariate optimisation of the design knobs
[drupal-abjs-2026-09] Drupal.org, A/B Test JS module (A/B and multivariate testing via JavaScript with a UI to create tests), read 2026-09-26 publisher · read in our library knowledge/fetched/cmp_cms_drupal_abjs_202609.html · pin h85619b840f539d692841a153cc110895cb94f64121ed9db0043a136a9eb50577 · accessed 2026-09-26 · vendor-docGrounds: Multivariate optimisation of the design knobs
[arxiv-arena-preference-2026-09] arXiv API listing, query (arena OR side-by-side) AND human preference AND (annotation OR evaluation), sorted by submission date, 30 entries as served 2026-09-28 (80,638 B): newest 2609.30221 WanPE (2026-09-24, 11K blind pairwise assessments of video prompts, on-subject as method only), newest on-subject 2609.19599 ANCHOR (2026-09-17) and 2609.15387 WebCraftBench (2026-09-14); read through nx_arxiv_digest. Mirrored index. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_202609_arena_human_preference.xml · pin h734cabfe83bf81a1deeaa816e3ee15c1b591bddeccf94e3946f007933aee2c58 · accessed 2026-09-28 · mirrored-indexGrounds: Dwell per verdict as gap evidence (a meta row: side shown, milliseconds on the pair, door)
[arxiv-pairwise-ui-2026-09] arXiv API listing, query pairwise AND preference AND (user interface OR web design OR website), sorted by submission date, 13 entries as served 2026-09-28 (37,953 B): newest 2609.03952 WorldReward (2026-09-03, a VLM pairwise reward model for camera-conditioned video, a method homonym), newest on-subject 2604.09876 (2026-04-10, 20 participants over 600 GenUI pairs at Krippendorff alpha 0.25) and DesignPref 2511.20513; read through nx_arxiv_digest. Mirrored index. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_202609_pairwise_ui_preference.xml · pin h5b9f1675c58d98a3e054356c74cee7fa5af093d375b33724ba969df5df772b37 · accessed 2026-09-28 · mirrored-indexGrounds: Inter-judge agreement as Krippendorff alpha beside the pairwise agreement
[designarena-home-2026-09] Design Arena, home as served 2026-09-28 (249,706 B): the largest crowdsourced benchmark for AI-generated design; its methodology strings read from the served bytes: each voting session randomly selects four models plus one backup, two randomly paired models are presented anonymously side by side and a choice is one vote, winner and loser brackets with a tie breaker to a final ranking, models ranked by the Bradley-Terry rating system denoted Elo starting at 1200, a margin of error as plus or minus the standard error, win rate and the most common placement per model. Vendor documentation. publisher · read in our library knowledge/fetched/cmp_cms_designarena_home.html · pin hfb86b7f846edf7c8b96ba3439298cbfd896f2614b928a18bc156ebe0c1e3f424 · accessed 2026-09-28 · vendor-docGrounds: Blind side-by-side pair judging surface (one pair at a time, Left and Right by a keyed coin, the click mapped back to the referee's A and B)
[designarena-leaderboard-2026-09] Design Arena, Leaderboards as served 2026-09-28 (310,681 B): the leaderboard families (website, web apps, UI component, data visualization, image, video, audio, builders); the ratings themselves are client-rendered and absent from the served bytes, so no number is declared from this page. Mirrored index. publisher · read in our library knowledge/fetched/cmp_cms_designarena_leaderboard.html · pin h1b2c6ab033bc02565d03389822067beb97ee1a907ac6a4042f74e01525b82648 · accessed 2026-09-28 · mirrored-indexGrounds: Standings on the judging surface: rating, W-L-T, win share with its Wilson interval
[lmarena-codearena-2026-09] LMArena, Code Arena (the WebDev arena's current name) as served 2026-09-28 (280,222 B): a script-rendered shell whose served text carries the title Code Arena, Build and Test with AI Coding Models and the Battle, Code and Leaderboard navigation; the vote mechanics are not in the served bytes. Mirrored index. publisher · read in our library knowledge/fetched/cmp_cms_lmarena_webdev_arena.html · pin ha5469c12710a96c1d7312ae7acb70257f1b3016fc7fb13889e6cccd7d2e7d546 · accessed 2026-09-28 · mirrored-indexGrounds: Blind side-by-side pair judging surface (one pair at a time, Left and Right by a keyed coin, the click mapped back to the referee's A and B)
[arxiv-2609-19599-anchor] Human-Anchored Inference for Ranking New Models with Large Language Model Judges (ANCHOR). arXiv:2609.19599, 2026-09-17 (historical human and LLM-judge comparisons give each judge's sensitivity to human score differences and its feature-dependent bias; a Neyman-orthogonal estimator infers a new model's human-reference score at the semiparametric efficiency bound; lowest score RMSE and insertion MAE on Chatbot Arena). Abstract read from the mirrored listing. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2609_19599_anchor.html · pin h081988d06566e2de6ec2f43d0c6791e445f6f4d26947da9b8d33b5917d9656e5 · accessed 2026-09-28 · published-paperGrounds: Human-anchored machine co-judge with a gap-calibrated vote
[arxiv-2609-15387-webcraftbench] WebCraftBench: Evaluating Web Application Generation from a Software Testing Perspective. arXiv:2609.15387 v3, 2026-09-14 (coverage-guided exploration of each generated app, a state-transition graph, three dimensions visual aesthetics, usability and requirement alignment, 369 requirements and 5,088 acceptance criteria, 85.3 percent agreement with human preference on 197 arena sessions, the agreement rising with the score difference between the pair). Abstract read from the mirrored listing. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2609_15387_webcraftbench.html · pin h296a35eb5f9c7c354f52f742c0671b533f3759a48d8f13469f71f6982e7e7251 · accessed 2026-09-28 · published-paperGrounds: Per-axis verdicts: aesthetics, usability, requirement alignment and overall
[arxiv-2602-09411-ksort-eval] K-Sort Eval: Efficient Preference Evaluation for Visual Generation via Corrected VLM-as-a-Judge. arXiv:2602.09411, 2026-02-10 (a VLM judge's posterior corrected by its consistency with human votes from K-Sort Arena, and a dynamic matching strategy that balances uncertainty and diversity to maximise the expected benefit of each comparison, typically under 90 model runs). Abstract read from the mirrored listing. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2602_09411_ksort_eval.html · pin h4a67cd11f66d7e04e4bf18cb0db7e604f6974f15cb8ce21048ad4144c26d208f · accessed 2026-09-28 · published-paperGrounds: Next pair by information: fewest judgments then the smallest rating gap
[arxiv-2511-20513-designpref] DesignPref: Capturing Personal Preferences in Visual Design Generation. arXiv:2511.20513, 2025-11-25 (12k pairwise comparisons of UI design generation by 20 professional designers with multi-level ratings; Krippendorff's alpha 0.25 for binary preferences; personalised models beat aggregated judges with 20 times fewer examples). Abstract read from the mirrored listing. publisher · read in our library knowledge/fetched/cmp_cms_arxiv_2511_20513_designpref.html · pin hc36aee2122f9023ca9e425fc03222c1d4156f058fdff13b254de07cf9c47465e · accessed 2026-09-28 · published-paperGrounds: Inter-judge agreement as Krippendorff alpha beside the pairwise agreement
[awwwards-sotd-2026-10] Awwwards. Winning websites -- web design inspiration, the Site of the Day listing as served 2026-10-01 (604,461 B): the newest entry is Colonia Zacamil by Adoratorio Studio, Site of the Day and Developer Award, dated Oct 01, 2026; the September entries follow it. Mirrored index. publisher · read in our library knowledge/fetched/cmp_cms_awwwards_sotd_202610.html · pin h8a0318d6529cd533b8477b249ddbbefc4b8ea53210da8acb5f4c1ec6103010e2 · accessed 2026-10-01 · mirrored-indexGrounds: Showcase presentation tier (image-led hero, display type, full-bleed, dark band)
[awwwards-evaluation-2026-10] Awwwards, Evaluation System (criteria, jury, voting), as served 2026-10-01 (216,181 B): Design 40, Usability 30, Creativity 20, Content 10; a minimum of 18 jury members, the 3 scores furthest from the average eliminated, a 5-day vote. Vendor documentation. publisher · read in our library knowledge/fetched/cmp_cms_awwwards_evaluation_202610.html · pin h85ae70e6df485ffad17a7d60f32ef0974d2c8baf475807ff037cd445cd082ebb · accessed 2026-10-01 · vendor-docGrounds: Award rubrics as referee axes

generated by nx_swcompare_matrix (sovereign NishiLang organ) from knowledge/compare/cms.matrix · source checks show implementation presence; runtime and user-outcome evidence are reported separately · JavaScript supports page controls