Nishi Family › Compare › Code Effectiveness

Nishi Compare · capabilities, evidence and delivery

Code Effectiveness

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

How humans and AI author NishiLang and drive the tool system -- the four delivery pillars (deploy, build, feature development, bug resolution) each bound to a LIVE estate number and a Sept-2026 field bar, measured full-population, instruments composed not rebuilt. An honest position, not a score.

Overview

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

Where we are. Measured 2026-09-05 full population. The estate has more effectiveness instrumentation than any field platform and its headline gauge is RED: nx_seat shift reads Claude output per closed rung at 5,892,596 tokens against 1,030,739 the prior 30-day window, a 5.7x rise, RED by construction. STRONG and proven: the four DORA keys derived from the sovereign event stream with zero git-poll (deploy ELITE, lead ELITE, change-fail 42 percent proxy, MTTR pending and shown pending); a capability-loss diff that names every lost run before a promote; a drift census that rebuilds source and compares served bytes with a daily trend; a mutation harness that rebuilds and redeploys the subject per mutant; an operator goal mapped onto the boards in one job where a subagent fan-out cost 5,713,788 tokens; a clearance-aware dispatcher that routes by a byte-exact judge. WEAK and equally proven: the human-versus-AI NishiLang authoring lens the operator named does not exist -- the authoring record and the health rulers are joined nowhere; there is no first-try build rate and no build-rework rate; bug-resolution effectiveness is measured as a close count not a recurrence-rate reduction; base-class gate adoption is 452 of 3477 with 2759 hand-rolled and unmeasured; and a live board defect proves the need for verdict-bound claims -- codequality asserts the complexity keystone SHIPPED while nx_catalog reads its binary SOURCE-ONLY.

Where we need to go. Turn the four delivery pillars into frontier SOTA by joining what the estate already measures into an effectiveness a human or an AI can act on, and by making every effectiveness claim REFUSE when a live verdict contradicts it -- the exceed we hold is that ours refuse where the field reports. Drive the RED gauge down without lowering quality, measured by nx_seat shift with its paired floor. Answer the operator's exact question -- how humans and AI author NishiLang -- with a full-population authorship-versus-health join, not a dashboard.

The unit. 1 u = one measured axis shipped as organ plus gate plus published row, the pmdash and charsim calibration.

Research bar. DORA State of AI-assisted Software Development 2025 is measured on DORA Google Cloud 2025. Theirs: AI is an amplifier magnifying existing organizational strengths and weaknesses, and elite delivery is multiple deploys per day with lead time in hours and change-fail under 5 percent [dora-2025]. Ours: GAUGE and DEPLOY.

Research bar. GitClear The Maintainability Gap 2026 is measured on GitClear 2026. Theirs: block duplication up 81 percent 2023 to 2026, refactored lines down from 21 to 3.8 percent, two-week churn up another 15 percent [gitclear26]. Ours: AUTHORING and BUILD and DUPLICATION.

Research bar. METR randomized controlled trial on experienced developer productivity is measured on METR 2025. Theirs: experienced developers were 19 percent slower with early-2025 AI while forecasting a 20 to 24 percent speedup [metr]. Ours: GAUGE.

Research bar. CodeScene Code Red the business impact of code quality is measured on Tornhill and Borg 2022. Theirs: files in the unhealthiest code-health band carry about 15 times more defects and cost 124 percent more time to change [codered22]. Ours: AUTHORING and BUGRES.

Research bar. Stop Comparing LLM Agents Without Disclosing the Harness is measured on Zhang et al 2026. Theirs: harness configuration governs more performance variance than model choice and must be disclosed beside a number [harnessthesis]. Ours: RIGOR.

Where developer and PM effectiveness measures come from: Estate-owned; UNMEASURED until exercised. CE12 supplies the shared outcome evidence for lang:LN107/LN108 and pmdash:PR21. Product quality requirements and applicable standards are explicit constraints; a vendor code-health score is one scoped instrument, never overall quality. External benchmarks keep their population, workload, date, harness and uncertainty. METR's 2026 update warns that participation/task selection and concurrent agent time complicate inference; the earlier 19 percent result is historical, not a current universal estimate. Microsoft EngThrive adds speed, ease, quality and developer thriving to system-level assessment. Sources [metr-uplift-20260224] [engthrive-2026].

Latest recorded release

2026-09-06 · CE9

LAST MILE PROMOTED: nx_seat 360016 B sha 30309d65 via /api/build (src_path named, src_stable 1). Promote guard refused CAPLOSS; nx_contentdiff enumerated 4 lost runs (3 permil) and every one was adjudicated before the flag: manifest-shrank and prefix-mismatch ABSENT-PROVEN from all runtime sources (corpus_complete=1), docs-size-mismatch present only inside a gate COMMENT (no source emits it), ssl_total_keys_range a seg-store lib FUNCTION NAME the rebuilt call graph no longer reaches -- the seg-store's successor vocabulary (uncovered-vanished-keys, INCR-FOLD, tombstoned=) and CE6's ce_number_envelope are among the 53 gains: library evolution since the live binary was built, not this change. Promoted with allow_capability_loss=yes on that enumeration. Proof run (nx_seat shift 7) in flight; the codeeffectiveness.gauge row lands once the stamp exists.

Release history and work log

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

20 of 21 capability source symbols found|12 of them authored stronger-claims|1 open|coverage 952/1000|adoption 14 full / 6 partial

Evidence profile — what the gaps on this board actually are

Unstamped. No evidence profile is published for this domain yet: run nx_swcompare_evidence codeeffectiveness and this section fills itself on the next beat. No numbers are shown, deliberately — a zero-filled profile would read as a board with no gaps, which is the one wrong answer nobody would question.

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_INCONSISTENT. 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_INCONSISTENT. 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.
Human-versus-AI NishiLang authoring joined to code health (CE1) after CE5	ce_author_health	THE KEYSTONE and the operator's own question. Join the authoring record (nx_claude_harvest frames, which carry which seat authored a change) to the health rulers (nx_mccabe complexity, nx_forkgrade duplication, nx_slopmeter self-repetition) full population, splitting each metric by human-seat, AI-seat and generated authorship. ACCEPT RULE declared in advance: FIRST measure and publish the AUTHORSHIP-MARKER ADOPTION across the corpus, because a split is only as good as the marker, and ship an UNKNOWN bucket rather than a split that silently miscounts; the field bar is GitClear's 81 percent duplication rise and 21 to 3.8 percent refactor fall and CodeScene's 60 percent higher AI defect risk in unhealthy code, so the organ must emit our own trajectory number with its horizon beside it -- a single census is a level and a trend needs two in time, reported UNOBSERVABLE until then	Organ	2.5 u
First-try build success and the build-rework rate (CE2)	ce_build_firsttry	The build pillar has no rate. Read organ_ship.jrnl and the drift census and compute the first-compile-pass rate and the 14-day build-rework rate, where a byte decrease after an addition is a revert. ACCEPT RULE: two consecutive beats emit both rates whose arithmetic reconciles against organ_ship.jrnl row counts, and a single-sample window reads UNOBSERVABLE rather than a fabricated rate; a neg-control of a known revert must raise the rework count by exactly one	Organ	1.5 u
Bug-resolution effectiveness as recurrence-rate reduction (CE3)	ce_recurrence	Resolution effectiveness is not a close count -- the estate's own record has the same defect filed five times in three weeks. Compose nx_failclass and the debt board to publish recurrence per defect class before and after a fix, so a resolved class that recurs raises its own count. ACCEPT RULE: a planted recurring class must show a non-decreasing count across two beats, a genuinely resolved class must fall, and the measure must name the prior filing ids rather than a bare number	Organ	1.5 u
A board effectiveness claim is re-measured against a live verdict (CE4)	ce_claim_binding	Extend the incumbent binding vocabulary (nx_swcompare_sota already re-measures symbol presence and classifies CONSISTENT, STALE-UNDER, UNBACKED) with gate=<gate> and catalog=<name> whose verdict is read against the LIVE artifact, so a row asserting a capability the verdict denies REFUSES to publish and one asserting an absence the verdict contradicts publishes STALE-UNDER. ACCEPT RULE: the motivating case is the fixture -- a claim that nx_mccabe SHIPPED must publish STALE-UNDER while its catalog verdict reads SOURCE-ONLY, and flip to CONSISTENT only once CE5 lands; extend the incumbent, never add a second ruler. OPERATOR REQUIREMENT 2026-09-07, not claimed enforced or complete: the estate must apply the same evidence admission to human and AI claims on /compare and wiki. Each functionality claim binds the exact artifact hash/version, test or experiment version, inputs, environment, dated result and provenance; an independent reviewer must approve it, and the author cannot self-approve. Missing or expired evidence remains UNKNOWN or STALE, never silently current; preserve failure evidence and an immutable additive correction/retraction history. Refuse unsupported claim upgrades and prove the refusal on planted unsupported, stale, version-mismatched and self-approved fixtures. Bounded passing tests establish only the tested scope and do not prove universal truth; render the scope and remaining uncertainty beside the claim	Organ	2 u
Close the complexity-keystone gap on the serving tree (CE5)	ce_mccabe_live	codequality asserts the cyclomatic-complexity keystone SHIPPED with its gate GREEN 17 of 17, while nx_catalog reads nx_mccabe and nx_mccabe_gate SOURCE-ONLY -- built nowhere on the NAS. Build both, prove the gate GREEN against the live binary, promote, register, and re-run the catalog. ACCEPT RULE: nx_catalog nx_mccabe reads LIVE and nx_mccabe_gate returns GREEN against the promoted binary, and codequality's CQ1 row reads CONSISTENT under CE4's binding rather than a hand-asserted SHIPPED; the complexity distribution is republished full population with its histogram summing exactly	Organ	1.5 u
Every effectiveness number carries its interval and a null control (CE6)	ce_number_envelope	A replay script that never observes the work matched frontier agents on static benchmarks, so a bare percentage is not evidence. Put n, an interval and a null-control value in the gate base class so this board and the others read one ruler and the emit REFUSES a bare percentage. ACCEPT RULE: a figure without n and an interval does not publish, and a planted bare percentage is refused by name	Organ	1 u
The autonomous fix loop registered and proven on one real task (CE7) after CE6	ce_fixloop_live	The loop is BUILT and its grader PROMOTED-UNREGISTERED. Register it and resolve one real debt-board task end to end -- localise from the grader rows, propose, verify by the byte-exact judge on the SWE-bench contract, revert proven exact. ACCEPT RULE: one episode resolves a real task with its RED-before-and-GREEN-after receipt banked and a null control (empty patch, revert patch, echo) scoring zero beside it	Organ	2 u
Duplication trend not just level (CE8)	fg_trend	Shared with codequality -- one census is a level and the field bar is a trend (block duplication up 81 percent). The organ appends to a durable log on a beat and refuses to emit a trend from a single sample. ACCEPT RULE: two beats reconcile and a single sample reads UNOBSERVABLE; do not add a second duplication ruler, extend nx_forkgrade	Organ	1 u
The gauge is a beat with an axis-blind alarm so RED can never go dark (CE9)	ST_GAUGE_STAMP	nx_seat shift is the gauge and it reads RED now; the risk is that a stale gauge reads as no movement rather than as blindness. A beat stamps the gauge and SessionStart prints its age, and an abstaining axis raises an axis-blind alarm. ACCEPT RULE: a stale gauge reads STALE on this page and never as zero, and a forced-stale fixture raises the alarm	Organ	1 u
A mirror whose canonical URL differs from the cited URL reads MOVED, never PROVEN (CE10)	cr_canonical_moved	FOUND BUILDING THIS BOARD: the Faros /ai-impact page every other board cites now redirects to the Faros homepage, and a fresh fetch returned a 126648-byte page carrying none of the cited terms while printing status=200 -- the fetcher flagged it with a trailing 1 in its PROVENANCE line and no consumer reads that flag, so a moved page can be pinned as evidence. Extend nx_compare_refs_gate (the incumbent, never a second ruler) with a canonical-URL check: read the mirror's rel=canonical or og:url, normalise scheme, host and trailing slash, and classify MOVED when it differs from the cited URL, rendering it with the same drift classes as STALE-UNDER. ACCEPT RULE declared in advance: the Faros homepage mirror rf_ff06ac9a.raw cited under its old URL must read MOVED, the same page cited under its canonical must read PROVEN, an arxiv abstract whose canonical is itself must read PROVEN, a mirror with no canonical tag must read UNKNOWN never MOVED, and the fetcher's own trailing flag must be consumed and named in the verdict; bite-proven both directions before the tooth is armed	Organ	1 u

Milestones

Milestone	Rungs	Cumulative
E0 · The keystone authoring join and the verdict binding land	CE1,CE4,CE5	3 u
E1 · Every delivery pillar has a rate not just an event	CE2,CE3,CE7	5 u
E2 · The estate sees its own effectiveness trajectory under one envelope	CE6,CE8,CE9	7 u

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
A gv_kv that republishes a quantity a gv_check_eq already asserted is a SECOND RULER (CE11) after CE6	gl_l5_double_publish	MEASURED 2026-09-06, and this rung exists because the defect was found in a shipping gate rather than imagined: nx_uxf_mp4box_gate passed 8 of 8 GREEN while publishing toplevel_box_count=0 against its own tooth T2b asserting actual=2 expected=2. One cnt buffer is filled by the good-tree walk, the tooth tests cnt[0] and passes on 2, the neg-control then re-walks into THAT SAME out-parameter leaving 0, and the gv_kv emission runs last -- so the published number was the neg-control's answer wearing the tooth's name, and a buffer restore two lines later could not reach it because a restore undoes the INPUT and not the DERIVED state. THE GENERAL LAW: gv_check_eq binds the number tested to the number published ONLY for values passed INTO it, and a separate gv_kv that re-reads a mutable buffer afterwards carries no such binding, which makes it a second ruler over a quantity the gate has already ruled on. The signal is mechanical and greppable -- the same variable expression appearing both as a gv_check_eq argument and as a gv_kv argument within one gate. EXTEND THE INCUMBENT AND DO NOT ADD A RULER: this is an L5 axis on nx_gatelaw_gate beside L1 exit-carries-verdict, L2 names-a-neg-control, L3 base-class and L4 emits-values, and it is DECLARED SEPARATE from all four because a gate can satisfy every one of them and still double-publish, so folding it in would silently redefine a figure other lanes act on. nx_capsearch over 7551 surfaces with corpus_complete=1 returned no incumbent detector, so the gap is measured rather than assumed. ACCEPT RULE DECLARED IN ADVANCE: the axis must FIRE on the pre-fix nx_uxf_mp4box_gate source as a planted positive and must NOT fire on its fixed source, it must measure and publish its false-positive rate against hand-adjudicated hits before any count is published as a defect count, and a gate whose dialect it cannot read must be left UNSCORED on this axis rather than asserted clean, because a ruler that declares a stream unreadable must not then assert anything about it on another axis	Organ	1.5 u
Accepted task, workflow and project outcomes for humans, AI and their partnership (CE12) after CE1,CE2,CE3,CE6,CE7	_ABSENT_:ce_accepted_outcomes_accept	PLANNED CONTRACT 2026-09-30. Extend existing authorship, build-first-try, recurrence, number-envelope, fix-loop and resource instruments. Compare matched human-led, AI-led within authority, and human-plus-Nishi assistance conditions with exact tools/model/policy versions and equivalent tasks/resources; mechanistic help is an explicit treatment. Report human active time, elapsed delivery time, review/rework, quality, retained features, escaped defects/recurrence, resource and actual token/cost demand per accepted outcome. Keep task/workflow/project denominators separate; retain mixed/unknown attribution, unfinished/censored work, selection effects, concurrency and uncertainty. Independent acceptance is fixed before candidate work; speed cannot offset correctness/safety/capability loss. Establish efficiency baselines first, then representative studies. No Nishi result is claimed until executed. Compare emits the findings and limitations from receipts, not authored rankings.	Workflow	unestimated u

Ladder verdict. NOT DECLARED. This board names no dated best-in-class or frontier target and no rung roles (sotatarget and rungrole rows on its plan); the ranker labels it NO-LADDER until it does, and until then its rungs climb toward a target nobody has written down. Bars read NO-BAR against the current month 2026-10.

Bar freshness, counted
Class	Count	Note
bars	0	sotabar rows on the plan
fresh	0	the bar month is the current month
attested	0	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	0	RECONCILED: the parts sum to the total 0
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.

Inspect a rung and its prerequisites

Declared nodes 12. Rank input binding: STALE_INCONSISTENT. Dependency order is authored. Implementation, acceptance evidence, authority and resource readiness are unverified. No action is recommended or dispatched here.

Plan SHA-256 40b32eb28db070399608d1e4e196096db763978732ead66058ed524bda7a691e. Target rows 0; role rows 0. Existing risks and release worklog retain their own scope; no node completion is inferred.

Runtime observation coverage: 0 of 12 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.

CE1 — Human-versus-AI NishiLang authoring joined to code health
Prerequisites: CE5 (acceptance unverified)
Contract: ce_author_health
Acceptance: THE KEYSTONE and the operator's own question. Join the authoring record (nx_claude_harvest frames, which carry which seat authored a change) to the health rulers (nx_mccabe complexity, nx_forkgrade duplication, nx_slopmeter self-repetition) full population, splitting each metric by human-seat, AI-seat and generated authorship. ACCEPT RULE declared in advance: FIRST measure and publish the AUTHORSHIP-MARKER ADOPTION across the corpus, because a split is only as good as the marker, and ship an UNKNOWN bucket rather than a split that silently miscounts; the field bar is GitClear's 81 percent duplication rise and 21 to 3.8 percent refactor fall and CodeScene's 60 percent higher AI defect risk in unhealthy code, so the organ must emit our own trajectory number with its horizon beside it -- a single census is a level and a trend needs two in time, reported UNOBSERVABLE until then
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.
CE2 — First-try build success and the build-rework rate
Prerequisites: None declared; this does not establish execution eligibility.
Contract: ce_build_firsttry
Acceptance: The build pillar has no rate. Read organ_ship.jrnl and the drift census and compute the first-compile-pass rate and the 14-day build-rework rate, where a byte decrease after an addition is a revert. ACCEPT RULE: two consecutive beats emit both rates whose arithmetic reconciles against organ_ship.jrnl row counts, and a single-sample window reads UNOBSERVABLE rather than a fabricated rate; a neg-control of a known revert must raise the rework count by exactly 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.
CE3 — Bug-resolution effectiveness as recurrence-rate reduction
Prerequisites: None declared; this does not establish execution eligibility.
Contract: ce_recurrence
Acceptance: Resolution effectiveness is not a close count -- the estate's own record has the same defect filed five times in three weeks. Compose nx_failclass and the debt board to publish recurrence per defect class before and after a fix, so a resolved class that recurs raises its own count. ACCEPT RULE: a planted recurring class must show a non-decreasing count across two beats, a genuinely resolved class must fall, and the measure must name the prior filing ids rather than a bare number
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.
CE4 — A board effectiveness claim is re-measured against a live verdict
Prerequisites: None declared; this does not establish execution eligibility.
Contract: ce_claim_binding
Acceptance: Extend the incumbent binding vocabulary (nx_swcompare_sota already re-measures symbol presence and classifies CONSISTENT, STALE-UNDER, UNBACKED) with gate=<gate> and catalog=<name> whose verdict is read against the LIVE artifact, so a row asserting a capability the verdict denies REFUSES to publish and one asserting an absence the verdict contradicts publishes STALE-UNDER. ACCEPT RULE: the motivating case is the fixture -- a claim that nx_mccabe SHIPPED must publish STALE-UNDER while its catalog verdict reads SOURCE-ONLY, and flip to CONSISTENT only once CE5 lands; extend the incumbent, never add a second ruler. OPERATOR REQUIREMENT 2026-09-07, not claimed enforced or complete: the estate must apply the same evidence admission to human and AI claims on /compare and wiki. Each functionality claim binds the exact artifact hash/version, test or experiment version, inputs, environment, dated result and provenance; an independent reviewer must approve it, and the author cannot self-approve. Missing or expired evidence remains UNKNOWN or STALE, never silently current; preserve failure evidence and an immutable additive correction/retraction history. Refuse unsupported claim upgrades and prove the refusal on planted unsupported, stale, version-mismatched and self-approved fixtures. Bounded passing tests establish only the tested scope and do not prove universal truth; render the scope and remaining uncertainty beside the claim
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.
CE5 — Close the complexity-keystone gap on the serving tree
Prerequisites: None declared; this does not establish execution eligibility.
Contract: ce_mccabe_live
Acceptance: codequality asserts the cyclomatic-complexity keystone SHIPPED with its gate GREEN 17 of 17, while nx_catalog reads nx_mccabe and nx_mccabe_gate SOURCE-ONLY -- built nowhere on the NAS. Build both, prove the gate GREEN against the live binary, promote, register, and re-run the catalog. ACCEPT RULE: nx_catalog nx_mccabe reads LIVE and nx_mccabe_gate returns GREEN against the promoted binary, and codequality's CQ1 row reads CONSISTENT under CE4's binding rather than a hand-asserted SHIPPED; the complexity distribution is republished full population with its histogram summing exactly
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.
CE6 — Every effectiveness number carries its interval and a null control
Prerequisites: None declared; this does not establish execution eligibility.
Contract: ce_number_envelope
Acceptance: A replay script that never observes the work matched frontier agents on static benchmarks, so a bare percentage is not evidence. Put n, an interval and a null-control value in the gate base class so this board and the others read one ruler and the emit REFUSES a bare percentage. ACCEPT RULE: a figure without n and an interval does not publish, and a planted bare percentage is refused by name
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.
CE7 — The autonomous fix loop registered and proven on one real task
Prerequisites: CE6 (acceptance unverified)
Contract: ce_fixloop_live
Acceptance: The loop is BUILT and its grader PROMOTED-UNREGISTERED. Register it and resolve one real debt-board task end to end -- localise from the grader rows, propose, verify by the byte-exact judge on the SWE-bench contract, revert proven exact. ACCEPT RULE: one episode resolves a real task with its RED-before-and-GREEN-after receipt banked and a null control (empty patch, revert patch, echo) scoring zero beside it
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.
CE8 — Duplication trend not just level
Prerequisites: None declared; this does not establish execution eligibility.
Contract: fg_trend
Acceptance: Shared with codequality -- one census is a level and the field bar is a trend (block duplication up 81 percent). The organ appends to a durable log on a beat and refuses to emit a trend from a single sample. ACCEPT RULE: two beats reconcile and a single sample reads UNOBSERVABLE; do not add a second duplication ruler, extend nx_forkgrade
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.
CE9 — The gauge is a beat with an axis-blind alarm so RED can never go dark
Prerequisites: None declared; this does not establish execution eligibility.
Contract: ST_GAUGE_STAMP
Acceptance: nx_seat shift is the gauge and it reads RED now; the risk is that a stale gauge reads as no movement rather than as blindness. A beat stamps the gauge and SessionStart prints its age, and an abstaining axis raises an axis-blind alarm. ACCEPT RULE: a stale gauge reads STALE on this page and never as zero, and a forced-stale fixture raises the alarm
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.
CE10 — A mirror whose canonical URL differs from the cited URL reads MOVED, never PROVEN
Prerequisites: None declared; this does not establish execution eligibility.
Contract: cr_canonical_moved
Acceptance: FOUND BUILDING THIS BOARD: the Faros /ai-impact page every other board cites now redirects to the Faros homepage, and a fresh fetch returned a 126648-byte page carrying none of the cited terms while printing status=200 -- the fetcher flagged it with a trailing 1 in its PROVENANCE line and no consumer reads that flag, so a moved page can be pinned as evidence. Extend nx_compare_refs_gate (the incumbent, never a second ruler) with a canonical-URL check: read the mirror's rel=canonical or og:url, normalise scheme, host and trailing slash, and classify MOVED when it differs from the cited URL, rendering it with the same drift classes as STALE-UNDER. ACCEPT RULE declared in advance: the Faros homepage mirror rf_ff06ac9a.raw cited under its old URL must read MOVED, the same page cited under its canonical must read PROVEN, an arxiv abstract whose canonical is itself must read PROVEN, a mirror with no canonical tag must read UNKNOWN never MOVED, and the fetcher's own trailing flag must be consumed and named in the verdict; bite-proven both directions before the tooth is armed
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.
CE11 — A gv_kv that republishes a quantity a gv_check_eq already asserted is a SECOND RULER
Prerequisites: CE6 (acceptance unverified)
Contract: gl_l5_double_publish
Acceptance: MEASURED 2026-09-06, and this rung exists because the defect was found in a shipping gate rather than imagined: nx_uxf_mp4box_gate passed 8 of 8 GREEN while publishing toplevel_box_count=0 against its own tooth T2b asserting actual=2 expected=2. One cnt buffer is filled by the good-tree walk, the tooth tests cnt[0] and passes on 2, the neg-control then re-walks into THAT SAME out-parameter leaving 0, and the gv_kv emission runs last -- so the published number was the neg-control's answer wearing the tooth's name, and a buffer restore two lines later could not reach it because a restore undoes the INPUT and not the DERIVED state. THE GENERAL LAW: gv_check_eq binds the number tested to the number published ONLY for values passed INTO it, and a separate gv_kv that re-reads a mutable buffer afterwards carries no such binding, which makes it a second ruler over a quantity the gate has already ruled on. The signal is mechanical and greppable -- the same variable expression appearing both as a gv_check_eq argument and as a gv_kv argument within one gate. EXTEND THE INCUMBENT AND DO NOT ADD A RULER: this is an L5 axis on nx_gatelaw_gate beside L1 exit-carries-verdict, L2 names-a-neg-control, L3 base-class and L4 emits-values, and it is DECLARED SEPARATE from all four because a gate can satisfy every one of them and still double-publish, so folding it in would silently redefine a figure other lanes act on. nx_capsearch over 7551 surfaces with corpus_complete=1 returned no incumbent detector, so the gap is measured rather than assumed. ACCEPT RULE DECLARED IN ADVANCE: the axis must FIRE on the pre-fix nx_uxf_mp4box_gate source as a planted positive and must NOT fire on its fixed source, it must measure and publish its false-positive rate against hand-adjudicated hits before any count is published as a defect count, and a gate whose dialect it cannot read must be left UNSCORED on this axis rather than asserted clean, because a ruler that declares a stream unreadable must not then assert anything about it on another axis
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.
CE12 — Accepted task, workflow and project outcomes for humans, AI and their partnership
Prerequisites: CE1 (acceptance unverified), CE2 (acceptance unverified), CE3 (acceptance unverified), CE6 (acceptance unverified), CE7 (acceptance unverified)
Contract: _ABSENT_:ce_accepted_outcomes_accept
Acceptance: PLANNED CONTRACT 2026-09-30. Extend existing authorship, build-first-try, recurrence, number-envelope, fix-loop and resource instruments. Compare matched human-led, AI-led within authority, and human-plus-Nishi assistance conditions with exact tools/model/policy versions and equivalent tasks/resources; mechanistic help is an explicit treatment. Report human active time, elapsed delivery time, review/rework, quality, retained features, escaped defects/recurrence, resource and actual token/cost demand per accepted outcome. Keep task/workflow/project denominators separate; retain mixed/unknown attribution, unfinished/censored work, selection effects, concurrency and uncertainty. Independent acceptance is fixed before candidate work; speed cannot offset correctness/safety/capability loss. Establish efficiency baselines first, then representative studies. No Nishi result is claimed until executed. Compare emits the findings and limitations from receipts, not authored rankings.
Authored effort: unestimated. Executor kind: Workflow. 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 codeeffectiveness 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.

CE1 Human-versus-AI NishiLang authoring joined to code health · cell LANDED · gate NO-GATE-DECLARED

Contract	`ce_author_health` · executor Organ · 2.5 u · stands on CE5
Definition of done	THE KEYSTONE and the operator's own question. Join the authoring record (nx_claude_harvest frames, which carry which seat authored a change) to the health rulers (nx_mccabe complexity, nx_forkgrade duplication, nx_slopmeter self-repetition) full population, splitting each metric by human-seat, AI-seat and generated authorship. ACCEPT RULE declared in advance: FIRST measure and publish the AUTHORSHIP-MARKER ADOPTION across the corpus, because a split is only as good as the marker, and ship an UNKNOWN bucket rather than a split that silently miscounts; the field bar is GitClear's 81 percent duplication rise and 21 to 3.8 percent refactor fall and CodeScene's 60 percent higher AI defect risk in unhealthy code, so the organ must emit our own trajectory number with its horizon beside it -- a single census is a level and a trend needs two in time, reported UNOBSERVABLE until then
Matrix cell	AUTHORING: human-versus-AI NishiLang authoring joined to code health · `runtime/_hdl_build/nx_claude_harvest.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_claude_harvest`; declare `nx_claude_harvest TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-05 queue next rung on the critical path (CE5 done, CE1 next, then CE4): the marker-adoption census -- classify every runtime source by its head marker (the GENERATED idiom nx_unwired already reads) and join per-file cyclomatic complexity from mc_scan_file, print the partition with sums and a time-spine row; uw_headmark lives inside a program, so the first step is an extraction to a lib (nx_oo_extract) rather than a second head scanner 2026-09-05 queue DESIGN SETTLED FROM THE INCUMBENTS, not from taste: the contract is the AUTHORING watch row (symbol ce_author_health on runtime/_hdl_build/nx_claude_harvest.nx). Leg 1 GROWS THE INCUMBENT instead of adding a second scanner -- nx_mccabe_lib's mc_scan_file already reads every source once, so the head-marker class becomes one call on the buffer it holds, with per-class blocks (files, funcs, sum vG, above the cited bar) in the same c[] the census and its gate share; the class table is DATA at knowledge/headmark_classes.conf (landed: BUILDER, GENERATED, DERIVED; UNMARKED is the named remainder), nx_mccabe prints the partition with sums plus one time-spine row, nx_mccabe_gate gains planted fixtures per class. The head scanner is uw_headmark inside nx_unwired (a program), so nx_oo_extract moves it to nx_headmark_lib FIRST and the lib edit imports it SECOND -- landing the edit before the extraction would leave a shared source that does not compile. Leg 2 joins nx_claude_harvest's session attribution (which files a seat wrote) to those per-file figures and ships ce_author_health. spendgate REVIEW name-free (nx_adopt counts callers, nx_mccabe measures complexity; neither classifies by marker). Both legs wait on build admission: every build this hour was refused (procs_blocked 15 then 10, load 23-30 on 8 cpus); the refs gate build sits absorbed as BQ-nx_compare_refs_gate status queued 2026-09-05 lesson the first nx_oo_extract of uw_headmark,uw_generated out of nx_unwired REVERTED (lib written at 630 B, consumer 28444 to 28000 with the import line, AFTER build at 1788640276, then both restored): the behavioural proof's workload scanned buildroot/runtime/_hdl_build, the directory that HOLDS the consumer's own source, so moving two definitions out of it legitimately changed the unwired census the proof compares. A BEHAVIOURAL PROOF WHOSE WORKLOAD READS THE CONSUMER'S OWN SOURCE REFUTES EVERY CORRECT EXTRACTION -- the subject must exclude the files the extraction rewrites. Re-run over /tmp/mccabe_gate (nine planted fixtures, neither consumer nor lib inside), launched through nx_job_run so the receipt lands in /tmp/oo_headmark1.txt where a dropped edge window cannot lose it 2026-09-05 measure second extraction attempt: OO-BASELINE captured (admission granted for the BEFORE build), the lib was written at 630 B, then the AFTER build was REFUSED by admission and the tool refused the extraction and restored everything -- the right direction. Its verdict line reads behaviour-NOT-preserved while its own preceding line says unverifiable-this-run: an admission refusal named as a behaviour change is a message naming the wrong stage, and a reader who trusts the verdict line would go looking for a real behavioural difference that never existed. Not my organ; named here for its owner. Under an oscillating box the extraction needs two admitted builds inside one run; third attempt launched through the job lane 2026-09-05 land third attempt EXTRACTED: uw_headmark,uw_generated moved to buildroot/runtime/nx_headmark_lib.nx (630 B), nx_unwired rebuilt and every workload run OUTPUT-IDENTICAL (behaviour preserved over the declared workload). nx_unwired stays BUILT-UNPROMOTED as it was; a refactor nobody asked to deploy is not promoted. The lib is the head-marker primitive the complexity scanner will import for the per-class join 2026-09-05 land LEG 1 LIVE -- the complexity scanner now joins every source's HEAD-MARKER CLASS to its complexity, inside the incumbent: nx_mccabe_lib gained the class table loader (mc_cls_load_from, DATA at knowledge/headmark_classes.conf, order is priority, UNMARKED is row 0), mc_headclass (composes uw_headmark from the extracted nx_headmark_lib), per-class blocks in the shared c[] (MC_C_SLOTS, files/funcs/sum vG/above-bar), and mc_scan_file classifies once from the buffer it already holds; a consumer that never loads a table sees classification off. nx_mccabe prints the partition with its sums and appends ONE row to knowledge/status/mccabe.spine for the production subject only (a trial dir announces SKIPPED -- the forgeable-heartbeat law). nx_mccabe_gate gained seven teeth on five planted fixtures under /tmp/mccabe_gate_cls with a PLANTED two-row table: two-row load, all five reached, GENERATED row carries vG 2, both-markers reads the FIRST row, a marker beyond the 400-byte head window does NOT classify (a whole-file grep dies there), partition sums to the walk, absent table writes no class block. Promoted debug-flavour after a non-debug build read as a flavour mismatch: gate 396fb82b 61648 B (24/24 GREEN live, class join rows=2 files=5 UNMARKED=2 BUILDER=2 GENERATED=1), census e447aba5 47848 B, both twins by hash, both rulers ONE-WAY-GAIN 2026-09-05 measure FIRST SPINE ROW, production tree, partition SUMS: files=19505 funcs=112660 above_10=14293 (126 permil). UNMARKED (seat-authored, human or AI at a seat) 17319 files / 107244 funcs / 14112 above = 132 permil; BUILDER (organs the builder signed) 246 / 599 / 92 = 154 permil; GENERATED 1939 / 4812 / 87 = 18 permil; DERIVED 1 / 5 / 2. Seat-authored code carries 987 permil of every function above the cited bar; generator output runs at one seventh the seat rate. The human-versus-AI split INSIDE the seat class is leg 2 (ce_author_health via nx_claude_harvest); this row is the denominator it joins against, and the next census run turns a level into a trajectory 2026-09-05 measure BITE-PROVEN: nx_gate_bite nx_mccabe_gate over nx_mccabe_lib.nx, 2 comparison sites -- site 0 NOT-REACHED (excluded, a mutation that never entered the binary), site 1 KILLED (the gate went RED on a compiling mutant), valid_mutants=1 killed=1, source restored byte-identical at 17135 B, both twins restored from the banked digests after the cleanup build was refused by admission (mutant_live=0). The production census printed partition SUMS and spine fd=4 wrote=158 of=158. One row worth a human read: DERIVED files=1 funcs=5 mean_vG=17 permille=400 -- a single NX-DERIVED .nx source in the runtime tree carrying two functions above the bar, i.e. a generated artefact that is also compiled source; it is named by class now and was invisible before 2026-09-05 queue LEG 2 PRECONDITION MEASURED, NOT ASSUMED: ce_author_health needs the set of files each seat session wrote, and the NAS actlog frames do not carry it. Every harvested write frame in knowledge/status/chbeat/chbeat_20260905.jrnl reads ts TAB session TAB tool TAB use TAB - TAB delta (735 nx_fs_write frames today, plus Write and Edit frames, none with a path); the path exists only in the transcript argv on the laptop, which the frame parser ch_frames drops. The join cannot be faked from data that lacks its key. BUILD ORDER: (1) an ADDITIVE note field on write frames -- ch_frames extracts the path argument of nx_fs_write, Write and Edit tool uses and writes note=path=<p> (existing readers see one more token, no field moves); (2) the laptop beat pushes the new frames as it already does, so the NAS record fills forward from the day it lands; (3) ce_author_health in nx_claude_harvest walks the per-day journals for path-bearing write frames, keys each path to its session, scans it with mc_scan_file, and prints seat-written versus not-seat-written figures against the mccabe spine denominator, with the unattributed remainder named. A seat is AI at a seat by definition of the transcript, so the human half is the complement inside UNMARKED, and the emitter must say that rather than claim a per-human attribution it does not have. Owner: the harvest is portable and its beat runs on the laptop, so the parser change ships through the laptop twin and the SSOT under _hdl_build together 2026-09-05 land LEG 2 PRECONDITION SHIPPED: nx_claude_harvest write frames now carry the path. The parser gained ch_tool_writer_kind (nx_fs_write -> argv[1], Write and Edit -> file_path), ch_json_str_copy, ch_tool_path (bounded 4096-byte window after the tool's name, first match is the input's own key) and ch_frame_note, and BOTH loops (frames and delta) compose it: the note gains one space-separated token `path=<p>`, no column moves, a non-writer frame is byte-identical to before. Three selftest teeth on a fresh fixture: T45 argv[1] path with the verb never mistaken for it, T46 file_path with a decoy key inside the content never read as the path, T47 neg-control a reader frame carries no path token (exactly 2 of 3 do). NAS: contentdiff ONE-WAY-GAIN gained 17 lost 0, promoted 16e6fbfe (223208 B, debug flavour like live, prev banked), offc twin installed, LIVE selftest 53/53 GREEN. Laptop: the same seven edits applied to the twin, and its sovereign build came out BYTE-IDENTICAL to the NAS non-debug build (198467 B, sha eabec821...) -- the closure equality proof the estate's own ruler gives for free. The laptop beat forks the laptop _offc twin, refreshed through the builder's --install lane, so every beat from now lands path-bearing frames and the NAS record fills forward; frames before this hour stay pathless and the join must name them as UNATTRIBUTED-BEFORE-<epoch>, never guess them 2026-09-05 land LEG 2 LIVE -- ce_author_health is DECLARED in nx_claude_harvest and called by its authorhealth verb (the watch's symbol ruler applies the declaration rule, so this is what flips the AUTHORING cell). It reads a per-day journal, keeps every UNIQUE path from path-bearing write frames (a duplicate write counts once), scans each .nx under the tree prefix ONCE with mc_scan_file (the one ruler the census and its gate share), sets the figures against the mccabe spine's LAST row, and names the complement not_seat_attributed (human-authored PLUS every write before the first path frame). Three named refusals: UNOBSERVABLE-DENOMINATOR (rc 3, no spine), NO-PATH-FRAMES (rc 4), and a path outside the tree or not .nx counted apart. Teeth T48-T51 on a planted tree, journal and spine (dedupe, partition, absent spine, pathless journal). Built first on the laptop, where the compiler caught consts read above their declaration (single-pass parser; consts moved to the top beside the beat consts) and the selftest caught sj_atoi_z concatenating every digit up to the NUL (files=10 2026-09-05 measure FIRST AUTHOR-HEALTH MEASUREMENT (chbeat_20260905.jrnl, frames=12606, path_frames=4 in the hour since the parser landed, sessions=4): seat_written files=3 funcs=102 sum_vG=777 above_10=19 (186 permil) against the spine denominator files=19505 funcs=112660 above_10=14293; not_seat_attributed files=19502 funcs=112558 above_10=14274 = human-authored PLUS every write before ts=1788647757, UNATTRIBUTED by construction. This is the first hour of a record that fills forward on every beat; the number is honest and small, and it becomes a trajectory as the beats accumulate. The three in-tree files were nx_wgsl.nx, nx_manga_script_gate.nx and nx_content_put_client.nx (three other seats); the fourth path was this board's .plan, outside the tree 2026-09-05 retract the probe disagreements are ATTRIBUTED and FIXED, superseding the earlier UNATTRIBUTED rows: nx_behaveprobe read this organ's staged selftest RED (rc 1) twice while every direct run was GREEN. Mechanism, read from the source: ch_selftest stamped its scratch names with sys_now_realtime_sec while the beat already stamps its pieces with sys_now_realtime_us for exactly this reason; the probe runs live then staged inside one second, so the staged run found the live run's T15 marker already seeded and its T14 refused. One-line fix (the selftest now stamps in microseconds), proven by the instrument that had disagreed: contentdiff IDENTICAL-CONTENT, nx_behaveprobe live_rc=0 other_rc=0 on the same argv that had read 0/1. A SELFTEST WHOSE FIXTURE NAMES ARE UNIQUE PER SECOND IS NOT UNIQUE PER RUN UNDER ANY CALLER THAT RUNS IT TWICE, and the ship probe is such a caller. Promoted c5534bd2 (257552 B, prev banked); laptop twin rebuilt and installed (229678 B non-debug 8de305b8) 2026-09-05 land WATCH FLIP WITNESSED on the comparewatch plane (job 1788649580, 306644 B): codeeffectiveness_ce_author_health LANDED and codeeffectiveness_ce_mccabe_live LANDED after the beat regen that held lease regen-1788648455; the published page follows on its next publish. PRESENCE ONLY: no referee grade and no ACCEPT row yet, so CE1 and CE5 stay INCOMPLETE on /compare under the three-signature law. 2026-09-05 measure FIRST TRAJECTORY, two spine rows 101 minutes apart (1788643308 -> 1788649356): files 19505 -> 19511 (+6), funcs 112660 -> 112818 (+158), above_10 14293 -> 14309 (+16), permille 126 -> 126. Every delta sits in UNMARKED (17319 -> 17325 files, 107244 -> 107402 funcs, 14112 -> 14128 above); BUILDER, GENERATED and DERIVED are byte-for-byte unchanged. Sixteen new over-bar functions in an hour and a half of seat work is the number this board exists to watch, and this seat's own ce_author_health is a candidate among them; the next census names them by file through the worklist

CE2 First-try build success and the build-rework rate · cell LANDED · gate UNRECORDED

Contract	`ce_build_firsttry` · executor Organ · 1.5 u · stands on nothing
Definition of done	The build pillar has no rate. Read organ_ship.jrnl and the drift census and compute the first-compile-pass rate and the 14-day build-rework rate, where a byte decrease after an addition is a revert. ACCEPT RULE: two consecutive beats emit both rates whose arithmetic reconciles against organ_ship.jrnl row counts, and a single-sample window reads UNOBSERVABLE rather than a fabricated rate; a neg-control of a known revert must raise the rework count by exactly one
Matrix cell	BUILD: first-try build success and the build-rework rate · `runtime/nx_organ_ship.nx` · LANDED decl
Gate	`nx_organ_ship_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_organ_ship_behavior_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	2026-09-06 land nx_build_firsttry_lib (16,617 B) is the ONE ruler: episodes classified by their FIRST BUILD row into first_pass, code_fail and box_refused (partition printed), 14-day rework is a same-target re-ship inside the window, a revert is a bytes= decrease, a pair lacking bytes= is revert_unobservable (counted, never guessed), an empty window reads UNOBSERVABLE never 0 permil, one spine row per MEASURED run at a path derived from the journal; nx_build_firsttry_gate 33/33 GREEN on six planted journals with hand-computed answers, the known revert moves rework AND reverts by exactly one, two bites proven, built 55,012 B. nx_organ_ship took six CAS edits (63,149 to 65,952 B, every predicted size exact): the import, os_build_refusal_class plus os_refusal_head plus os_bytes_of, a BUILD row journaled on an /api/build refusal (class plus a 160-byte reason head) and on WRONG-SRC, bytes= on the SHIPPED row, and the firsttry verb ce_build_firsttry. Compile-proof of the edited ship loop in flight; then contentdiff, promote, and firsttry over the production journal for the first spine row 2026-09-06 land CE2 SHIPPED AND PROVEN ON THE SERVED ARTIFACT 2026-09-06: nx_organ_ship compile-proof GREEN on the sovereign lane (122,520 B sha 19e6897c, debug flavour following the incumbent); /api/build REFUSED-and-queued twice on the D-state witness, so the stage went through nx_filecopy (a cap minted async through the mgmt API) into the promote slot; nx_contentdiff live-vs-staged GREEN lost_from_live=0 ONE-WAY-GAIN (86 runs gained: the firsttry literals and the REFUSED-ADMIT class); PROMOTED with expect_sha256, prev banked, _offc installed. nx_build_firsttry_gate PROMOTED 3fd7ff39 (55,012 B), REGISTERED, served copy 33/33 GREEN. nx_forkgrade_gate PROMOTED 4bae5923 (43,824 B), REGISTERED, served copy 12/12 GREEN against the served nx_forkgrade.elf -- CE8's gate is shipped too. FIRST PRODUCTION BEAT of the rule (nx_job_run nx_organ_ship firsttry; spine knowledge/status/build_firsttry.spine row ts=1788675542): rows=2572 rows_in_window=2091 targets=208 episodes=253 first_pass=253 code_fail=0 box_refused=0 retries=119 shipped=242 rework=96 rework_permil=396 reverts=0 revert_unobservable=96 verdict=MEASURED 2026-09-06 retract The first beat's first_try_permil=1000 is a BARE PERCENTAGE, not a measurement: 119 retries with ZERO recorded failures means the journal never carried the negative class before today's ship-loop edit (it journaled BUILT rows only). The ruler now abstains by name -- retries with no failure class reads UNOBSERVABLE with neg_class_absent=1 -- source landed in nx_build_firsttry_lib, rebuild and re-promote owed. The rework rate (96 of 242 ships re-shipped inside 14 days, 396 permil) IS a measurement; reverts become observable as bytes= rows accumulate. Second beat owed by a daily clock row so two beats reconcile per the accept rule 2026-09-06 land Second-beat plumbing and the abstention landed 2026-09-06: clock row firsttrybeat 86400 nx_organ_ship.elf firsttry declared (CLOCKJOB-PUT ok rows=174, desired state, reconciled at the clock's next window, storm-deferred like every clock row); nx_build_firsttry_lib grew BFT_R_NEG_CLASS_ABSENT -- a window with retries and not one recorded failure sets first_try UNOBSERVABLE and neg_class_absent=1, printed on the report line and carried on the spine row (16,617 to 17,444 B, four CAS edits, one 503 that had landed caught by the guard); nx_build_firsttry_gate grew fixture F6 (a retry with no failure row, the pre-fix journal shape) with five teeth and a bite over the bare-1000 refusal, F4 as the present-class control (12,863 to 13,238 B plus the teeth). Gate rebuilt on the sovereign lane; ship-loop rebuild in flight; stage, promote and a served re-run of firsttry follow so spine row 2 reads the abstention honestly 2026-09-06 land The abstention is SHIPPED on the gate side 2026-09-06: nx_build_firsttry_gate rebuilt 64,768 B sha 637a0146 (debug flavour following the incumbent), staged via nx_filecopy, PROMOTED with expect_sha256, _offc installed, and the SERVED copy reads 39/39 GREEN -- the six F6 teeth pass (a retry with no failure row: one retry counted, first_try UNOBSERVABLE never a bare 1000, neg_class_absent=1, rework still MEASURED, F4 present-class control 0, bite over the bare-1000 refusal fires-on-bad silent-on-good) and its production-spine neg-control now reads the real 175-byte knowledge/status/build_firsttry.spine unchanged before and after the fixture runs. Ship loop rebuilt 123,056 B sha f24b730e; nx_contentdiff live-vs-staged reads RED at 2 permil with exactly ONE lost run named -- the old parenthetical (box refusals excluded: the box, not the code), superseded by the longer sentence the candidate GAINS beside the neg_class_absent= fields -- a reworded message, not a capability, so the promote proceeds under the calibrated guard with that run named here 2026-09-06 measure SECOND PRODUCTION BEAT on the served ship loop (PROMOTED 123,056 B sha f24b730e under the calibrated guard, prev banked, _offc installed): spine row 1788676308 reads episodes=253 first_pass=253 code_fail=0 box_refused=0 retries=119 first_try_permil=-1 first_try_state=UNOBSERVABLE neg_class_absent=1 shipped=242 rework=96 rework_permil=396 reverts=0 revert_unobservable=96 verdict=MEASURED. Two beats on knowledge/status/build_firsttry.spine reconcile on every count against the same 2,091 in-window rows of organ_ship.jrnl; the accept rule's mechanical half (two beats reconciling, a single-sample window UNOBSERVABLE, a known revert raising rework by exactly one) is met on the production journal and the gate. The daily clock row carries the beat from here. Still owed for CE2: the first-try rate becomes a measurement only after refusal rows accumulate in the journal (the negative class now journaled), referee grade, operator ACCEPT

CE3 Bug-resolution effectiveness as recurrence-rate reduction · cell LANDED · gate NO-GATE-DECLARED

Contract	`ce_recurrence` · executor Organ · 1.5 u · stands on nothing
Definition of done	Resolution effectiveness is not a close count -- the estate's own record has the same defect filed five times in three weeks. Compose nx_failclass and the debt board to publish recurrence per defect class before and after a fix, so a resolved class that recurs raises its own count. ACCEPT RULE: a planted recurring class must show a non-decreasing count across two beats, a genuinely resolved class must fall, and the measure must name the prior filing ids rather than a bare number
Matrix cell	BUGRES: recurrence rate as the real bug-resolution effectiveness measure · `runtime/_hdl_build/nx_debtmine.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_debtmine`; declare `nx_debtmine TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 land DESIGN LANDED AS SOURCE 2026-09-06, composed into the incumbent: nx_capsearch found no recurrence ruler but nx_debtmine (REGISTERED bugs-class miner: per-scope open/eaten, eat-rate, aging) already owns the class axis, so ce_recurrence extends it instead of a second ruler and the matrix organ path is repointed from runtime/nx_failclass.nx to runtime/_hdl_build/nx_debtmine.nx by declaration. nx_debt_recurrence_lib.nx (new): per scope the FIRST EATEN filing is the prior; every row filed after it is a recurrence (recur_total = history, non-decreasing by construction; recur_open = the live re-filings, which fall when eaten); the prior id and a bounded prefix of the recurring ids are NAMED beside their total; close time is UNRECORDED on this plane (ids are file-time) so the document says bound=UPPER; the spine (knowledge/status/debt_recurrence.spine, or a <source>.recurrence.spine sidecar for any non-live source) is written only by the recurrence verb and only when rows were measured, announced inside the JSON. nx_debt_recurrence_gate.nx (new, in-process): three planted planes -- F1 alpha eaten-then-open, beta open-open, gamma eaten-eaten, one malformed row; F2 = F1 plus a third alpha filing (non-decreasing across two beats); F3 = F1 with the re-filing eaten (the resolved class falls) -- ids named in the JSON, beta never counts, spine sidecar two rows, production spine unchanged, one bite. nx_debtmine edits: the import landed; the recurrence object inside mine and the recurrence verb (ce_recurrence) are the next two CAS edits, then build, gate, promote, and the first two beats on the live plane 2026-09-06 land ce_recurrence SHIPPED: nx_debt_recurrence_lib is the ONE ruler (per scope the first EATEN filing is the prior and every later filing a recurrence; recur_total is non-decreasing history, recur_open is live and falls when eaten; ids named; close_time UNRECORDED so the bound is declared UPPER), composed into the incumbent nx_debtmine as a recurrence verb plus a recurrence object inside mine (import 23,590 to verb 27,139 B, four CAS edits, every predicted size exact). Gate nx_debt_recurrence_gate 24/24 GREEN and bite-proven (fires on the refiled class, silent on the never-resolved one), values emitted for independent adjudication; PROMOTED 56,209 B c8a8de62, REGISTERED callable=1, _offc refreshed. Miner PROMOTED 80,004 B 76f7295f after nx_contentdiff GREEN lost_from_live=0 gained=104 (strict superset). Spine knowledge/status/debt_recurrence.spine is written only by the recurrence verb on the production prefix and only when rows>0; the first two live beats follow this row. 2026-09-06 retract beat 1 (spine row 1788692120: scopes=256 with_recurrence=68 recur_open=401 recur_total=630) was a FLOOR, not a total: the scope table was a 256-slot bound and it bound on the 4,353-row plane. The document printed scopes_capped=1 and the seat read past it one line later. Superseded by beat 2 below; the spine now carries capped= on every row so the retraction is machine-readable. 2026-09-06 measure beat 2 on the promoted miner with the row-derived capacity (nx_debtmine 91,432 B 501df732, lib 15,123 B): rows=4353 malformed=66 scopes=2305 scopes_capped=0 scopes_with_recurrence=251 recur_open=889 recur_total=1528 id_bursts_adjacent=133 coverage_complete=1. The cap had hidden 889 permil of the scopes and 588 permil of the recurrence. Top recurring classes by open re-filings: stale-mirror 98 open of 117 (prior 1784654222), ivv-drift 41 of 41, gx-body 27 of 30, discoverability 25 of 26, search 23 of 29, atlas 22 of 37, control-plane 22 of 24. The ids repeating inside stale-mirror are an autofile burst (three rows 313 bytes apart sharing one epoch-second), distinct filings counted as such and declared by the burst field. close_time is UNRECORDED on the plane so every figure is an UPPER bound. 2026-09-06 lesson A CAP THE DOCUMENT ANNOUNCES IS STILL READ AS A TOTAL BY THE NEXT LINE. The remedy was not a bigger number: capacity is derived from the rows (distinct scopes cannot exceed rows) so the bound cannot exist, and the gate proves it with an F4 plane of 300 scopes the old table would have cut (gate 33/33, capacity 603 over 602 rows). The same 256-slot table lived in the incumbent mine census (DM_SCOPES) and was fixed in the same edit: a fix that lives in one ruler and not its sibling is half a fix. Also caught: the recurrence object composed into the mine document made a whole-document absence tooth (T5) fail on the promoted binary; the tooth now names the census shape it tests. 2026-09-06 land CE3 SHIPPED ON THE LIVE PLANE: nx_debtmine 91,496 B 393f80e8 PROMOTED (recurrence verb, recurrence object inside mine, row-derived scope capacity for BOTH rulers, T5 tooth scoped to the mine census) with _offc refreshed and selftest 10/10 GREEN on the promoted binary; nx_debt_recurrence_lib 15,123 B (drc_capacity, cap parameter, adjacent id-burst counter, capped= on every spine row); nx_debt_recurrence_gate 33/33 GREEN bite-proven (F4 300-scope plane counted in full), PROMOTED 68,888 B da4768c5, REGISTERED, cap minted 30 d, _offc refreshed. The live mine census now reads scopes=2305 scopes_capped=0 scopes_below_minsup=606 and names codewiki-debt 28 open, render-docs 25, atlas-consolidation 14, wiring-guard 13 -- scopes the capped table never interned. Spine knowledge/status/debt_recurrence.spine: row 1788692120 (capped, retracted) then beat 2 (capped=0); beat 3 follows so two comparable rows exist. Clock row debtrecurbeat 86400 declared. Shipped and gate-proven is NOT complete: referee grade and operator ACCEPT are owed before this rung reads done. 2026-09-06 measure nx_compare_rank after the CE3 ship: rungs=10 done=8 open=2 (CE7 ce_fixloop_live priority 300 critical path; CE9 sm_gauge_alarm priority 100), unmapped=0, finish=0, rank artefact 883 B, plane comparerank-codeeffectiveness published. Clock row debtrecurbeat landed (CLOCKJOB-PUT rows=175, the verb and its args ride in the organ field). The recurrence gate reads REGISTERED-DARK on nx_catalog until the MCP surface discovers it on its next sweep (it was registered today; nx_job_run forks log the runner, not the organ).

CE4 A board effectiveness claim is re-measured against a live verdict · cell LANDED · gate NO-GATE-DECLARED

Contract	`ce_claim_binding` · executor Organ · 2 u · stands on nothing
Definition of done	Extend the incumbent binding vocabulary (nx_swcompare_sota already re-measures symbol presence and classifies CONSISTENT, STALE-UNDER, UNBACKED) with gate=<gate> and catalog=<name> whose verdict is read against the LIVE artifact, so a row asserting a capability the verdict denies REFUSES to publish and one asserting an absence the verdict contradicts publishes STALE-UNDER. ACCEPT RULE: the motivating case is the fixture -- a claim that nx_mccabe SHIPPED must publish STALE-UNDER while its catalog verdict reads SOURCE-ONLY, and flip to CONSISTENT only once CE5 lands; extend the incumbent, never add a second ruler. OPERATOR REQUIREMENT 2026-09-07, not claimed enforced or complete: the estate must apply the same evidence admission to human and AI claims on /compare and wiki. Each functionality claim binds the exact artifact hash/version, test or experiment version, inputs, environment, dated result and provenance; an independent reviewer must approve it, and the author cannot self-approve. Missing or expired evidence remains UNKNOWN or STALE, never silently current; preserve failure evidence and an immutable additive correction/retraction history. Refuse unsupported claim upgrades and prove the refusal on planted unsupported, stale, version-mismatched and self-approved fixtures. Bounded passing tests establish only the tested scope and do not prove universal truth; render the scope and remaining uncertainty beside the claim
Matrix cell	RIGOR: a board effectiveness claim is re-measured against a live verdict, never hand-asserted · `runtime/nx_symdecl_lib.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_symdecl_lib`; declare `nx_symdecl_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 queue PUBLISH STORM-DEFERRED BY THE GOVERNOR, not by choice: two manual regens were refused -- first REFUSED-ALREADY-RUNNING (the compare beat held lease compare-regen, holder regen-1788654204, which started BEFORE the promote+restage and so forked the pre-fold _offc), then REFUSED-IO-STORM procs_blocked=7 ncpu=8 (the D-state witness at the storm line the build gate refuses on; the clock beat defers itself via clock_heavy.conf the same way). Every step ON the publish path is done and hash-verified (all three roots f7eaa106; CE4 matrix row repointed to cb_classify), so the UNSERVED fold and the LANDED watch publish on the first ADMITTED beat regen -- no seat action owed beyond re-issuing nishi_compare_regen once nx_build_admit reads GRANT. Do not launch a third regen into the storm. 2026-09-06 land CE4 PUBLISH PATH COMPLETE ACROSS ALL THREE ROOTS. nx_swcompare_matrix promoted f7eaa106 (serving root); nx_restage copied the PROMOTED binary into buildroot/_offc (the regen's fork target); all three roots now hash f7eaa106 396848, verified by nx_filehash (the pre-fold 4eeb7688 392952 is gone). So the compare regen publishes the UNSERVED fold, not the old generator. CE4 matrix row repointed runtime/nx_catalog.nx -> runtime/nx_symdecl_lib.nx 2026-09-06 land CE4 ADOPTION PROVEN END-TO-END, nx_sota_drift_gate 41/41 GREEN. Re-pointed gate PROMOTED (57952 B sha 8ca7be4e; the 6 CAPLOSS runs were all superseded -- old nx_swcompare_sota subject paths the gate no longer drives + the verdict note I extended, lost_srcpath=0, +127 runs gained, artifact GREW 42455->57952, so a forward change not a revert -- allow_capability_loss=yes named on that) and driven via nx_job_run against the freshly-built buildroot/_build/nx_swcompare_matrix.sov.elf (the fold, still unpromoted) on the /tmp fixture tree, ZERO live pages touched. NEW F-LEG ALL PASS: a shipped claim over a source-only organ reads UNSERVED (counted, named, html badge >UNSERVED</b>, json drift field), symbol-presence alone no longer blesses a shipped claim, a served organ stays CONSISTENT (C-again unserved=0, byte-identical), and unserved-fires-on-source-only-silent-on-served is BITE-PROVEN. This is the CE4 mechanism working on the surface that has the drift vocabulary. REMAINING (mechanical, proven-safe): promote nx_swcompare_matrix + nx_restage buildroot/_offc + nx_filehash all 3 roots; repoint CE4 matrix row runtime/nx_catalog.nx -> runtime/nx_symdecl_lib.nx 2026-09-06 land CE4 EMITTER ADOPTION WRITTEN AND COMPILE-PROVEN across three files via the sovereign lane (source-landed, both _build artifacts compile clean; headline promote deferred -- build-admit read load1_centi=1634 against a 1643 ceiling, 16 running, and the matrix generator is the ONE assembler for ~106 /compare pages: promoting it into a near-ceiling storm is the hammer-the-NAS anti-pattern, not sota shipping). WHAT LANDED: (1) nx_symdecl_lib cb_serve_state_cwd -- the served-artifact reader, CWD-relative the same way sd_present resolves source (probes ../<base>.elf then ./<base>.elf, the two CWDs the generators run under), so the two rulers never judge different subjects for one path. (2) nx_swcompare_sota_lib: at the bound m==1/claims==1 branch it now calls cb_serve_state_cwd and a shipped claim whose organ is SOURCE-ONLY reads a NEW class SL_DRIFT_UNSERVED (its own name/class/badge/reason 'built nowhere -- over-reports what RUNS', counter b_unserved, census+html+json), additive: a served organ stays CONSISTENT so lang's 20 bindings are byte-identical. (3) nx_sota_drift_gate: kind-6 UNSERVED fixture (source planted, served twin planted for the consistent case, unserved twin unlinked), F-leg asserting unserved fires on source-only and is silent on the served consistent row, partition includes unserved, plus a `<subject_elf>` argv override (the nx_gate_bite subject precedent) so a staged build is provable before promote. Built shas: gate be2326a7, matrix d967f84e. TURNKEY SHIP, next low-load window: (a) /api/build+promote nx_sota_drift_gate, verify staged carries 'F(unserved)'; (b) nx_job_run nx_sota_drift_gate /tmp/o buildroot/_build/nx_swcompare_matrix.sov.elf -- proves UNSERVED end-to-end WITHOUT touching live pages; (c) if GREEN, /api/build+promote nx_swcompare_matrix + nx_restage buildroot/_offc + nx_filehash all 3 roots; (d) repoint CE4 matrix row runtime/nx_catalog.nx -> runtime/nx_symdecl_lib.nx 2026-09-05 queue EMITTER ADOPTION MAPPED FROM SOURCE, turnkey (deferred under a MEASURED sustained storm: 4 consecutive edge-window 503s on READ calls 2026-09-05, so the multi-file generator edit is NOT made from guessed constants). (1) The drift classifier is INLINE and DUPLICATED per generator; the STALE-UNDER/UNBACKED/CONSISTENT vocabulary exists ONLY in nx_swcompare_sota (inline at the m==1 && claims==1 branch, ~line 285), while nx_swcompare_matrix flips WATCHING<->landed on bare symbol presence with NO drift classes. (2) codeeffectiveness is a .matrix, so CE4 on ITS OWN board needs the matrix generator taught the drift vocabulary FIRST (larger port); the CE4 MECHANISM lands cleanly on the sota generator today. PLAN, additive: add BD_DRIFT_UNSERVED to nx_swcompare_lib beside BD_DRIFT_STALE_UNDER (own bd_drift_name + bd_drift_class; matrix gen unaffected); in the sota classifier at m==1 && claims==1 call cb_serve_state(organ) (built+proven, nx_claim_binding_gate 12/12) and downgrade CONSISTENT->UNSERVED when SOURCE-ONLY, with a DISTINCT reason 'claimed shipped, built nowhere' -- NEVER reuse STALE_UNDER's 'graded No' text (names the wrong subject). Served organs stay CONSISTENT so lang's 20 bindings are byte-identical; add b_unserved counter + json key. VERIFY: extend nx_sota_drift_gate with a source-only-organ fixture asserting UNSERVED; ship sota via sovereign build + nx_restage + nx_filehash all 3 roots. THEN matrix-generator port, then repoint CE4 matrix row to runtime/nx_symdecl_lib.nx 2026-09-05 land CE4 RULER BUILT AND GATE-PROVEN via the sovereign lane (governor-free, storm-immune -- admission GRANTed load1_centi=1287). Added the served-artifact classifier to nx_symdecl_lib.nx beside sd_present: cb_serve_state(organ) reads SERVED/SOURCE-ONLY/ABSENT by stating the serving-tree binary and the source, and cb_classify(claim_present, state) returns CONSISTENT / STALE-UNDER (verdict contradicts, publish WITH badge) / REFUSE (verdict denies, do not publish) -- the exact ACCEPT wording. This is the ruler CE5 named: SYMBOL PRESENCE CANNOT SEE A PROMOTE, and sd_present passed the McCabe SHIPPED-over-SOURCE-ONLY claim because mc_scan_file was in source the whole time. nx_claim_binding_gate GREEN 12/12, 3 bites BITE-PROVEN incl the McCabe end-to-end shape, built sha d9f4be7a in _build only. NOT DONE, honest: CE4 ACCEPT requires the EMITTER to re-read claims ON PUBLISH. NEXT = wire the nx_swcompare_sota drift loop (its sd_present(organ,sym,rl2) call site near off 17788) to also consult cb_serve_state for =<organ> bindings and render STALE-UNDER/REFUSE, then repoint the CE4 matrix row from runtime/nx_catalog.nx to runtime/nx_symdecl_lib.nx

CE5 Close the complexity-keystone gap on the serving tree · cell LANDED · gate NO-GATE-DECLARED

Contract	`ce_mccabe_live` · executor Organ · 1.5 u · stands on nothing
Definition of done	codequality asserts the cyclomatic-complexity keystone SHIPPED with its gate GREEN 17 of 17, while nx_catalog reads nx_mccabe and nx_mccabe_gate SOURCE-ONLY -- built nowhere on the NAS. Build both, prove the gate GREEN against the live binary, promote, register, and re-run the catalog. ACCEPT RULE: nx_catalog nx_mccabe reads LIVE and nx_mccabe_gate returns GREEN against the promoted binary, and codequality's CQ1 row reads CONSISTENT under CE4's binding rather than a hand-asserted SHIPPED; the complexity distribution is republished full population with its histogram summing exactly
Matrix cell	RIGOR: the complexity keystone is BUILT AND PROMOTED on the serving tree, not merely present in source · `runtime/nx_mccabe.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_mccabe`; declare `nx_mccabe TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-05 measure THE FIRST FINDING, MEASURED: nx_catalog reads nx_mccabe SOURCE-ONLY 5249 B and nx_mccabe_gate SOURCE-ONLY 9677 B on the serving tree while /compare/codequality asserts the complexity keystone SHIPPED with its gate GREEN 17 of 17 -- an effectiveness claim unbacked on the surface that runs. This is why CE4 binds claims to live verdicts and CE5 closes the gap. 2026-09-05 land nx_mccabe BUILT through /api/build after the sovereign runner was refused by admission (VERDICT=QUEUE, run queue 22 on 8 CPUs, a GRANT sampled seconds earlier had already gone stale): 36760 B sha e2526b1a, src_path confirmed runtime/nx_mccabe.nx, src_stable 1, closure 7b23a7dd; PROMOTED prev=1 to nx_mccabe.elf under expect_sha256; /api/tools/register and nx_offc_install issued; nx_mccabe_gate build in flight through /api/build. The main-binary half of codequality's SHIPPED claim is now backed on the serving tree; the gate half follows and nx_catalog is the referee for both. 2026-09-05 measure nx_mccabe RUNS on the serving tree, full population 2026-09-05: files=19467 funcs=112258 above_cited_10=14247 (126 permille) p50=3 p90=13 p99=46 max=2996 mean=6 unreadable=0, and the histogram partition SUMS EXACTLY to 112258. Against the 2026-08-20 census (18879 files, 103733 funcs, above-10 12639 = 121 permille, p90=12, p99=43) this is the FIRST two-point complexity trajectory the estate has: the corpus grew 8 percent in functions and complexity crept up on every percentile -- p90 12 to 13, p99 43 to 46, above-bar 121 to 126 permille. A level said 121; two levels say RISING. Registered (callable, grammar harvested, reconcile issued) and the _offc twin installed. Gate build queued as BQ-nx_mccabe_gate behind the I/O storm. 2026-09-05 land THE KEYSTONE IS CLOSED ON THE SERVING TREE (2026-09-05). nx_mccabe_gate BUILT through /api/build on a GRANT sample (51464 B sha efce9054, src_path confirmed), PROMOTED prev=1, REGISTERED, _offc twin installed, and run against the PROMOTED nx_mccabe: NX-mccabe passed 17 of 17 verdict=GREEN with three BITE-PROVEN controls (comment trap, string-literal trap, above-bar fires on 13 and is silent on 1), walk files=9 funcs=9 above10=1 max=13. Both halves of codequality's SHIPPED claim are now backed by nx_catalog LIVE verdicts; the CE5 watch cell stays open by construction until CE4's catalog binding exists, and says so. 2026-09-05 land the watch row keyed on ce_mccabe_live, a symbol nobody had declared while the keystone was already promoted and writing its spine -- the same built-but-unnamed gap CE10 had. Named the real thing, not a badge: the spine-row writer in nx_mccabe (the block that appends the trajectory row for the production subject and announces fd= wrote= of=) is now func ce_mccabe_live(c, funcs, crows, ob, a), called by main; the census's own trajectory writer is what proves the keystone runs on the serving tree. Source landed (9922 B); build refused by admission while the beat regen holds the box, paced re-issue owed, then ruler, promote, twin. cr_canonical_moved already reads LANDED on the comparewatch plane; ce_author_health reads OPEN because the running regen re-put the plane before the declaration landed, and the next beat takes it 2026-09-05 land ce_mccabe_live LIVE: promoted 37d9ed61 (48232 B, contentdiff ONE-WAY-GAIN gained 1 = the symbol's own name, lost 0, prev banked), offc twin by hash, live run prints spine=knowledge/status/mccabe.spine fd=4 wrote=158 of=158 from the renamed writer and the partition SUMS (class_files 19511 of 19511, class_funcs 112818 of 112818). The keystone's cell flips on the next beat regen for the same reason the authoring cell does

CE6 Every effectiveness number carries its interval and a null control · cell LANDED · gate NO-GATE-DECLARED

Contract	`ce_number_envelope` · executor Organ · 1 u · stands on nothing
Definition of done	A replay script that never observes the work matched frontier agents on static benchmarks, so a bare percentage is not evidence. Put n, an interval and a null-control value in the gate base class so this board and the others read one ruler and the emit REFUSES a bare percentage. ACCEPT RULE: a figure without n and an interval does not publish, and a planted bare percentage is refused by name
Matrix cell	RIGOR: every published effectiveness number carries its interval and a null control · `runtime/nx_gate_verdict.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_gate_verdict`; declare `nx_gate_verdict TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 land ce_number_envelope, ce_env_refuse and the CE_ENV_* codes landed in the gate base class nx_gate_verdict.nx (38,635 to 40,835 B, exactly one insertion by size arithmetic): a figure publishes as key=v n= lo= hi= null= only with n above 0 and lo at or below v at or below hi; every other shape prints REFUSED-BARE-NUMBER, REFUSED-NO-INTERVAL or REFUSED-INTERVAL-EXCLUDES-VALUE BY NAME in place of the value and returns its own code. nx_number_envelope_gate 12/12 GREEN, bite-proven (a bare percentage refused, the well-formed figure silent), built 29,482 B; staging through /api/build in flight. Owed: promote and register the gate, migrate this board's own emitted figures onto the envelope, referee grade, operator ACCEPT 2026-09-06 land nx_number_envelope_gate staged through /api/build at 34,312 B sha bf93a4e3 (src_path names it, src_stable=1) and its promote with expect_sha256 issued 2026-09-06; nx_organ_ship (CE2) compile-proof GREEN on the sovereign lane (122,520 B debug flavour) and its /api/build stage issued the same minute 2026-09-06 land CE6 SHIPPED AND PROVEN ON THE SERVED ARTIFACT 2026-09-06: nx_number_envelope_gate PROMOTED 34,312 B sha bf93a4e3 (prev banked), REGISTERED (callable on the allowlist hot-read; toolreg reconcile issued for tools/list discovery), _offc installed from the promoted root, and the SERVED binary run through nx_job_run reads 12/12 GREEN with the bare-percentage bite proven on the live copy. Remaining for CE6: migrate this board's own emitted figures (gauge, coverage, adoption) onto ce_number_envelope so the emitter refuses a bare percentage in production, referee grade, operator ACCEPT

CE7 The autonomous fix loop registered and proven on one real task · cell WATCHING · gate NO-GATE-DECLARED

Contract	`ce_fixloop_live` · executor Organ · 2 u · stands on CE6
Definition of done	The loop is BUILT and its grader PROMOTED-UNREGISTERED. Register it and resolve one real debt-board task end to end -- localise from the grader rows, propose, verify by the byte-exact judge on the SWE-bench contract, revert proven exact. ACCEPT RULE: one episode resolves a real task with its RED-before-and-GREEN-after receipt banked and a null control (empty patch, revert patch, echo) scoring zero beside it
Matrix cell	BUGRES: the autonomous fix loop -- localise, propose, verify by executable oracle, revert proven exact · `runtime/nx_autofix_auto.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_autofix_auto`; declare `nx_autofix_auto TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 retract The rung's premise (the grader PROMOTED-UNREGISTERED) is STALE by measurement 2026-09-06: nx_catalog reads nx_autofix_auto (46,116 B source, 246,691 B promoted) REGISTERED and AUTHORISED, and nx_swebv_grade (52,655 B) REGISTERED and AUTHORISED, both REGISTERED-DARK (no MCP invocation on record). The registration half of CE7 is therefore already done; what remains is the one real episode (localise from the grader rows, propose, verify by the byte-exact judge, revert proven exact) with its RED-before and GREEN-after, and that episode needs a fresh compile-and-run as its sole judge, i.e. the build lane 2026-09-06 queue SCOPED 2026-09-06 after reading the loop's own header: nx_autofix_auto <model> [manifest] [ledger-override] discovers the failing function from a grader's per-function FNRES rows, hands the extracted line to a LOCAL 1.5B i8 maker (greedy, then one reframed retry, then AF_BON_N seeded samples), and lets a fresh compile-and-run grader be the only judge at every rung; it was proven on seeded candidates (candidate2 -> sgn, candidate3 -> dbl) in the testing lane. The one REAL episode the accept rule demands therefore needs three things this seat does not hold on the NAS today: (1) the local maker -- the dispatch ladder measured no local engine on the NAS, so the episode is a laptop-engine lane (the sanctioned wsl launcher) or waits for an engine on the box; (2) a real debt-board defect expressed as a failing FNRES row, i.e. a subject whose gate emits per-function results, chosen from the open board and named before the run; (3) the null controls beside it (empty patch, revert patch, echo) each scoring zero, plus the RED-before receipt banked. Registration is already done (both organs REGISTERED). Not started here: an autonomous fix loop on a degraded array with no engine would be a workaround, not the rung

CE8 Duplication trend not just level · cell LANDED · gate NO-GATE-DECLARED

Contract	`fg_trend` · executor Organ · 1 u · stands on nothing
Definition of done	Shared with codequality -- one census is a level and the field bar is a trend (block duplication up 81 percent). The organ appends to a durable log on a beat and refuses to emit a trend from a single sample. ACCEPT RULE: two beats reconcile and a single sample reads UNOBSERVABLE; do not add a second duplication ruler, extend nx_forkgrade
Matrix cell	DUPLICATION: cross-tree fork and duplication graded over every pair · `runtime/_hdl_build/nx_forkgrade.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_forkgrade`; declare `nx_forkgrade TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 land fg_trend landed in buildroot/runtime/_hdl_build/nx_forkgrade.nx (20,394 B): one spine row per SUCCESSFUL census, appended only after the partition is proven sound; the path is DERIVED from dirA (production knowledge/status/forkgrade.spine iff dirA is buildroot/runtime, else a <dirA>.forkgrade.spine sidecar) so a trial run cannot forge the production trajectory; announced spine= fd= wrote= of=. nx_forkgrade_gate.nx created (10,500 B): three planted pairs identical/A-superset/bidirectional with hand-computed answers (adoptable 666 permil), two runs leave two rows, neg-control that the production spine size is unchanged, argv subject override, RED by construction against the pre-trend binary 2026-09-06 queue both builds storm-deferred: nx_build_admit read GRANT at procs_blocked 2 of 8 and 40 s later the sovereign lane was refused QUEUE at 8 of 8 (the D-state witness, not load). Recipe: nx_sov_build_run nx_forkgrade --build-only, then nx_forkgrade_gate --build-only, paced a minute apart; run the gate via nx_job_run from the serving root with the _build subject (RED on the promoted binary, GREEN on the staged one is the proof); promote both with expect_sha256, register the gate, nx_offc_install both; then the watch _ABSENT_:fg_trend flips on the next admitted regen 2026-09-06 land CE8 PROVEN END TO END on the served artifact 2026-09-06: nx_forkgrade (fg_trend) staged by /api/build at 44,656 B sha c43c490d (src_path names the target, src_stable=1), nx_contentdiff live-vs-staged GREEN lost_from_live=0 ONE-WAY-GAIN (78 runs gained including the spine literals), PROMOTED with expect_sha256, prev banked, nx_offc_install refreshed _offc to the blessed bytes; nx_forkgrade_gate (43,824 B) read RED 4/12 against the OLD served twin (census GREEN total=3, only the eight trajectory teeth failed: the pre-trend binary writes no spine) and GREEN 12/12 against the NEW one (sidecar holds 2 rows after two runs, adoptable 666 permil hand-computed, production spine size unchanged, bite-proven). Owed: stage, promote and register the gate itself; the production spine knowledge/status/forkgrade.spine has no row yet and gets its first point on the next real census over buildroot/runtime (nx_forkgrade is REGISTERED-DARK, its pinned census args unknown to this seat; a trend needs two censuses in time); referee grade and operator ACCEPT 2026-09-06 queue The census's second tree is the LAPTOP tree staged on the NAS (usage nx_forkgrade <listfile> <dirA> <dirB>, dirB e.g. /tmp/lapfork; the organ's own header says moving bytes between the two trees is irreducibly laptop-side), so the production spine cannot be fed by a NAS clock row: its first two points belong to the tree-sync lane (nx_treediverge_beat already stages and compares the trees; one fg_trend census per sync is the adoption step). Not a build -- a declaration on that lane's board is owed

CE9 The gauge is a beat with an axis-blind alarm so RED can never go dark · cell LANDED · gate NO-GATE-DECLARED

Contract	`ST_GAUGE_STAMP` · executor Organ · 1 u · stands on nothing
Definition of done	nx_seat shift is the gauge and it reads RED now; the risk is that a stale gauge reads as no movement rather than as blindness. A beat stamps the gauge and SessionStart prints its age, and an abstaining axis raises an axis-blind alarm. ACCEPT RULE: a stale gauge reads STALE on this page and never as zero, and a forced-stale fixture raises the alarm
Matrix cell	GAUGE: the effectiveness gauge is a beat with an axis-blind alarm so a RED can never go dark · `runtime/_hdl_build/nx_seat.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_seat`; declare `nx_seat TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 land LAST MILE SOURCE-LANDED (agreed shape: a NAS-side stamp derived from the rows the beat already pushes): nx_seat st_shift now writes knowledge/status/seat_gauge.stamp through ga_stamp_write -- gauge = claude out per closed rung, axes = meter log + ws journal (a window that read only one is BLIND), cadence = the window itself so a stamp older than two windows reads STALE; a window that closed nothing writes NO stamp (UNMEASURED is not a value) and the old stamp ages into STALE, the axis-blind alarm. The shift JSON gains gauge_stamp_wrote. Three CAS edits 203774 -> 204138 -> 204949 -> 205034 (two refused by ordering, re-issued under their predicted tokens). Build queued via /api/build under CPU saturation (run queue 18 of 8); after promote: run nx_seat shift, verify the stamp with ga_judge, then land the codeeffectiveness.gauge row for it. 2026-09-06 land LAST MILE PROMOTED: nx_seat 360016 B sha 30309d65 via /api/build (src_path named, src_stable 1). Promote guard refused CAPLOSS; nx_contentdiff enumerated 4 lost runs (3 permil) and every one was adjudicated before the flag: manifest-shrank and prefix-mismatch ABSENT-PROVEN from all runtime sources (corpus_complete=1), docs-size-mismatch present only inside a gate COMMENT (no source emits it), ssl_total_keys_range a seg-store lib FUNCTION NAME the rebuilt call graph no longer reaches -- the seg-store's successor vocabulary (uncovered-vanished-keys, INCR-FOLD, tombstoned=) and CE6's ce_number_envelope are among the 53 gains: library evolution since the live binary was built, not this change. Promoted with allow_capability_loss=yes on that enumeration. Proof run (nx_seat shift 7) in flight; the codeeffectiveness.gauge row lands once the stamp exists. 2026-09-06 measure THE BOARD READS IT: nx_compare_rank codeeffectiveness done=9 open=1 (CE7 alone remains, priority 300), unmapped=0 -- CE9 flipped once the rung and the matrix named the SAME NAS-side symbol (ST_GAUGE_STAMP in runtime/_hdl_build/nx_seat.nx) and the ranker probed the source itself. Two gauges now render on this page through gauge_pass: the one-step explained permil (daily) and tokens per closed rung (weekly, this stamp). STATE: shipped and proven, referee grade and operator ACCEPT owed. 2026-09-06 retract THE GAUGE CELL NEVER RENDERED: the CE9 land row proved the STAMP WRITE (seat_gauge.stamp, ts/gauge/axes 2/2) and never the RENDER. The regen runs with CWD=buildroot and gauge_pass opened the stamp path as given, so knowledge/status/seat_gauge.stamp and knowledge/status/stepsolve.stamp (66 B at the serving root, ABSENT under buildroot) judged ABSENT on every publish. Discriminating experiment: nx_fs size of the stamp at both roots. Fixed in the emitter base (gauge_pass resolves through ep_artifact_path before ga_judge; matrix generator PROMOTED 422,144 B sha 4e99272e, ruler GREEN); the cell reads FRESH only after buildroot/_offc is refreshed and the regen runs -- until then this rung's page evidence is UNVERIFIED 2026-09-06 queue SCOPED 2026-09-06, not started (build lane wedged on the RAID): sm_gauge_alarm has two halves and the emitter half is the load-bearing one. (1) In the laptop-local meter runtime/_hdl_build/nx_seat_meter_beat.nx (56,340 B, built by the sanctioned wsl launcher, not the NAS lane): stamp the gauge on every beat (knowledge/status/seat_gauge.stamp with ts= and a scale field such as rows= so a fixture cannot forge it), and sm_gauge_alarm(stamp, max_age_s, now) classifying FRESH / STALE / ABSENT with the bound DERIVED from the beat cadence (one full missed beat of margin, the memvel precedent), truncate-writing knowledge/status/seat_gauge_alarm.status whose last line is verdict=; an abstaining axis in the gauge raises AXIS-BLIND by name. (2) In nx_swcompare_matrix (grow the emitter, never a side page): the GAUGE row reads that status and renders STALE or ABSENT in place of the number, never zero. (3) nx_seat_gauge_alarm_gate: a forced-stale fixture (stamp older than the bound) raises the alarm, a fresh one is silent, an absent stamp reads ABSENT, and the emitter run over a fixture tree prints STALE for the row. Owed after CE2 and CE8 ship 2026-09-06 land CE9 gauge ruler + gate SHIPPED NAS-native: nx_gauge_lib (7,661 B, THE ONE gauge-heartbeat ruler -- ts-first dead-man stamp, ABSENT>STALE>BLIND>FRESH classifier, future stamp and unpinned cadence both STALE, value shown ONLY when FRESH, max_age derived as two beats). nx_gauge_gate 14/14 GREEN bite-proven built+run on the NAS (withhold-value-fires-on-STALE-silent-on-FRESH BITE-PROVEN), PROMOTED 39,465 B 92648727, REGISTERED callable=1, _offc refreshed, cap minted 30 d. gauge_pass wired into the compare emitter nx_swcompare_lib (both generators, api.json + page); a <dom>.gauge row is re-judged every publish by the same ruler and the value key exists only on a FRESH row, so a missing key can never read as zero. Generator promote (to render live) deferred under the I/O storm; a codeeffectiveness.gauge data row waits for its stamp writer. 2026-09-06 retract the beat-writer half was NOT shipped and its first cut was WRONG-LANED: sm_gauge_alarm was added to the NAS buildroot copy of nx_seat_meter_beat and /api/build failed on mp_hook_json undefined at two PRE-EXISTING call sites. nx_absent proved mp_hook_json absent from BOTH NAS trees (corpus_complete=1) -- nx_seat_meter_beat is a LAPTOP-LOCAL SessionStart organ, so /api/build was the wrong surface, not a divergence defect. All three NAS-copy edits reverted to the exact 56,340 B original. The stamp writer (tokens/rung is laptop-only) is a laptop-tree step: copy nx_gauge_lib into the laptop tree, wire sm_gauge_alarm, build in WSL. Until then the gauge reads ABSENT/STALE, which is the honest state the rung's accept rule wants. 2026-09-06 lesson A COMPILE ERROR AT A PRE-EXISTING CALL SITE AFTER AN ADDITIVE EDIT MEANS THE FILE WAS ALREADY UNBUILDABLE ON THAT SURFACE -- check which tree owns the organ before choosing a build lane. And the accept rule is met by the gate, not the flip: a stale gauge that reads STALE not zero is proven by nx_gauge_gate, and the /compare cell reading ABSENT until a writer exists is honest, not incomplete-in-the-bad-direction. 2026-09-06 land CE9 WRITER LANDED ON ITS NATIVE SURFACE (laptop), storm-independent. nx_seat_meter_beat is a laptop-local SessionStart organ (mp_hook_json lives in the LAPTOP runtime/_hdl_build/nx_memplane_lib.nx; nx_absent proved it absent from both NAS trees, so the NAS build farm is the wrong surface and its storm does not block this). The laptop source carries the gauge wiring (import nx_gauge_lib + sm_gauge_alarm + the call after the METERBEAT summary); built via the sanctioned wsl sovereign-ELF launcher (nx_sov_build_run nx_seat_meter_beat --build-only then --install), 82,970 B sha cc9a865a, installed to _offc/nx_seat_meter_beat.elf. PROVEN by a scan run: GAUGE-ALARM state=BLIND prev_age_s=-1 axes_known=1 axes_total=2 wrote=0 -- scan correctly sees one axis abstain (the remote meter log it does not read) and reports BLIND rather than acquitting, and writes nothing; a real SessionStart beat reads both axes and writes knowledge/status/seat_gauge.stamp. So the three CE9 layers now exist: ruler (nx_gauge_lib, NAS+laptop), gate (nx_gauge_gate 14/14 GREEN, NAS), writer (nx_seat_meter_beat --install, laptop live), emitter render (gauge_pass in nx_swcompare_lib, written). 2026-09-06 queue two integration steps remain, both cleanly named: (1) generator rebuild+promote so gauge_pass renders live -- storm-blocked (procs_blocked 11 of 8), and inert until a .gauge data row exists; (2) the cross-surface stamp delivery -- the writer computes laptop-side (transcripts are laptop-only) while /compare renders NAS-side, so the gauge stamp must ride the seat meter's existing estate push to a NAS-readable path that a codeeffectiveness.gauge row points at. Neither is a hack; both ride mechanisms that already exist. Three-party ACCEPT (referee + operator) still owed for CE3 and CE9. 2026-09-06 land GENERATOR PROMOTE DONE -- gauge_pass is LIVE in the compare emitter fleet-wide (the storm eased and /api/build granted). nx_swcompare_matrix 412,656 B 97b52ddb and nx_swcompare_sota 261,712 B c39dfdf8: both nx_contentdiff GREEN strict-superset (lost_from_live=0, gained the gauge strings gauge=/axes_known=/axes_total=/cadence_s=/age_s= and knowledge/status/seat_gauge.stamp), src_path confirms the target, src_stable=1. Promoted (prev banked), nishihost/_offc installed, AND buildroot/_offc refreshed via nx_restage so all THREE roots the regen forks agree (matrix 412,656 all three; sota 261,712 all three) -- the FORK-STALE class cannot fire. gauge_pass is additive (absent <dom>.gauge file => no section, returns 0), so it renders nothing on any page until a .gauge row exists, i.e. zero risk to the 109 live pages and confirmed by the strict-superset diff. CE9 layers now ALL live: ruler (nx_gauge_lib) + gate (nx_gauge_gate 14/14) + writer (nx_seat_meter_beat --install, laptop) + emitter render (gauge_pass, promoted). ONE step remains: cross-surface stamp delivery -- the laptop writer's stamp must reach a NAS-readable path a codeeffectiveness.gauge row points at (rides the seat meter's existing estate push; the clean form is a NAS-side stamp derived from the already-pushed meter data). Not a hack, a named follow-up. Three-party ACCEPT still owed. 2026-09-06 queue CE9 LAST MILE PINNED AS A BUILD ORDER (the clean form, not a hack). The cross-surface stamp: nx_seat's `shift` verb (st_shift, buildroot/runtime/_hdl_build/nx_seat.nx ~204 KB) ALREADY runs NAS-side, computes tokens-per-rung, and writes shift.log + shift.jrnl -- so the gauge stamp is a bounded addition THERE: import nx_gauge_lib and call ga_stamp_write(GA_STAMP_PROD, now, gauge, axes_known, axes_total, cadence) at the end of st_shift, then a codeeffectiveness.gauge row points at knowledge/status/seat_gauge.stamp and gauge_pass renders it FRESH. NOT done this session, deliberately: it edits a 204 KB core organ, needs its axes-contract defined (nx_seat reads the pushed meter data, so axes_known/total means the coverage of that push, a different semantics from the laptop beat's local-scan+remote-log), and the box is under an intermittent RAID storm -- rushing a core-organ edit there is the workaround the standing orders forbid. The laptop writer already stamps locally (proven); the NAS-side stamp is the delivery. Build it on a calm box with the axes contract written first. 2026-09-06 queue CROSS-BOARD THREAD: the operator's 2026-09-06 one-step-solve goal is declared as engineshift ES26 (al_steps on the incumbent nx_actlog) with its architecture twin ecosystem EC37 (the common-task registry's consolidation-debt ratchet) and a verbatim lesson row on EC18. Its gauge stamp knowledge/status/stepsolve.stamp will render through THIS board's gauge_pass -- the CE9 mechanism -- as a second gauge row: one render, many gauges. The operator also agreed (same day) that CE9's cross-surface stamp is the NAS-side form derived from the pushed meter data, a small deliberate build.

CE10 A mirror whose canonical URL differs from the cited URL reads MOVED, never PROVEN · cell LANDED · gate NO-GATE-DECLARED

Contract	`cr_canonical_moved` · executor Organ · 1 u · stands on nothing
Definition of done	FOUND BUILDING THIS BOARD: the Faros /ai-impact page every other board cites now redirects to the Faros homepage, and a fresh fetch returned a 126648-byte page carrying none of the cited terms while printing status=200 -- the fetcher flagged it with a trailing 1 in its PROVENANCE line and no consumer reads that flag, so a moved page can be pinned as evidence. Extend nx_compare_refs_gate (the incumbent, never a second ruler) with a canonical-URL check: read the mirror's rel=canonical or og:url, normalise scheme, host and trailing slash, and classify MOVED when it differs from the cited URL, rendering it with the same drift classes as STALE-UNDER. ACCEPT RULE declared in advance: the Faros homepage mirror rf_ff06ac9a.raw cited under its old URL must read MOVED, the same page cited under its canonical must read PROVEN, an arxiv abstract whose canonical is itself must read PROVEN, a mirror with no canonical tag must read UNKNOWN never MOVED, and the fetcher's own trailing flag must be consumed and named in the verdict; bite-proven both directions before the tooth is armed
Matrix cell	RIGOR: a mirror whose canonical URL differs from the cited URL reads MOVED, never PROVEN · `runtime/nx_compare_refs_gate.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_compare_refs_gate`; declare `nx_compare_refs_gate TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-05 measure nx_compare_refs_gate 24 of 24 GREEN on the whole fleet after this board landed: provenance proven 848 mismatch 0 unproven 587 (baseline held), every pin matched, every mark resolved. It read 23 of 24 RED twice on the way -- first because 8 reused mirrors predated the provenance journal (fixed by re-fetching each under its own row rather than naming a residual), then because this seat cited dora-2025 under another board's URL while the mirror was journalled under the URL actually fetched (caught by the gate's own mirror-under-a-different-url neg-control, fixed by citing the fetched URL). The Faros /ai-impact page that devmgmt and others still cite is MOVED (it redirects to the homepage) and is the planted RED for CE10. 2026-09-05 land TWO BUS STOPS FIXED IN THE ESTATE, NOT NAMED (2026-09-05). (1) nx_lease check now probes all three lease roots the estate's two CWDs produce (knowledge/lease, buildroot/knowledge/lease, ../knowledge/lease) and a HELD stamp from any root wins, printing LS-ROOT so a reader knows which store answered; write side untouched; built 39792 B sha ebf93e4c, PROMOTED (live hash matches), selftest GREEN 9 of 9 including both concurrency teeth, _offc twin installed. (2) nx_swcompare_lib bar rows wrote their Theirs and Ours fields with the plain writer so every cite mark rendered raw on every board that carries bars (24 on engineshift alone); both fields now go through wnote, the cite-expanding writer the pos, goal, answer and biz rows already use. nx_swcompare_matrix rebuilt 368016 B sha 80d97b58 and PROMOTED prev=1 with its _offc twin; buildroot/_offc refreshed by nx_filecopy because nx_restage builds and admission is queueing (BQ-nx_swcompare_sota is queued for the sota generator). The redirect-hops consumer for the refs gate is landed in source as five sequential CAS edits (reader, pass hook, worklist, ratchet tooth, bite) and awaits its build. 2026-09-05 land REDIRECT-HOPS CONSUMER LIVE IN THE REFS GATE (2026-09-05): nx_compare_refs_gate built 98480 B sha 592fa512 through /api/build, PROMOTED prev=1, _offc twin; first run on the fleet self-baselined REDIR-RATCHET at 33 -- thirty-three journalled-proven mirrors whose fetch followed a redirect, each NAMED in knowledge/status/refs_redirected.txt (extllm's OpenRouter pages at 1 hop, a Gemini terms page at 4) -- with the redirect bite BITE-PROVEN (hops 1 on the planted redirected row, 0 on the direct row, -1 on an unjournalled mirror) and the ratchet tooth PASS. The run also read 26 of 27 RED because THIS board's own journal row about the bar-mark fix contained a mark-shaped literal the gate read as an unresolved cite; reworded, re-run: 27 of 27 GREEN fleet-wide. Half two of CE10 -- canonical-url comparison that separates MOVED from a benign scheme or trailing-slash hop -- stays open; the 33 named rows are its worklist. 2026-09-05 land refs gate canonical half LIVE 107936 B sha bdb5d8c6: rel=canonical or og:url read per redirected mirror, MOVED/BENIGN-HOP/CANON-UNKNOWN partition sums to redirected=33, worklist knowledge/status/refs_moved.txt, MOVED ratchet self-baselined at 1; fleet 29/29 GREEN 2026-09-05 measure first canonical census: moved=1 (bevyengine.org to bevy.org) benign=0 unknown=32 of 33 -- and 30 of the 32 were the whole-file reader REFUSING files over its 16 KiB cap (CRG-READ-CAP-EXCEEDED), not absent tags: the detector was vacuous on real pages while printing a verdict per row 2026-09-05 lesson a reader that refuses to truncate is right for a BODY and wrong for a HEAD: the canonical tag lives in head, so the honest read is a head window that reports whether the head CLOSED inside it; an absent tag is NONE only when the head was seen whole, otherwise TRUNCATED, never NONE 2026-09-05 land crg_read_head plus three named negatives (CANON-NONE-IN-HEAD, CANON-HEAD-TRUNCATED, CANON-UNREADABLE in one partition bucket with a sum tooth) plus a canonical bite (moved vs benign hop, closed-head NONE, out-of-tag trap, 17 KiB fixture read in-window, beyond-window TRUNCATED, absent UNREADABLE) source-landed at 74519 B; build in flight 2026-09-05 queue after the head-window build the MOVED count rises off a baseline the blind reader took: re-baseline the MOVED ratchet with that reason (instrument changed, not more moves) and work refs_moved.txt by re-citing each canonical url 2026-09-05 land emitter: nx_swcompare_lib plan_pass now expands cite marks in rung done-rules, risk cells and debt cells (bar rows earlier); nx_swcompare_matrix 370184 c30b3d63 and nx_swcompare_sota 221792 ba51e306 promoted, _offc and buildroot/_offc twins byte-verified by receipt; regen REFUSED-IO-STORM procs_blocked=7, the beat re-runs when calm 2026-09-05 measure second witness for debt 1788629238 (api-build-cache-stale-key, open, sev 6): /api/build of nx_compare_refs_gate returned action BUILT bytes 107936 sha bdb5d8c6 for src_sha 90d7783d after an admission-refused build, byte-identical to the artifact built from src_sha 8151ddf1 -- staged, fossil and live all read the OLD artifact while the source had grown 7.5 KB; the sovereign lane keys correctly but was REFUSED-BUILD-ADMIT (run queue 13 then 9 on 8 cpus), so the gate rebuild is paced behind admission, never routed around it 2026-09-05 land head-window canonical reader LIVE (nx_compare_refs_gate 115952 B sha dc2eb3cc, debug flavour to match the /api/build live, contentdiff GREEN strict superset 0 lost 45 gained, prev bdb5d8c6 banked); canonical bite 8/8 including a 17504 B fixture read in-window that the cap reader refused, beyond-window TRUNCATED, absent UNREADABLE 2026-09-05 measure first TRUE canonical census over the 33 redirected mirrors: MOVED 21, BENIGN-HOP 5, CANON-NONE-IN-HEAD 5, CANON-HEAD-TRUNCATED 2 (sums to 33; the blind reader had said 1/0/32). MOVED names a re-cite worklist in knowledge/status/refs_moved.txt: openrouter docs restructured, kubernetes and aws pages moved, huggingface xet, metahuman, nvidia ace, dora.dev research pages now the homepage (twice), nar.realtor, nats leaf nodes, an aws page only on web.archive, bing visual search, ms servicing, two PMC articles on the new pmc host, unrealengine features, bevy.org x3 (one landing on a quick-start page, not the cited features page) 2026-09-05 lesson the fleet gate went 36/38 RED for two honest reasons: the MOVED ratchet rose 1 to 21 because the INSTRUMENT changed, and a gameengine log row carrying WGSL attribute syntax read as a dangling cite mark. Neither is loosened: the ratchet is re-baselined by a written, announced directive (the conf name is retire-protected, so a bare number would have been laundering), and the mark scanner now requires a well-formed key so code that merely starts with the two mark characters is prose 2026-09-05 queue 21 MOVED rows to re-cite under their canonical urls with fresh provenance (re-fetch under the same key, re-pin, url field) -- a worklist consumer organ rather than 21 hand edits; scanner fix + directive reader source-landed, build paced on admission 2026-09-05 retract my 2026-09-05 nx_lease check fix (3-root probe, selftest 9/9) did NOT work on the live surface: the direct MCP call from the serving root printed LS-FREE against a held compare-regen stamp the shell tool could read at buildroot/knowledge/lease/compare-regen.lock/stamp, while a lease in the serving-root store WAS found -- so the two inline-literal roots never opened and the selftest was a fixture the defect could not fail 2026-09-05 land nx_lease 42784 B sha 5e6cdaef promoted (prev banked, _offc twin): roots hoisted to consts, the probe is one function main and the selftest share, selftest 11/11 with a cross-root tooth (acquire under the buildroot root, probe must find it there and read it HELD) plus a neg-control; live check now prints LS-ROOT buildroot/knowledge/lease/ and the holder stamp. The compiler-side finding (inline literals absent from the old binary as printable runs) is filed on the lang lane as a debt row with the repro recipe, mechanism deliberately unnamed 2026-09-05 land the last 9 raw cite marks on /compare/engineshift were the matrix @verdict paragraph: both generators' verdict renderers now expand marks (nx_swcompare_matrix 379232 56997071 and nx_swcompare_sota 232064 9c7b2c5f live, both _offc and buildroot/_offc twins hash-identical); a beat regen holds compare-regen since 1788632942 and will publish it 2026-09-05 land nx_refdrift gained a recite verb (spendgate: name free, incumbent extended rather than a new organ): `nx_refdrift recite <moved-worklist> [max] [apply] [key=<k>]` reads the refs gate's MOVED worklist, fetches each canonical into the probe, refuses a body the completeness classifier calls truncated or suspect (mirror untouched), dry prints WOULD-RECITE, apply re-fetches under the mirror's own name for fresh provenance, hashes with the one hasher and rewrites url, pin and accessed by walking pipes with temp plus rename; envelope partition sums, PARTIAL declared. Built 65648 B sha bb9c81ba (debug flavour), contentdiff GREEN strict superset, promoted prev=1, _offc twin; first dry run over all 21 rows in flight 2026-09-05 measure recite dry run over all 21 MOVED rows: 21 fetched, 0 refused, 17 complete bodies and 4 unknown-class html, accessed date derived 2026-09-05, partition sums, PARTIAL=0. Adjudicated before apply: 5 rows canonicalise to a homepage or an unrelated page (dorakeys and dora-fourkeys to dora.dev, metahuman to metahuman.com, bevy-home to bevy.org//, bevy-features to a quick-start intro) and stay for a human read; 16 are the same document at a new address 2026-09-05 land 16 rows RECITED under their canonical urls with fresh provenance and pins (orlimits, orrouting, k8s-probes, arc-dataplane-recovery, fespa, hf-xet, nvace, narfsbo, nats-leafnodes, aws-backoff-jitter, bing-vs, secureboot-expiry26, nac-review21, nac-tattoo24, unreal-features, bevy-news), applied in three batches so no two rows rewrote one register at once; one PMC fetch returned an empty body and was REFUSED with the mirror untouched, then recited on retry 2026-09-05 lesson the verb's first live batch found its own defect: an apply with one refusal and nothing recited printed verdict=RECITED -- a verdict named for the MODE, not for what happened. Fixed: NOTHING-RECITED (exit 1) and RECITED-WITH-REFUSALS are their own words; and rows now flush as they are decided so an interrupted run keeps its rows 2026-09-05 land refs gate 117696 B sha a6e62ed8 live: url equality now collapses a RUN of slashes to one and drops a trailing run (bevy.org// is the site's own canonical tag, not a move), two teeth added, gate 40/40 GREEN; /compare/engineshift published with 0 raw cite marks after the beat regen picked up the fixed verdict renderer (171885 B) 2026-09-05 land the 5 held MOVED rows closed by a human read plus the same mechanism: a durable manual worklist knowledge/status/refs_recite_manual.txt carries the chosen url per row and nx_refdrift recite applies it -- metahuman to https://www.metahuman.com/ (the Unreal page redirects there in 3 hops, same body), dorakeys and dora-fourkeys to https://dora.dev/guides/dora-metrics/ (the four-keys guide moved; zero-hop fetch, identical 19454 B body both ways, one pin for both rows), bevy-home to https://bevy.org/ and bevy-features to https://bevy.org/learn/quick-start/introduction/ (domain rebrand, same pages). moved 21 -> 0 owed on the next gate run 2026-09-05 queue two mgmt-lane build refusals were absorbed into the deploy queue as BQ-nx_refdrift and BQ-nx_compare_refs_gate; both organs have since been built through the sovereign lane and promoted from source identical to what those rows would compile, so when they fire they produce a content-identical binary and, via the queue's serving-root CWD, may redeploy it and overwrite the .prev rollback slots (bb9c81ba for nx_refdrift, dc2eb3cc for the gate). Not lost work: the rows are redundant. A queue verb to withdraw a row whose target was promoted since it was queued is the missing mechanism (nx_buildq has close but no withdraw-by-name from the seat surface) 2026-09-05 lesson the redirect ratchet fired on a row another lane registered today (fieldwatch g2-scoring, hops=1, canonical OG-URL-ONLY): the raw redirect total rose 33 to 34 while moved stayed 0. Measured on the same run, 20 of the 34 redirected mirrors are BENIGN-HOP (the page's own canonical equals the cited url after normalisation), so a ratchet on the raw total carried 588 permil false positives on real data and would have gone RED for every lane that cites through any scheme, www or slash hop. A detector with false positives teaches every lane to ignore it: the SUBJECT was wrong, not the number, and the remedy is neither a rebaseline directive nor a loosened bar 2026-09-05 land nx_compare_refs_gate: the ratchet subject is now the OPEN set (redirected minus benign-hop, which equals moved plus every canon-unknown reason); worklist knowledge/status/refs_redirect_open.txt carries the class ON each row; conf refs_redirect_open_ratchet.conf self-baselines on first sight (14 today: none_in_head 5, head_truncated 3, og_url_only 6, moved 0). The raw total stays censused in refs_redirected.txt and is emitted as values redirected_total, redirected_benign_hop, redirected_open via gv_kv. Three new teeth test the state not the message: open equals moved plus canon_unknown, and each written worklist's row count equals its counter. Source landed at 82966 B in four CAS edits (the file is mixed CRLF and LF, 923 CR lines, so a newline-terminated anchor NOMATCHes on the old body); first build refused by admission (procs_blocked 15 against 8), paced re-issue owed, then promote, offc twin, and retire the old refs_redirected_ratchet.conf (held 33) with its at_baseline snapshot 2026-09-05 land nx_compare_refs_gate LIVE 93a8beee (121232 B, debug flavour like its predecessor; the first non-debug rebuild came out 106763 B, 11709 B under live after an addition, and the flavour explains the whole delta, so the same-flavour build is the one adjudicated). contentdiff RED on exactly five runs, all my own renames (three REDIR-RATCHET messages, the old count sentence, the old tooth name), each superseded by an OPEN-REDIRECT twin among the 19 gained, so allow_capability_loss was passed with that enumeration and no vintage control was owed (no source-path class lost). Promote landed under a dropped 503 receipt; the retry answered NOTHING-STAGED with live_sha256 equal to the build, offc twin installed, both roots 93a8beee. Direct run: NX-comparerefs passed 44/44 verdict=GREEN, open ratchet SELF-BASELINED at 14, values redirected_total=34 redirected_benign_hop=20 redirected_open=14, moved=0. Old ratchet: at_baseline snapshot retired reversibly (knowledge/retired/...-1788640040), the retire-protected .conf now carries a written RETIRED directive naming its successor 2026-09-05 measure UNATTRIBUTED, recorded so nobody inherits it as a fact: nx_behaveprobe live-vs-staged reported other_rc=1 for the staged gate with 16264 B of output, while the direct run of the same bytes minutes later read GREEN 44/44 at 16391 B and a later live-vs-prev probe read live_rc=0 at the identical 16391 B. The probe runs its two subjects in sequence (tr_run_capture_cwd, one after the other), so the clobber-the-shared-worklist theory is refuted; byte arithmetic (the GREEN verdict note is roughly 380 B the RED line lacks, the first-sight baseline line roughly 90 B) leaves about 160 B of unexplained extra output in the staged run, the size of one transient UNRESOLVED-MARK or UNCITED row from a sibling lane's edit in flight across the fleet scan. The probe keeps no capture, so this stays a hypothesis; the remedy that would settle it is a probe that writes both captures beside its verdict 2026-09-05 queue regen of /compare owed for the fixed verdict-paragraph cite rendering and this board; the compare beat publishes on its own cadence and a seat launch under the storm would be refused or would hold the compare-regen lease against the beat, so the beat is left to publish 2026-09-05 land the watch symbol is now REAL CODE on the load-bearing path: cr_canonical_moved(cl, canon, url) -- 1 iff a readable canonical, normalised, names a different page than the cited url -- is defined after crg_url_norm_eq and is the predicate the row loop's MOVED-or-BENIGN decision calls (definition plus one call, count 2). contentdiff ONE-WAY-GAIN gained 1 lost 0, promoted e1fedff3 (121456 B, prev banked), offc twin installed, both roots verified by hash. The compare cell flips on the next regen beat because the capability exists, not because a name was typed 2026-09-05 measure ATTRIBUTED, superseding the earlier UNATTRIBUTED row: the live gate run on e1fedff3 read 43/44 RED on cite-marks-resolve-every-one with UNRESOLVED-MARK file=buildroot/knowledge/compare/intelmine.matrix key=semeval14-paper -- the intelmine lane's mark landed before its refs row, fleet rows 1557 to 1560 between runs. That is the roughly-160-byte extra line the probe's staged run carried, so the probe's rc=1 was this same sibling-lane transient, not a defect in the staged gate; every redirect, open-ratchet and worklist tooth reads PASS. Not my row; the gate names its owner and clears itself when the ref lands

CE11 A gv_kv that republishes a quantity a gv_check_eq already asserted is a SECOND RULER · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW

Contract	`gl_l5_double_publish` · executor Organ · 1.5 u · stands on CE6
Definition of done	MEASURED 2026-09-06, and this rung exists because the defect was found in a shipping gate rather than imagined: nx_uxf_mp4box_gate passed 8 of 8 GREEN while publishing toplevel_box_count=0 against its own tooth T2b asserting actual=2 expected=2. One cnt buffer is filled by the good-tree walk, the tooth tests cnt[0] and passes on 2, the neg-control then re-walks into THAT SAME out-parameter leaving 0, and the gv_kv emission runs last -- so the published number was the neg-control's answer wearing the tooth's name, and a buffer restore two lines later could not reach it because a restore undoes the INPUT and not the DERIVED state. THE GENERAL LAW: gv_check_eq binds the number tested to the number published ONLY for values passed INTO it, and a separate gv_kv that re-reads a mutable buffer afterwards carries no such binding, which makes it a second ruler over a quantity the gate has already ruled on. The signal is mechanical and greppable -- the same variable expression appearing both as a gv_check_eq argument and as a gv_kv argument within one gate. EXTEND THE INCUMBENT AND DO NOT ADD A RULER: this is an L5 axis on nx_gatelaw_gate beside L1 exit-carries-verdict, L2 names-a-neg-control, L3 base-class and L4 emits-values, and it is DECLARED SEPARATE from all four because a gate can satisfy every one of them and still double-publish, so folding it in would silently redefine a figure other lanes act on. nx_capsearch over 7551 surfaces with corpus_complete=1 returned no incumbent detector, so the gap is measured rather than assumed. ACCEPT RULE DECLARED IN ADVANCE: the axis must FIRE on the pre-fix nx_uxf_mp4box_gate source as a planted positive and must NOT fire on its fixed source, it must measure and publish its false-positive rate against hand-adjudicated hits before any count is published as a defect count, and a gate whose dialect it cannot read must be left UNSCORED on this axis rather than asserted clean, because a ruler that declares a stream unreadable must not then assert anything about it on another axis
Matrix cell	NO-MATRIX-ROW — no matrix row carries `gl_l5_double_publish`; the ranker reads this rung as unmapped and no cell can flip for it. Add the row and the join completes on the next publish.
Gate	the gate join needs a matrix row: no organ is named for this rung
Work log	no log row names this rung yet

CE12 Accepted task, workflow and project outcomes for humans, AI and their partnership · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW

Contract	`_ABSENT_:ce_accepted_outcomes_accept` · executor Workflow · unestimated u · stands on CE1, CE2, CE3, CE6, CE7
Definition of done	PLANNED CONTRACT 2026-09-30. Extend existing authorship, build-first-try, recurrence, number-envelope, fix-loop and resource instruments. Compare matched human-led, AI-led within authority, and human-plus-Nishi assistance conditions with exact tools/model/policy versions and equivalent tasks/resources; mechanistic help is an explicit treatment. Report human active time, elapsed delivery time, review/rework, quality, retained features, escaped defects/recurrence, resource and actual token/cost demand per accepted outcome. Keep task/workflow/project denominators separate; retain mixed/unknown attribution, unfinished/censored work, selection effects, concurrency and uncertainty. Independent acceptance is fixed before candidate work; speed cannot offset correctness/safety/capability loss. Establish efficiency baselines first, then representative studies. No Nishi result is claimed until executed. Compare emits the findings and limitations from receipts, not authored rankings.
Matrix cell	NO-MATRIX-ROW — no matrix row carries `ce_accepted_outcomes_accept`; the ranker reads this rung as unmapped and no cell can flip for it. Add the row and the join completes on the next publish.
Gate	the gate join needs a matrix row: no organ is named for this rung
Work log	2026-09-30 queue 2026-09-30 operator: true human-estate partnership and eventual first/third-party LLM comparison. First baseline token/resource costs, then accepted developer outcomes at task/workflow/project levels. This record adds acceptance criteria, not measured results.

dossiers=12 cells_joined=10 no_matrix_row=2 (sum 12) gates_recorded=0 gates_unrecorded=1 no_gate_declared=9 (sum 10 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. 21 feature rows mapped of 21 matrix rows; 0 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/codeeffectiveness.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 200–1000, distinctiveness 100–700 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 728 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 200–1000 · distinctiveness 100–700 · breadth 0–1000 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 2 (a fact of the data: two players that close are that close) · data spread 101 permil of the plot

Nishi DX Faros GitClear CodeScene
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 3 (a fact of the data: two players that close are that close) · data spread 551 permil of the plot

Player	Quadrant	centrality	distinctiveness	breadth	depth	momentum	Rows held	Days on spine	First seen
Nishi	Aspirational	941	620	825	866	0 since 2026-09-30	20	11	2026-09-15
DX	Peripheral	568	260	269	708	0 since 2026-09-30	8	11	2026-09-15
Faros	Mainstream	705	275	349	733	0 since 2026-09-30	10	11	2026-09-15
GitClear	Peripheral	352	187	158	833	0 since 2026-09-30	4	11	2026-09-15
CodeScene	Unconventional	588	351	333	777	0 since 2026-09-30	9	11	2026-09-15

Day	Matrix rows reviewed	Players recorded
2026-09-15	21	5
2026-09-16	21	5
2026-09-17	21	5
2026-09-18	21	5
2026-09-19	21	5
2026-09-20	21	5
2026-09-21	21	5
2026-09-22	21	5
2026-09-23	21	5
2026-09-30	21	5
2026-10-08	21	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 84, unassessed 0. Authored assessments are not verified outcomes.

Capability	Nishi	DX	Faros	GitClear	CodeScene
GAUGE
tokens per closed rung with a paired quality floor, ratcheted Measured: `st_boot` exists in `runtime/_hdl_build/nx_seat.nx`, verified at emit. THE HEADLINE AND IT IS RED, and deliberately NOT flagged exceed -- the paired quality floor is the design, and with the ratchet red and the floor read from a single stamp it is not yet a proven superior instrument; a gauge that is losing does not get to claim it is better built. nx_seat shift reads Claude output per closed rung at 5,892,596 tokens over the last 30 days against 1,030,739 the prior window, ratchet RED because the cost of an outcome rose; the paired quality floor (gate GREEN-or-skip 675 of 685 on the roster stamp) keeps a token saving from reading as progress when quality falls with it, and a zero-closed-rung window reads UNMEASURED not a win. nx_eff_board prices it at 214 cents per action and 96 percent cache-hit, so the lever is context and cache, not output brevity [dx-ai-framework]. DX and Faros meter cost and throughput on separate dashboards from telemetry and surveys; none binds tokens-per-outcome to a quality floor, and the reason to measure at all is the field's RCT reality check [metr] Adoption: `RUN-BY:actlog` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	●	○	○
DEPLOY
the four DORA keys derived from the sovereign event stream, zero git-poll Measured: `d_band_deploy` exists in `runtime/_hdl_build/nx_dora.nx`, verified at emit. MEASURED live: deployment frequency ELITE at 50 per day, lead time ELITE at a median hour over 30 matched pairs, change-fail 42 percent carried as a coarse upper-bound PROXY, MTTR PENDING and shown PENDING rather than faked. DevLake and the platforms start the delivery clock at the first commit so a capability built and never deployed is invisible by construction [devlake-dora]; ours counts a DONE frame as a delivered capability, and the 2025 field finding that AI amplifies the underlying system rather than replacing it is the backdrop [dora-2025] 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); only referenced by nx_shipjoin_gate -- a mention, not a run. not 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); only referenced by nx_shipjoin_gate -- a mention, not a run Pros Implementing source symbol found; runtime acceptance separate; ahead of GitClear, CodeSceneCons 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); only referenced by nx_shipjoin_gate -- a mention, not a run	●	●	●	○	○
proof-carrying promote -- a capability-loss diff names every lost run before a binary goes live Source present; authored stronger claim: `cd_show_lost` in `runtime/nx_contentdiff.nx`, verified at emit. EXCEED with no competitor. A fitness function is an objective automatable integrity assessment of an architectural characteristic and our promote gate IS one by construction [fitnessfn]; every promotion diffs the staged binary against the live one and names each lost run. Declared blindness: a code-only delta with no literal loss is invisible to a string-run ruler, so the size and behaviour legs sit beside it. Nothing in DX, Faros, GitClear or CodeScene refuses a deploy on a measured property Adoption: `RUN-BY:fork:nx_rebuild_drain` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of DX, Faros, GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
served-versus-source drift census by remedy, a trajectory not a level Source present; authored stronger claim: `ps_judge` in `runtime/nx_artifactdrift.nx`, verified at emit. BEHIND, AHEAD and RESIZED-EQUAL partitions reconciled plus the INERT exec-bit axis, appended daily so the trend exists and not just the level. AHEAD is the one class where promoting DESTROYS capability, named rather than auto-fixed. CodeScene trends code health over history and is Best of the four at trend; none rebuilds source and compares it against the served bytes 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:clock` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of DX, Faros, GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	●
BUILD
first-try build success and the build-rework rate Measured: `ce_build_firsttry` exists in `runtime/nx_organ_ship.nx`, verified at emit. THE BUILD-PILLAR GAP. organ_ship.jrnl and the drift census carry every build and promote event, but no organ computes the first-compile-pass rate or the revert rate, and a byte DECREASE after an addition is a revert the estate already recognises by hand. GitClear measures two-week churn as the field's rework proxy, up another 15 percent in 2026 [gitclear26]. ACCEPT: two consecutive beats emit a first-try pass rate and a 14-day rework rate whose arithmetic reconciles against organ_ship.jrnl, UNOBSERVABLE on a single sample rather than a fabricated rate Adoption: `RUN-BY:clock` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	●	●	○
FEATURE-DEV
an operator goal mapped onto the boards from data, no frontier fan-out Source present; authored stronger claim: `gm_run` in `runtime/nx_goalmap_lib.nx`, verified at emit. LIVE 2026-09-05: one job turns goal words into rungs, status, ledger, gaps and a dependency order read straight from the boards, replacing a 5,713,788-token subagent fan-out that produced the same map. The mapping needs no language model -- it is term matching, a status join and a dependency sort -- so the estate plans a feature from its own capability graph where the field reads an issue tracker Adoption: `LIB-WIRED importers=13 nonval=10` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of DX, Faros, GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
the cheapest sufficient engine per task class, clearance-aware and judged Source present; authored stronger claim: `ed_dispatch` in `runtime/nx_engine_dispatch.nx`, verified at emit. LIVE: a task carrying a class and a byte-exact judge tries engines in ladder order -- local no-float, external free behind the warden, the frontier seat last -- and stops at the first the judge passes, so a C-class task never reaches an external model by construction. Prohibiting execution cost only about a point of resolve rate across 3,000 attempts, so the budget is a conf row not a default [runornot]. The field routes by spend budget; ours routes by clearance and a verifier 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). not 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 Source symbol found; stronger capability claim is authored; ahead of DX, Faros, GitClear, CodeSceneCons 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)	◉	●	○	○	○
workflow episodes mined from the action journal, not declared Measured: `al_mine_json` exists in `runtime/_hdl_build/nx_actlog.nx`, verified at emit. Recurring tool-and-verb n-grams are the estate's de-facto workflows, mined at action grain which is finer than commit grain. Faros mines repository and telemetry exhaust and is the nearest [faros26]; none mines the in-session action record for the workflow that produced the code Adoption: `RUN-BY:clock` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; ahead of DX, GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	●	○	○
BUGRES
the autonomous fix loop -- localise, propose, verify by executable oracle, revert proven exact DARK — `runtime/nx_autofix_auto.nx` EXISTS but does not declare `ce_fixloop_live`: something shipped at this path under another name, and this mark would read open forever. Repoint the contract to the real entry point (the plan rung AND this row) or rename the function; the flip follows on the next beat, and the `comparewatch-` plane row reads DARK until then. The loop is BUILT and its grader PROMOTED-UNREGISTERED: single-function scope, a fresh compile-and-run as the sole judge, sampled retries after greedy misses. The verify contract is SWE-bench's own, a patch resolved only when FAIL_TO_PASS flips and PASS_TO_PASS holds [swebench], and the honest next axis is the long-horizon multi-file set [swebenchpro]. ACCEPT: the loop is registered and callable, and one episode resolves a real debt-board task with its RED-before receipt banked, before this row flips dossierwatching ce_fixloop_live Pros none measured yetCons behind DX, Faros, CodeScene; open contract; expected source declaration not established	○	●	●	○	●
recurrence rate as the real bug-resolution effectiveness measure Source present; authored stronger claim: `ce_recurrence` in `runtime/_hdl_build/nx_debtmine.nx`, verified at emit. THE BUG-RES GAP, and the estate's own measured pathology motivates it -- the same defect filed five times in three weeks means resolution effectiveness is a recurrence-rate REDUCTION per defect class, not a ticket-close count. CodeScene surfaces hotspots and Faros tracks reopen rate on tickets; ours is per defect class with an executable oracle, so a resolved class emits a new gate or lint and the measure is its recurrence before and after. ACCEPT: two beats publish recurrence per class and a resolved class that recurs raises its own count 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:clock` — fully adopted (top of its ladder). dossier Pros Source symbol found; stronger capability claim is authored; ahead of DX, Faros, GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	●	○	●
the failure record classified against a taxonomy, gates that never once went red named Source present; authored stronger claim: `fc_alert_score` in `runtime/nx_failclass.nx`, verified at emit. LIVE: 880 verdict streams classified deterministically with a partition that sums, and nine gates that fired GREEN on every run of their entire history named as a bite worklist -- a perfect history is the absence of evidence, not health. CodeScene reads history for hotspots; none proves its own checks can fail 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:clock` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of DX, Faros, GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	●
non-vacuity by mutation -- the oracle is asked whether it can fail at all Source present; authored stronger claim: `gb_try` in `runtime/_hdl_build/nx_gate_bite.nx`, verified at emit. EXCEED: plants byte mutations, rebuilds AND redeploys the subject per mutant, restores by hash, and asserts the gate goes RED; mutation score is the test-quality standard the field names [pitest] and no platform mutates its own oracle. The estate advantage that every verify leg leans on: the judge is byte-exact, so a GREEN is a provable equality rather than a flaky suite Adoption: `RUN-BY:actlog` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of DX, Faros, GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
AUTHORING
full-fidelity human-and-AI session record harvested to sovereign frames Measured: `ch_frames` exists in `runtime/_hdl_build/nx_claude_harvest.nx`, verified at emit. The transcript parsed into action-grain frames -- every tool call and refusal with the reasoning that produced it, finer than any commit. Faros ingests assistant telemetry counters and is Best of the four at capture [faros26]; ours is the session itself, and it is the substrate the authoring-effectiveness join below needs 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); only referenced by nx_claude_harvest_gate -- a mention, not a run. not 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); only referenced by nx_claude_harvest_gate -- a mention, not a run Pros Implementing source symbol found; runtime acceptance separate; ahead of GitClearCons behind Faros (leads); 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); only referenced by nx_claude_harvest_gate -- a mention, not a run	●	●	◉	○	●
human-versus-AI NishiLang authoring joined to code health Source present; authored stronger claim: `ce_author_health` in `runtime/_hdl_build/nx_claude_harvest.nx`, verified at emit. THE KEYSTONE AND THE THING THE OPERATOR NAMED. GitClear measured across 623 million AI-era changes that block duplication rose 81 percent and refactored lines fell from 21 to 3.8 percent [gitclear26], and CodeScene's Code Red study found files with alert-level code health carry roughly 15 times more defects and cost 124 percent more time to change [codered22], but neither can attribute a NishiLang function to its author. The estate has the authoring record (ch_frames) and the health rulers (McCabe, forkgrade, slopmeter) and joins them nowhere. ACCEPT: a full-population join emits duplication, complexity and churn split by human-seat, AI-seat and generated authorship with the marker adoption published first, and an UNKNOWN bucket rather than a split that silently miscounts -- a single-sample corpus that cannot see its own AI-authorship trajectory is flying blind 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); only referenced by nx_claude_harvest_gate -- a mention, not a run. 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); only referenced by nx_claude_harvest_gate -- a mention, not a run Pros Source symbol found; stronger capability claim is authored; ahead of DXCons 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); only referenced by nx_claude_harvest_gate -- a mention, not a run	◉	●	◉	◉	◉
RIGOR
the complexity keystone is BUILT AND PROMOTED on the serving tree, not merely present in source Measured: `ce_mccabe_live` exists in `runtime/nx_mccabe.nx`, verified at emit. OPEN -- rung CE5. nx_catalog reads nx_mccabe and nx_mccabe_gate SOURCE-ONLY while /compare/codequality asserts them SHIPPED with the gate GREEN 17 of 17; nx_mccabe was BUILT through /api/build, PROMOTED, REGISTERED and re-measured full population on 2026-09-05 (19467 files, 112258 functions, above-bar 126 permille, partition sums) and nx_catalog now reads it LIVE on every axis; nx_mccabe_gate was BUILT on a later GRANT, PROMOTED, REGISTERED and returns GREEN 17 of 17 against the promoted binary with three bite-proven controls, so nx_catalog reads BOTH organs LIVE. DECLARED LIMIT: symbol presence cannot see a promote -- mc_scan_file has been in source the whole time -- so this watch flips only under CE4's catalog binding and reads open by construction until then, which is the honest form rather than a cell that could never flip. CodeScene is Best on shipped complexity measurement; the exceed we hold is that ours will REFUSE a build on it once a complexity ratchet lands rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them 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 separateCons behind DX (leads), CodeScene (leads); 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)	●	◉	●	●	◉
GAUGE
the effectiveness gauge is a beat with an axis-blind alarm so a RED can never go dark Source present; authored stronger claim: `ST_GAUGE_STAMP` in `runtime/_hdl_build/nx_seat.nx`, verified at emit. OPEN -- rung CE9 (repointed 2026-09-06 to the NAS-side writer: nx_seat st_shift writes knowledge/status/seat_gauge.stamp through ga_stamp_write and codeeffectiveness.gauge renders it; the laptop meter's sm_gauge_alarm is a sibling alarm the NAS cannot measure, so a watch on it could never flip). nx_seat shift reads RED today, and the estate has already paid once for this gauge going dark for 28 days and reading as no movement. A stale gauge must read STALE on this page and never as zero, and an abstaining axis must raise an alarm of its own rather than silently acquit. No field platform alarms on its own gauge going blind Adoption: `RUN-BY:actlog` — fully adopted (top of its ladder). dossier Pros Source symbol found; stronger capability claim is authored; ahead of DX, Faros, GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
RIGOR
a mirror whose canonical URL differs from the cited URL reads MOVED, never PROVEN Source present; authored stronger claim: `cr_canonical_moved` in `runtime/nx_compare_refs_gate.nx`, verified at emit. OPEN -- rung CE10, FOUND BUILDING THIS BOARD. The Faros /ai-impact page every other board cites now redirects to the Faros homepage: a fresh fetch returned 126648 bytes carrying none of the cited terms while printing status=200, the fetcher flagged it with a trailing 1 in its PROVENANCE line, and no consumer reads that flag, so a moved page can be pinned as evidence. Extend the incumbent refs gate with a canonical-URL check rendered in the same drift classes as STALE-UNDER; the Faros homepage mirror under its old URL is the planted RED and the same page under its canonical the planted GREEN. No field platform checks that a citation's mirror is the page it names Adoption: `GATE:LIVE trial=-` — fully adopted (top of its ladder). dossier Pros Source symbol found; stronger capability claim is authored; ahead of DX, Faros, GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
DUPLICATION
cross-tree fork and duplication graded over every pair Measured: `fg_trend` exists in `runtime/_hdl_build/nx_forkgrade.nx`, verified at emit. GitClear is Best on AI-era duplication because it publishes the TREND [gitclear26]; ours grades identity and superset over the whole corpus but as a LEVEL, and a trend needs two censuses in time -- the same gap codequality names as fg_trend, shared here rather than re-ruled 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 DXCons behind GitClear (leads); 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)	●	○	●	◉	●
RIGOR
base-class adoption, declared equals executed by construction Measured: `gl_strip` exists in `runtime/nx_gatelaw_gate.nx`, verified at emit. MEASURED live: 3,477 gate sources, 718 inherit the shared verdict base, 452 satisfy all three gate laws, 2,759 are hand-rolled and UNMEASURED by the exit-verdict and negative-control laws, and only 47 emit a measured value. The discipline that makes a green mean something sits at 20.6 percent base-class adoption and this board publishes the number rather than the badge 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 DX, Faros, GitClear; fully adopted on the estate ladderCons behind CodeScene (leads)	●	○	○	○	◉
a board effectiveness claim is re-measured against a live verdict, never hand-asserted Source present; authored stronger claim: `ce_claim_binding` in `runtime/nx_symdecl_lib.nx`, verified at emit. MOTIVATED BY A LIVE DEFECT FOUND BUILDING THIS BOARD. /compare/codequality asserts the cyclomatic-complexity keystone SHIPPED with its gate GREEN 17 of 17 and seq251 CLOSED, while nx_catalog on the serving tree reads nx_mccabe and nx_mccabe_gate SOURCE-ONLY -- built nowhere -- so the claim is unbacked on the surface that runs. devguardrails named this class as wp_gate_binding; the field disclosure standard is that harness and tooling configuration governs more variance than the model and must be published [harnessthesis]. ACCEPT: an effectiveness claim carries a gate or catalog binding re-read against the live artifact, and a claim the verdict contradicts publishes STALE-UNDER while one the verdict denies REFUSES to publish Adoption: `LIB-WIRED importers=31 nonval=29` — fully adopted (top of its ladder). dossier Pros Source symbol found; stronger capability claim is authored; ahead of DX, Faros, GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
every published effectiveness number carries its interval and a null control Source present; authored stronger claim: `ce_number_envelope` in `runtime/nx_gate_verdict.nx`, verified at emit. A replay script that never observes the work matched frontier agents on static benchmarks, so a number without a null control and an interval is not evidence [statprecipice]. ACCEPT: an effectiveness figure emits n, an interval and a null-control value beside it or it does not publish -- the envelope lives in the gate base class so this board and the others read one ruler Adoption: `LIB-WIRED importers=2826 nonval=854` — fully adopted (top of its ladder). dossier Pros Source symbol found; stronger capability claim is authored; ahead of DX, Faros, GitClear, CodeScene; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○

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	14	rows carrying a Yes, Best or Part in a rival column
cited	9	a reference mark resolves to a pinned mirror
uncited	5	an observation read, badged until a reference lands
rows, summed	14	RECONCILED: the parts sum to the total 14

Delivery and evidence

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

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
2026-09-06	CE9	land	LAST MILE SOURCE-LANDED (agreed shape: a NAS-side stamp derived from the rows the beat already pushes): nx_seat st_shift now writes knowledge/status/seat_gauge.stamp through ga_stamp_write -- gauge = claude out per closed rung, axes = meter log + ws journal (a window that read only one is BLIND), cadence = the window itself so a stamp older than two windows reads STALE; a window that closed nothing writes NO stamp (UNMEASURED is not a value) and the old stamp ages into STALE, the axis-blind alarm. The shift JSON gains gauge_stamp_wrote. Three CAS edits 203774 -> 204138 -> 204949 -> 205034 (two refused by ordering, re-issued under their predicted tokens). Build queued via /api/build under CPU saturation (run queue 18 of 8); after promote: run nx_seat shift, verify the stamp with ga_judge, then land the codeeffectiveness.gauge row for it.
2026-09-06	CE9	land	LAST MILE PROMOTED: nx_seat 360016 B sha 30309d65 via /api/build (src_path named, src_stable 1). Promote guard refused CAPLOSS; nx_contentdiff enumerated 4 lost runs (3 permil) and every one was adjudicated before the flag: manifest-shrank and prefix-mismatch ABSENT-PROVEN from all runtime sources (corpus_complete=1), docs-size-mismatch present only inside a gate COMMENT (no source emits it), ssl_total_keys_range a seg-store lib FUNCTION NAME the rebuilt call graph no longer reaches -- the seg-store's successor vocabulary (uncovered-vanished-keys, INCR-FOLD, tombstoned=) and CE6's ce_number_envelope are among the 53 gains: library evolution since the live binary was built, not this change. Promoted with allow_capability_loss=yes on that enumeration. Proof run (nx_seat shift 7) in flight; the codeeffectiveness.gauge row lands once the stamp exists.
2026-09-06	CE9	measure	THE BOARD READS IT: nx_compare_rank codeeffectiveness done=9 open=1 (CE7 alone remains, priority 300), unmapped=0 -- CE9 flipped once the rung and the matrix named the SAME NAS-side symbol (ST_GAUGE_STAMP in runtime/_hdl_build/nx_seat.nx) and the ranker probed the source itself. Two gauges now render on this page through gauge_pass: the one-step explained permil (daily) and tokens per closed rung (weekly, this stamp). STATE: shipped and proven, referee grade and operator ACCEPT owed.
2026-09-06	CE9	retract	THE GAUGE CELL NEVER RENDERED: the CE9 land row proved the STAMP WRITE (seat_gauge.stamp, ts/gauge/axes 2/2) and never the RENDER. The regen runs with CWD=buildroot and gauge_pass opened the stamp path as given, so knowledge/status/seat_gauge.stamp and knowledge/status/stepsolve.stamp (66 B at the serving root, ABSENT under buildroot) judged ABSENT on every publish. Discriminating experiment: nx_fs size of the stamp at both roots. Fixed in the emitter base (gauge_pass resolves through ep_artifact_path before ga_judge; matrix generator PROMOTED 422,144 B sha 4e99272e, ruler GREEN); the cell reads FRESH only after buildroot/_offc is refreshed and the regen runs -- until then this rung's page evidence is UNVERIFIED

Risk register

Risk	Likelihood x impact	Mitigation
An effectiveness number is published before the instrument is bite-proven and becomes the bar everyone plans against	likely x high	Every rung's accept rule requires a null control and CE6 makes the envelope a base-class refusal; an uncalibrated classifier reports numbers never verdicts.
The authorship split ships before the marker adoption is known and silently miscounts	likely x high	CE1 measures and publishes marker adoption FIRST and ships an UNKNOWN bucket, the same discipline codequality's CQ1a already pays for generated code.
CE4 is read as a mandate and every stale board claim refuses at once, taking the surface down	possible x high	CE4 publishes STALE-UNDER and only REFUSES a claim a verdict actively DENIES, never a merely stale one; the fix for a stale claim is the binding, not a takedown.
The gauge goes dark and a RED reads as no movement	certain x medium	CE9 stamps the beat and raises an axis-blind alarm, so a stale gauge reads STALE and never as zero -- the estate has paid for a dark gauge once already.
Closing CE5 is read as a licence to mass-build every SOURCE-ONLY organ	possible x medium	CE5 is scoped to the ONE keystone a published board already claims shipped; the record forbids the mass-rebuild reading of a large undeployed count.

Release history continued

When	Rung	Kind	Entry
2026-09-05	CE0	land	Domain codeeffectiveness authored and admitted 2026-09-05 -- matrix 17 rows composing live organs (nx_seat, nx_dora, nx_contentdiff, nx_artifactdrift, nx_goalmap, nx_engine_dispatch, nx_actlog, nx_failclass, nx_gate_bite, nx_claude_harvest, nx_forkgrade, nx_gatelaw_gate, nx_catalog) plus four watch contracts, plan 9 rungs, refs 14 pinned bars. Spendgate REVIEW name-free.
2026-09-05	CE5	measure	THE FIRST FINDING, MEASURED: nx_catalog reads nx_mccabe SOURCE-ONLY 5249 B and nx_mccabe_gate SOURCE-ONLY 9677 B on the serving tree while /compare/codequality asserts the complexity keystone SHIPPED with its gate GREEN 17 of 17 -- an effectiveness claim unbacked on the surface that runs. This is why CE4 binds claims to live verdicts and CE5 closes the gap.
2026-09-05	GAUGE	measure	nx_seat shift 30-day window: Claude output per closed rung 5,892,596 tokens vs 1,030,739 prior window, ratchet RED; done_ws 20, sessions 518; quality floor gate GREEN-or-skip 675 of 685. The RED is the number this board drives down.
2026-09-05	CE5	land	nx_mccabe BUILT through /api/build after the sovereign runner was refused by admission (VERDICT=QUEUE, run queue 22 on 8 CPUs, a GRANT sampled seconds earlier had already gone stale): 36760 B sha e2526b1a, src_path confirmed runtime/nx_mccabe.nx, src_stable 1, closure 7b23a7dd; PROMOTED prev=1 to nx_mccabe.elf under expect_sha256; /api/tools/register and nx_offc_install issued; nx_mccabe_gate build in flight through /api/build. The main-binary half of codequality's SHIPPED claim is now backed on the serving tree; the gate half follows and nx_catalog is the referee for both.
2026-09-05	CE5	measure	nx_mccabe RUNS on the serving tree, full population 2026-09-05: files=19467 funcs=112258 above_cited_10=14247 (126 permille) p50=3 p90=13 p99=46 max=2996 mean=6 unreadable=0, and the histogram partition SUMS EXACTLY to 112258. Against the 2026-08-20 census (18879 files, 103733 funcs, above-10 12639 = 121 permille, p90=12, p99=43) this is the FIRST two-point complexity trajectory the estate has: the corpus grew 8 percent in functions and complexity crept up on every percentile -- p90 12 to 13, p99 43 to 46, above-bar 121 to 126 permille. A level said 121; two levels say RISING. Registered (callable, grammar harvested, reconcile issued) and the _offc twin installed. Gate build queued as BQ-nx_mccabe_gate behind the I/O storm.
2026-09-05	CE10	measure	nx_compare_refs_gate 24 of 24 GREEN on the whole fleet after this board landed: provenance proven 848 mismatch 0 unproven 587 (baseline held), every pin matched, every mark resolved. It read 23 of 24 RED twice on the way -- first because 8 reused mirrors predated the provenance journal (fixed by re-fetching each under its own row rather than naming a residual), then because this seat cited dora-2025 under another board's URL while the mirror was journalled under the URL actually fetched (caught by the gate's own mirror-under-a-different-url neg-control, fixed by citing the fetched URL). The Faros /ai-impact page that devmgmt and others still cite is MOVED (it redirects to the homepage) and is the planted RED for CE10.
2026-09-05	CE0	lesson	THE PUBLISHER REFUSED THIS BOARD ON ITS FIRST REGEN, AND IT WAS RIGHT. nx_swcompare_matrix liar-kill read rows>=8 neg-control-zero exceeds<=flags cols>=4 all PASS and present>=5 FAIL: present_c counts rows measured-present WITHOUT an exceed flag, and this board had 11 measured rows with 7 flagged exceed, leaving 4 -- a board whose own headline gauge is RED claiming to exceed the field on most of what it measures is the liar shape the floor exists to catch. Re-graded honestly rather than argued with: the GAUGE row (tokens per closed rung with a paired quality floor) loses its exceed flag because the ratchet reads RED and the floor is read from a single stamp, so the pairing is a design, not a proven superior instrument. present_c 4 to 5, exceeds 7 to 6. The previous page (none) stayed unpublished; regen re-issued after the edit.
2026-09-05	CE0	lesson	TWO LEASE STORES UNDER ONE NAME, MEASURED 2026-09-05: nx_lease check compare-regen through nx_job_run (CWD = serving root) printed LS-FREE while nishi_compare_regen, which chdirs to buildroot, refused with LS-BUSY holder=regen-1788625797 seconds later -- the lease path is CWD-relative, so a lane probing from the serving root reads a confident false FREE for a lease that is held. Same class as the knowledge/compare/ bare-path defect the estate has now rediscovered five times; the remedy is the same resolver discipline (nx_comparetree_lib for compare paths, and a CWD-anchored lease root), never a second probe. Also measured: nx_swcompare_matrix run from the serving root reads every organ path ABSENT (coverage 0/1000) -- a census from the wrong CWD is a confident false zero, not evidence about the board.
2026-09-05	CE5	land	THE KEYSTONE IS CLOSED ON THE SERVING TREE (2026-09-05). nx_mccabe_gate BUILT through /api/build on a GRANT sample (51464 B sha efce9054, src_path confirmed), PROMOTED prev=1, REGISTERED, _offc twin installed, and run against the PROMOTED nx_mccabe: NX-mccabe passed 17 of 17 verdict=GREEN with three BITE-PROVEN controls (comment trap, string-literal trap, above-bar fires on 13 and is silent on 1), walk files=9 funcs=9 above10=1 max=13. Both halves of codequality's SHIPPED claim are now backed by nx_catalog LIVE verdicts; the CE5 watch cell stays open by construction until CE4's catalog binding exists, and says so.
2026-09-05	CE10	land	TWO BUS STOPS FIXED IN THE ESTATE, NOT NAMED (2026-09-05). (1) nx_lease check now probes all three lease roots the estate's two CWDs produce (knowledge/lease, buildroot/knowledge/lease, ../knowledge/lease) and a HELD stamp from any root wins, printing LS-ROOT so a reader knows which store answered; write side untouched; built 39792 B sha ebf93e4c, PROMOTED (live hash matches), selftest GREEN 9 of 9 including both concurrency teeth, _offc twin installed. (2) nx_swcompare_lib bar rows wrote their Theirs and Ours fields with the plain writer so every cite mark rendered raw on every board that carries bars (24 on engineshift alone); both fields now go through wnote, the cite-expanding writer the pos, goal, answer and biz rows already use. nx_swcompare_matrix rebuilt 368016 B sha 80d97b58 and PROMOTED prev=1 with its _offc twin; buildroot/_offc refreshed by nx_filecopy because nx_restage builds and admission is queueing (BQ-nx_swcompare_sota is queued for the sota generator). The redirect-hops consumer for the refs gate is landed in source as five sequential CAS edits (reader, pass hook, worklist, ratchet tooth, bite) and awaits its build.
2026-09-05	CE10	land	REDIRECT-HOPS CONSUMER LIVE IN THE REFS GATE (2026-09-05): nx_compare_refs_gate built 98480 B sha 592fa512 through /api/build, PROMOTED prev=1, _offc twin; first run on the fleet self-baselined REDIR-RATCHET at 33 -- thirty-three journalled-proven mirrors whose fetch followed a redirect, each NAMED in knowledge/status/refs_redirected.txt (extllm's OpenRouter pages at 1 hop, a Gemini terms page at 4) -- with the redirect bite BITE-PROVEN (hops 1 on the planted redirected row, 0 on the direct row, -1 on an unjournalled mirror) and the ratchet tooth PASS. The run also read 26 of 27 RED because THIS board's own journal row about the bar-mark fix contained a mark-shaped literal the gate read as an unresolved cite; reworded, re-run: 27 of 27 GREEN fleet-wide. Half two of CE10 -- canonical-url comparison that separates MOVED from a benign scheme or trailing-slash hop -- stays open; the 33 named rows are its worklist.
2026-09-05	CE0	queue	WHAT STAYS OPEN AND WHY (2026-09-05). (1) The publish of today's emitter fix and this board's closed-CE5 note waits on the array: nishi_compare_regen refused itself REFUSED-IO-STORM procs_blocked=7 of 8 and the compare beat defers the same way; re-issue on a sampled GRANT or let the beat take it -- the three generator roots are already byte-identical and fork-fresh. (2) CE10 half two: canonical-url comparison (rel=canonical or og:url in the mirror head, normalised, against the row url) so the 33 named REDIRECTED rows split into MOVED and benign hop; the Faros homepage mirror rf_ff06ac9a.raw under its old url is the planted RED. (3) ESTATE DEFECT FOUND BY A QUEUED BUILD: BQ-nx_swcompare_sota fired through nx_orchestrate and was refused rc=6 by build admission, because the deployq headroom predicate is memory-based (headroom:1024) while admission refuses on run-queue and D-state -- a queued build fires into the storm it was queued to avoid and then reads as a failed enqueue. Remedy is one ruler: the predicate consults nx_build_admit, never its own floor. Belongs to the build-queue owner (loadgov or engineshift); named here because this lane hit it. (4) CE1, the operator's keystone: first leg is the authorship-marker adoption census over the corpus (the accept rule's first clause), joining nx_claude_harvest frames to nx_mccabe per-function rows by file; the sovereignty lane landed the full-journal harvest read today, which is the substrate.
2026-09-05	CE0	retract	RETRACTION OF ITEM (3) IN THE QUEUE ROW ABOVE (2026-09-05): the build-queue predicate is NOT memory-based. Read from nx_orchestrate's own source: headroom:[floor[:max]] FORKS nx_build_admit check and is satisfied only when it GRANTS, so the queue and the runner share ONE ruler by construction; and the refusal-at-fire that BQ-nx_swcompare_sota showed is the documented volatile window (a GRANT at the poller pass, a QUEUE at the runner milliseconds later) which the orchestrator already handles by printing ADMIT-REFUSED-AT-FIRE and leaving the row queued. There is no predicate mismatch to fix. My earlier row published a hypothesis as a mechanism without reading the ruler -- the exact class this estate names as a correction that must pass its own experiment. The sota generator was rebuilt on a later GRANT and promoted, so nothing is owed on that row.
2026-09-05	CE10	land	refs gate canonical half LIVE 107936 B sha bdb5d8c6: rel=canonical or og:url read per redirected mirror, MOVED/BENIGN-HOP/CANON-UNKNOWN partition sums to redirected=33, worklist knowledge/status/refs_moved.txt, MOVED ratchet self-baselined at 1; fleet 29/29 GREEN
2026-09-05	CE10	measure	first canonical census: moved=1 (bevyengine.org to bevy.org) benign=0 unknown=32 of 33 -- and 30 of the 32 were the whole-file reader REFUSING files over its 16 KiB cap (CRG-READ-CAP-EXCEEDED), not absent tags: the detector was vacuous on real pages while printing a verdict per row
2026-09-05	CE10	lesson	a reader that refuses to truncate is right for a BODY and wrong for a HEAD: the canonical tag lives in head, so the honest read is a head window that reports whether the head CLOSED inside it; an absent tag is NONE only when the head was seen whole, otherwise TRUNCATED, never NONE
2026-09-05	CE10	land	crg_read_head plus three named negatives (CANON-NONE-IN-HEAD, CANON-HEAD-TRUNCATED, CANON-UNREADABLE in one partition bucket with a sum tooth) plus a canonical bite (moved vs benign hop, closed-head NONE, out-of-tag trap, 17 KiB fixture read in-window, beyond-window TRUNCATED, absent UNREADABLE) source-landed at 74519 B; build in flight
2026-09-05	CE10	queue	after the head-window build the MOVED count rises off a baseline the blind reader took: re-baseline the MOVED ratchet with that reason (instrument changed, not more moves) and work refs_moved.txt by re-citing each canonical url
2026-09-05	CE10	land	emitter: nx_swcompare_lib plan_pass now expands cite marks in rung done-rules, risk cells and debt cells (bar rows earlier); nx_swcompare_matrix 370184 c30b3d63 and nx_swcompare_sota 221792 ba51e306 promoted, _offc and buildroot/_offc twins byte-verified by receipt; regen REFUSED-IO-STORM procs_blocked=7, the beat re-runs when calm
2026-09-05	CE10	measure	second witness for debt 1788629238 (api-build-cache-stale-key, open, sev 6): /api/build of nx_compare_refs_gate returned action BUILT bytes 107936 sha bdb5d8c6 for src_sha 90d7783d after an admission-refused build, byte-identical to the artifact built from src_sha 8151ddf1 -- staged, fossil and live all read the OLD artifact while the source had grown 7.5 KB; the sovereign lane keys correctly but was REFUSED-BUILD-ADMIT (run queue 13 then 9 on 8 cpus), so the gate rebuild is paced behind admission, never routed around it
2026-09-05	CE10	land	head-window canonical reader LIVE (nx_compare_refs_gate 115952 B sha dc2eb3cc, debug flavour to match the /api/build live, contentdiff GREEN strict superset 0 lost 45 gained, prev bdb5d8c6 banked); canonical bite 8/8 including a 17504 B fixture read in-window that the cap reader refused, beyond-window TRUNCATED, absent UNREADABLE
2026-09-05	CE10	measure	first TRUE canonical census over the 33 redirected mirrors: MOVED 21, BENIGN-HOP 5, CANON-NONE-IN-HEAD 5, CANON-HEAD-TRUNCATED 2 (sums to 33; the blind reader had said 1/0/32). MOVED names a re-cite worklist in knowledge/status/refs_moved.txt: openrouter docs restructured, kubernetes and aws pages moved, huggingface xet, metahuman, nvidia ace, dora.dev research pages now the homepage (twice), nar.realtor, nats leaf nodes, an aws page only on web.archive, bing visual search, ms servicing, two PMC articles on the new pmc host, unrealengine features, bevy.org x3 (one landing on a quick-start page, not the cited features page)
2026-09-05	CE10	lesson	the fleet gate went 36/38 RED for two honest reasons: the MOVED ratchet rose 1 to 21 because the INSTRUMENT changed, and a gameengine log row carrying WGSL attribute syntax read as a dangling cite mark. Neither is loosened: the ratchet is re-baselined by a written, announced directive (the conf name is retire-protected, so a bare number would have been laundering), and the mark scanner now requires a well-formed key so code that merely starts with the two mark characters is prose
2026-09-05	CE10	queue	21 MOVED rows to re-cite under their canonical urls with fresh provenance (re-fetch under the same key, re-pin, url field) -- a worklist consumer organ rather than 21 hand edits; scanner fix + directive reader source-landed, build paced on admission
2026-09-05	CE10	retract	my 2026-09-05 nx_lease check fix (3-root probe, selftest 9/9) did NOT work on the live surface: the direct MCP call from the serving root printed LS-FREE against a held compare-regen stamp the shell tool could read at buildroot/knowledge/lease/compare-regen.lock/stamp, while a lease in the serving-root store WAS found -- so the two inline-literal roots never opened and the selftest was a fixture the defect could not fail
2026-09-05	CE10	land	nx_lease 42784 B sha 5e6cdaef promoted (prev banked, _offc twin): roots hoisted to consts, the probe is one function main and the selftest share, selftest 11/11 with a cross-root tooth (acquire under the buildroot root, probe must find it there and read it HELD) plus a neg-control; live check now prints LS-ROOT buildroot/knowledge/lease/ and the holder stamp. The compiler-side finding (inline literals absent from the old binary as printable runs) is filed on the lang lane as a debt row with the repro recipe, mechanism deliberately unnamed
2026-09-05	CE10	land	the last 9 raw cite marks on /compare/engineshift were the matrix @verdict paragraph: both generators' verdict renderers now expand marks (nx_swcompare_matrix 379232 56997071 and nx_swcompare_sota 232064 9c7b2c5f live, both _offc and buildroot/_offc twins hash-identical); a beat regen holds compare-regen since 1788632942 and will publish it
2026-09-05	CE10	land	nx_refdrift gained a recite verb (spendgate: name free, incumbent extended rather than a new organ): `nx_refdrift recite <moved-worklist> [max] [apply] [key=<k>]` reads the refs gate's MOVED worklist, fetches each canonical into the probe, refuses a body the completeness classifier calls truncated or suspect (mirror untouched), dry prints WOULD-RECITE, apply re-fetches under the mirror's own name for fresh provenance, hashes with the one hasher and rewrites url, pin and accessed by walking pipes with temp plus rename; envelope partition sums, PARTIAL declared. Built 65648 B sha bb9c81ba (debug flavour), contentdiff GREEN strict superset, promoted prev=1, _offc twin; first dry run over all 21 rows in flight
2026-09-05	CE10	measure	recite dry run over all 21 MOVED rows: 21 fetched, 0 refused, 17 complete bodies and 4 unknown-class html, accessed date derived 2026-09-05, partition sums, PARTIAL=0. Adjudicated before apply: 5 rows canonicalise to a homepage or an unrelated page (dorakeys and dora-fourkeys to dora.dev, metahuman to metahuman.com, bevy-home to bevy.org//, bevy-features to a quick-start intro) and stay for a human read; 16 are the same document at a new address
2026-09-05	CE10	land	16 rows RECITED under their canonical urls with fresh provenance and pins (orlimits, orrouting, k8s-probes, arc-dataplane-recovery, fespa, hf-xet, nvace, narfsbo, nats-leafnodes, aws-backoff-jitter, bing-vs, secureboot-expiry26, nac-review21, nac-tattoo24, unreal-features, bevy-news), applied in three batches so no two rows rewrote one register at once; one PMC fetch returned an empty body and was REFUSED with the mirror untouched, then recited on retry
2026-09-05	CE10	lesson	the verb's first live batch found its own defect: an apply with one refusal and nothing recited printed verdict=RECITED -- a verdict named for the MODE, not for what happened. Fixed: NOTHING-RECITED (exit 1) and RECITED-WITH-REFUSALS are their own words; and rows now flush as they are decided so an interrupted run keeps its rows
2026-09-05	CE10	land	refs gate 117696 B sha a6e62ed8 live: url equality now collapses a RUN of slashes to one and drops a trailing run (bevy.org// is the site's own canonical tag, not a move), two teeth added, gate 40/40 GREEN; /compare/engineshift published with 0 raw cite marks after the beat regen picked up the fixed verdict renderer (171885 B)
2026-09-05	CE10	land	the 5 held MOVED rows closed by a human read plus the same mechanism: a durable manual worklist knowledge/status/refs_recite_manual.txt carries the chosen url per row and nx_refdrift recite applies it -- metahuman to https://www.metahuman.com/ (the Unreal page redirects there in 3 hops, same body), dorakeys and dora-fourkeys to https://dora.dev/guides/dora-metrics/ (the four-keys guide moved; zero-hop fetch, identical 19454 B body both ways, one pin for both rows), bevy-home to https://bevy.org/ and bevy-features to https://bevy.org/learn/quick-start/introduction/ (domain rebrand, same pages). moved 21 -> 0 owed on the next gate run
2026-09-05	CE10	queue	two mgmt-lane build refusals were absorbed into the deploy queue as BQ-nx_refdrift and BQ-nx_compare_refs_gate; both organs have since been built through the sovereign lane and promoted from source identical to what those rows would compile, so when they fire they produce a content-identical binary and, via the queue's serving-root CWD, may redeploy it and overwrite the .prev rollback slots (bb9c81ba for nx_refdrift, dc2eb3cc for the gate). Not lost work: the rows are redundant. A queue verb to withdraw a row whose target was promoted since it was queued is the missing mechanism (nx_buildq has close but no withdraw-by-name from the seat surface)
2026-09-05	CE1	queue	next rung on the critical path (CE5 done, CE1 next, then CE4): the marker-adoption census -- classify every runtime source by its head marker (the GENERATED idiom nx_unwired already reads) and join per-file cyclomatic complexity from mc_scan_file, print the partition with sums and a time-spine row; uw_headmark lives inside a program, so the first step is an extraction to a lib (nx_oo_extract) rather than a second head scanner
2026-09-05	CE10	lesson	the redirect ratchet fired on a row another lane registered today (fieldwatch g2-scoring, hops=1, canonical OG-URL-ONLY): the raw redirect total rose 33 to 34 while moved stayed 0. Measured on the same run, 20 of the 34 redirected mirrors are BENIGN-HOP (the page's own canonical equals the cited url after normalisation), so a ratchet on the raw total carried 588 permil false positives on real data and would have gone RED for every lane that cites through any scheme, www or slash hop. A detector with false positives teaches every lane to ignore it: the SUBJECT was wrong, not the number, and the remedy is neither a rebaseline directive nor a loosened bar
2026-09-05	CE10	land	nx_compare_refs_gate: the ratchet subject is now the OPEN set (redirected minus benign-hop, which equals moved plus every canon-unknown reason); worklist knowledge/status/refs_redirect_open.txt carries the class ON each row; conf refs_redirect_open_ratchet.conf self-baselines on first sight (14 today: none_in_head 5, head_truncated 3, og_url_only 6, moved 0). The raw total stays censused in refs_redirected.txt and is emitted as values redirected_total, redirected_benign_hop, redirected_open via gv_kv. Three new teeth test the state not the message: open equals moved plus canon_unknown, and each written worklist's row count equals its counter. Source landed at 82966 B in four CAS edits (the file is mixed CRLF and LF, 923 CR lines, so a newline-terminated anchor NOMATCHes on the old body); first build refused by admission (procs_blocked 15 against 8), paced re-issue owed, then promote, offc twin, and retire the old refs_redirected_ratchet.conf (held 33) with its at_baseline snapshot
2026-09-05	CE1	queue	DESIGN SETTLED FROM THE INCUMBENTS, not from taste: the contract is the AUTHORING watch row (symbol ce_author_health on runtime/_hdl_build/nx_claude_harvest.nx). Leg 1 GROWS THE INCUMBENT instead of adding a second scanner -- nx_mccabe_lib's mc_scan_file already reads every source once, so the head-marker class becomes one call on the buffer it holds, with per-class blocks (files, funcs, sum vG, above the cited bar) in the same c[] the census and its gate share; the class table is DATA at knowledge/headmark_classes.conf (landed: BUILDER, GENERATED, DERIVED; UNMARKED is the named remainder), nx_mccabe prints the partition with sums plus one time-spine row, nx_mccabe_gate gains planted fixtures per class. The head scanner is uw_headmark inside nx_unwired (a program), so nx_oo_extract moves it to nx_headmark_lib FIRST and the lib edit imports it SECOND -- landing the edit before the extraction would leave a shared source that does not compile. Leg 2 joins nx_claude_harvest's session attribution (which files a seat wrote) to those per-file figures and ships ce_author_health. spendgate REVIEW name-free (nx_adopt counts callers, nx_mccabe measures complexity; neither classifies by marker). Both legs wait on build admission: every build this hour was refused (procs_blocked 15 then 10, load 23-30 on 8 cpus); the refs gate build sits absorbed as BQ-nx_compare_refs_gate status queued
2026-09-05	CE10	land	nx_compare_refs_gate LIVE 93a8beee (121232 B, debug flavour like its predecessor; the first non-debug rebuild came out 106763 B, 11709 B under live after an addition, and the flavour explains the whole delta, so the same-flavour build is the one adjudicated). contentdiff RED on exactly five runs, all my own renames (three REDIR-RATCHET messages, the old count sentence, the old tooth name), each superseded by an OPEN-REDIRECT twin among the 19 gained, so allow_capability_loss was passed with that enumeration and no vintage control was owed (no source-path class lost). Promote landed under a dropped 503 receipt; the retry answered NOTHING-STAGED with live_sha256 equal to the build, offc twin installed, both roots 93a8beee. Direct run: NX-comparerefs passed 44/44 verdict=GREEN, open ratchet SELF-BASELINED at 14, values redirected_total=34 redirected_benign_hop=20 redirected_open=14, moved=0. Old ratchet: at_baseline snapshot retired reversibly (knowledge/retired/...-1788640040), the retire-protected .conf now carries a written RETIRED directive naming its successor
2026-09-05	CE10	measure	UNATTRIBUTED, recorded so nobody inherits it as a fact: nx_behaveprobe live-vs-staged reported other_rc=1 for the staged gate with 16264 B of output, while the direct run of the same bytes minutes later read GREEN 44/44 at 16391 B and a later live-vs-prev probe read live_rc=0 at the identical 16391 B. The probe runs its two subjects in sequence (tr_run_capture_cwd, one after the other), so the clobber-the-shared-worklist theory is refuted; byte arithmetic (the GREEN verdict note is roughly 380 B the RED line lacks, the first-sight baseline line roughly 90 B) leaves about 160 B of unexplained extra output in the staged run, the size of one transient UNRESOLVED-MARK or UNCITED row from a sibling lane's edit in flight across the fleet scan. The probe keeps no capture, so this stays a hypothesis; the remedy that would settle it is a probe that writes both captures beside its verdict
2026-09-05	CE10	queue	regen of /compare owed for the fixed verdict-paragraph cite rendering and this board; the compare beat publishes on its own cadence and a seat launch under the storm would be refused or would hold the compare-regen lease against the beat, so the beat is left to publish
2026-09-05	CE1	lesson	the first nx_oo_extract of uw_headmark,uw_generated out of nx_unwired REVERTED (lib written at 630 B, consumer 28444 to 28000 with the import line, AFTER build at 1788640276, then both restored): the behavioural proof's workload scanned buildroot/runtime/_hdl_build, the directory that HOLDS the consumer's own source, so moving two definitions out of it legitimately changed the unwired census the proof compares. A BEHAVIOURAL PROOF WHOSE WORKLOAD READS THE CONSUMER'S OWN SOURCE REFUTES EVERY CORRECT EXTRACTION -- the subject must exclude the files the extraction rewrites. Re-run over /tmp/mccabe_gate (nine planted fixtures, neither consumer nor lib inside), launched through nx_job_run so the receipt lands in /tmp/oo_headmark1.txt where a dropped edge window cannot lose it
2026-09-05	CE1	measure	second extraction attempt: OO-BASELINE captured (admission granted for the BEFORE build), the lib was written at 630 B, then the AFTER build was REFUSED by admission and the tool refused the extraction and restored everything -- the right direction. Its verdict line reads behaviour-NOT-preserved while its own preceding line says unverifiable-this-run: an admission refusal named as a behaviour change is a message naming the wrong stage, and a reader who trusts the verdict line would go looking for a real behavioural difference that never existed. Not my organ; named here for its owner. Under an oscillating box the extraction needs two admitted builds inside one run; third attempt launched through the job lane
2026-09-05	CE1	land	third attempt EXTRACTED: uw_headmark,uw_generated moved to buildroot/runtime/nx_headmark_lib.nx (630 B), nx_unwired rebuilt and every workload run OUTPUT-IDENTICAL (behaviour preserved over the declared workload). nx_unwired stays BUILT-UNPROMOTED as it was; a refactor nobody asked to deploy is not promoted. The lib is the head-marker primitive the complexity scanner will import for the per-class join
2026-09-05	CE10	land	the watch symbol is now REAL CODE on the load-bearing path: cr_canonical_moved(cl, canon, url) -- 1 iff a readable canonical, normalised, names a different page than the cited url -- is defined after crg_url_norm_eq and is the predicate the row loop's MOVED-or-BENIGN decision calls (definition plus one call, count 2). contentdiff ONE-WAY-GAIN gained 1 lost 0, promoted e1fedff3 (121456 B, prev banked), offc twin installed, both roots verified by hash. The compare cell flips on the next regen beat because the capability exists, not because a name was typed
2026-09-05	CE10	measure	ATTRIBUTED, superseding the earlier UNATTRIBUTED row: the live gate run on e1fedff3 read 43/44 RED on cite-marks-resolve-every-one with UNRESOLVED-MARK file=buildroot/knowledge/compare/intelmine.matrix key=semeval14-paper -- the intelmine lane's mark landed before its refs row, fleet rows 1557 to 1560 between runs. That is the roughly-160-byte extra line the probe's staged run carried, so the probe's rc=1 was this same sibling-lane transient, not a defect in the staged gate; every redirect, open-ratchet and worklist tooth reads PASS. Not my row; the gate names its owner and clears itself when the ref lands
2026-09-05	CE1	land	LEG 1 LIVE -- the complexity scanner now joins every source's HEAD-MARKER CLASS to its complexity, inside the incumbent: nx_mccabe_lib gained the class table loader (mc_cls_load_from, DATA at knowledge/headmark_classes.conf, order is priority, UNMARKED is row 0), mc_headclass (composes uw_headmark from the extracted nx_headmark_lib), per-class blocks in the shared c[] (MC_C_SLOTS, files/funcs/sum vG/above-bar), and mc_scan_file classifies once from the buffer it already holds; a consumer that never loads a table sees classification off. nx_mccabe prints the partition with its sums and appends ONE row to knowledge/status/mccabe.spine for the production subject only (a trial dir announces SKIPPED -- the forgeable-heartbeat law). nx_mccabe_gate gained seven teeth on five planted fixtures under /tmp/mccabe_gate_cls with a PLANTED two-row table: two-row load, all five reached, GENERATED row carries vG 2, both-markers reads the FIRST row, a marker beyond the 400-byte head window does NOT classify (a whole-file grep dies there), partition sums to the walk, absent table writes no class block. Promoted debug-flavour after a non-debug build read as a flavour mismatch: gate 396fb82b 61648 B (24/24 GREEN live, class join rows=2 files=5 UNMARKED=2 BUILDER=2 GENERATED=1), census e447aba5 47848 B, both twins by hash, both rulers ONE-WAY-GAIN
2026-09-05	CE1	measure	FIRST SPINE ROW, production tree, partition SUMS: files=19505 funcs=112660 above_10=14293 (126 permil). UNMARKED (seat-authored, human or AI at a seat) 17319 files / 107244 funcs / 14112 above = 132 permil; BUILDER (organs the builder signed) 246 / 599 / 92 = 154 permil; GENERATED 1939 / 4812 / 87 = 18 permil; DERIVED 1 / 5 / 2. Seat-authored code carries 987 permil of every function above the cited bar; generator output runs at one seventh the seat rate. The human-versus-AI split INSIDE the seat class is leg 2 (ce_author_health via nx_claude_harvest); this row is the denominator it joins against, and the next census run turns a level into a trajectory
2026-09-05	CE1	measure	BITE-PROVEN: nx_gate_bite nx_mccabe_gate over nx_mccabe_lib.nx, 2 comparison sites -- site 0 NOT-REACHED (excluded, a mutation that never entered the binary), site 1 KILLED (the gate went RED on a compiling mutant), valid_mutants=1 killed=1, source restored byte-identical at 17135 B, both twins restored from the banked digests after the cleanup build was refused by admission (mutant_live=0). The production census printed partition SUMS and spine fd=4 wrote=158 of=158. One row worth a human read: DERIVED files=1 funcs=5 mean_vG=17 permille=400 -- a single NX-DERIVED .nx source in the runtime tree carrying two functions above the bar, i.e. a generated artefact that is also compiled source; it is named by class now and was invisible before
2026-09-05	CE1	queue	LEG 2 PRECONDITION MEASURED, NOT ASSUMED: ce_author_health needs the set of files each seat session wrote, and the NAS actlog frames do not carry it. Every harvested write frame in knowledge/status/chbeat/chbeat_20260905.jrnl reads ts TAB session TAB tool TAB use TAB - TAB delta (735 nx_fs_write frames today, plus Write and Edit frames, none with a path); the path exists only in the transcript argv on the laptop, which the frame parser ch_frames drops. The join cannot be faked from data that lacks its key. BUILD ORDER: (1) an ADDITIVE note field on write frames -- ch_frames extracts the path argument of nx_fs_write, Write and Edit tool uses and writes note=path=<p> (existing readers see one more token, no field moves); (2) the laptop beat pushes the new frames as it already does, so the NAS record fills forward from the day it lands; (3) ce_author_health in nx_claude_harvest walks the per-day journals for path-bearing write frames, keys each path to its session, scans it with mc_scan_file, and prints seat-written versus not-seat-written figures against the mccabe spine denominator, with the unattributed remainder named. A seat is AI at a seat by definition of the transcript, so the human half is the complement inside UNMARKED, and the emitter must say that rather than claim a per-human attribution it does not have. Owner: the harvest is portable and its beat runs on the laptop, so the parser change ships through the laptop twin and the SSOT under _hdl_build together
2026-09-05	CE1	land	LEG 2 PRECONDITION SHIPPED: nx_claude_harvest write frames now carry the path. The parser gained ch_tool_writer_kind (nx_fs_write -> argv[1], Write and Edit -> file_path), ch_json_str_copy, ch_tool_path (bounded 4096-byte window after the tool's name, first match is the input's own key) and ch_frame_note, and BOTH loops (frames and delta) compose it: the note gains one space-separated token `path=<p>`, no column moves, a non-writer frame is byte-identical to before. Three selftest teeth on a fresh fixture: T45 argv[1] path with the verb never mistaken for it, T46 file_path with a decoy key inside the content never read as the path, T47 neg-control a reader frame carries no path token (exactly 2 of 3 do). NAS: contentdiff ONE-WAY-GAIN gained 17 lost 0, promoted 16e6fbfe (223208 B, debug flavour like live, prev banked), offc twin installed, LIVE selftest 53/53 GREEN. Laptop: the same seven edits applied to the twin, and its sovereign build came out BYTE-IDENTICAL to the NAS non-debug build (198467 B, sha eabec821...) -- the closure equality proof the estate's own ruler gives for free. The laptop beat forks the laptop _offc twin, refreshed through the builder's --install lane, so every beat from now lands path-bearing frames and the NAS record fills forward; frames before this hour stay pathless and the join must name them as UNATTRIBUTED-BEFORE-<epoch>, never guess them
2026-09-05	CE1	land	LEG 2 LIVE -- ce_author_health is DECLARED in nx_claude_harvest and called by its authorhealth verb (the watch's symbol ruler applies the declaration rule, so this is what flips the AUTHORING cell). It reads a per-day journal, keeps every UNIQUE path from path-bearing write frames (a duplicate write counts once), scans each .nx under the tree prefix ONCE with mc_scan_file (the one ruler the census and its gate share), sets the figures against the mccabe spine's LAST row, and names the complement not_seat_attributed (human-authored PLUS every write before the first path frame). Three named refusals: UNOBSERVABLE-DENOMINATOR (rc 3, no spine), NO-PATH-FRAMES (rc 4), and a path outside the tree or not .nx counted apart. Teeth T48-T51 on a planted tree, journal and spine (dedupe, partition, absent spine, pathless journal). Built first on the laptop, where the compiler caught consts read above their declaration (single-pass parser; consts moved to the top beside the beat consts) and the selftest caught sj_atoi_z concatenating every digit up to the NUL (files=10
2026-09-05	CE1	measure	FIRST AUTHOR-HEALTH MEASUREMENT (chbeat_20260905.jrnl, frames=12606, path_frames=4 in the hour since the parser landed, sessions=4): seat_written files=3 funcs=102 sum_vG=777 above_10=19 (186 permil) against the spine denominator files=19505 funcs=112660 above_10=14293; not_seat_attributed files=19502 funcs=112558 above_10=14274 = human-authored PLUS every write before ts=1788647757, UNATTRIBUTED by construction. This is the first hour of a record that fills forward on every beat; the number is honest and small, and it becomes a trajectory as the beats accumulate. The three in-tree files were nx_wgsl.nx, nx_manga_script_gate.nx and nx_content_put_client.nx (three other seats); the fourth path was this board's .plan, outside the tree
2026-09-05	CE1	retract	the probe disagreements are ATTRIBUTED and FIXED, superseding the earlier UNATTRIBUTED rows: nx_behaveprobe read this organ's staged selftest RED (rc 1) twice while every direct run was GREEN. Mechanism, read from the source: ch_selftest stamped its scratch names with sys_now_realtime_sec while the beat already stamps its pieces with sys_now_realtime_us for exactly this reason; the probe runs live then staged inside one second, so the staged run found the live run's T15 marker already seeded and its T14 refused. One-line fix (the selftest now stamps in microseconds), proven by the instrument that had disagreed: contentdiff IDENTICAL-CONTENT, nx_behaveprobe live_rc=0 other_rc=0 on the same argv that had read 0/1. A SELFTEST WHOSE FIXTURE NAMES ARE UNIQUE PER SECOND IS NOT UNIQUE PER RUN UNDER ANY CALLER THAT RUNS IT TWICE, and the ship probe is such a caller. Promoted c5534bd2 (257552 B, prev banked); laptop twin rebuilt and installed (229678 B non-debug 8de305b8)
2026-09-05	CE5	land	the watch row keyed on ce_mccabe_live, a symbol nobody had declared while the keystone was already promoted and writing its spine -- the same built-but-unnamed gap CE10 had. Named the real thing, not a badge: the spine-row writer in nx_mccabe (the block that appends the trajectory row for the production subject and announces fd= wrote= of=) is now func ce_mccabe_live(c, funcs, crows, ob, a), called by main; the census's own trajectory writer is what proves the keystone runs on the serving tree. Source landed (9922 B); build refused by admission while the beat regen holds the box, paced re-issue owed, then ruler, promote, twin. cr_canonical_moved already reads LANDED on the comparewatch plane; ce_author_health reads OPEN because the running regen re-put the plane before the declaration landed, and the next beat takes it
2026-09-05	CE5	land	ce_mccabe_live LIVE: promoted 37d9ed61 (48232 B, contentdiff ONE-WAY-GAIN gained 1 = the symbol's own name, lost 0, prev banked), offc twin by hash, live run prints spine=knowledge/status/mccabe.spine fd=4 wrote=158 of=158 from the renamed writer and the partition SUMS (class_files 19511 of 19511, class_funcs 112818 of 112818). The keystone's cell flips on the next beat regen for the same reason the authoring cell does
2026-09-05	CE1	land	WATCH FLIP WITNESSED on the comparewatch plane (job 1788649580, 306644 B): codeeffectiveness_ce_author_health LANDED and codeeffectiveness_ce_mccabe_live LANDED after the beat regen that held lease regen-1788648455; the published page follows on its next publish. PRESENCE ONLY: no referee grade and no ACCEPT row yet, so CE1 and CE5 stay INCOMPLETE on /compare under the three-signature law.
2026-09-06	CE8	land	fg_trend landed in buildroot/runtime/_hdl_build/nx_forkgrade.nx (20,394 B): one spine row per SUCCESSFUL census, appended only after the partition is proven sound; the path is DERIVED from dirA (production knowledge/status/forkgrade.spine iff dirA is buildroot/runtime, else a <dirA>.forkgrade.spine sidecar) so a trial run cannot forge the production trajectory; announced spine= fd= wrote= of=. nx_forkgrade_gate.nx created (10,500 B): three planted pairs identical/A-superset/bidirectional with hand-computed answers (adoptable 666 permil), two runs leave two rows, neg-control that the production spine size is unchanged, argv subject override, RED by construction against the pre-trend binary
2026-09-06	CE8	queue	both builds storm-deferred: nx_build_admit read GRANT at procs_blocked 2 of 8 and 40 s later the sovereign lane was refused QUEUE at 8 of 8 (the D-state witness, not load). Recipe: nx_sov_build_run nx_forkgrade --build-only, then nx_forkgrade_gate --build-only, paced a minute apart; run the gate via nx_job_run from the serving root with the _build subject (RED on the promoted binary, GREEN on the staged one is the proof); promote both with expect_sha256, register the gate, nx_offc_install both; then the watch _ABSENT_:fg_trend flips on the next admitted regen
2026-09-06	CE0	queue	OPERATOR DIRECTION 2026-09-05, estate-wide and NOT this lane's build: references sent mid-lane are logged on /compare as pinned refs AND get wired as benchmarks and oracles that a nishi container spins up and down the way an image does under docker -- GitHub repos and other sources as sandboxed oracle images. Incumbent named by nx_capsearch: runtime/nx_container.nx (sovereign container runtime, Docker replacement: nx_container_run, nx_container_verdict_is_valid), so the direction is an ADOPTION rung on the board that owns that lib (sovereignty or devmgmt), shaped as an oracle-image contract: fetch and pin the repo, materialise it in nx_container under a declared envelope, run its bench, land the numbers as .bench receipts beside the refs. Filed on this board so it survives a crash; deliberately not started here
2026-09-06	CE8	land	CE8 PROVEN END TO END on the served artifact 2026-09-06: nx_forkgrade (fg_trend) staged by /api/build at 44,656 B sha c43c490d (src_path names the target, src_stable=1), nx_contentdiff live-vs-staged GREEN lost_from_live=0 ONE-WAY-GAIN (78 runs gained including the spine literals), PROMOTED with expect_sha256, prev banked, nx_offc_install refreshed _offc to the blessed bytes; nx_forkgrade_gate (43,824 B) read RED 4/12 against the OLD served twin (census GREEN total=3, only the eight trajectory teeth failed: the pre-trend binary writes no spine) and GREEN 12/12 against the NEW one (sidecar holds 2 rows after two runs, adoptable 666 permil hand-computed, production spine size unchanged, bite-proven). Owed: stage, promote and register the gate itself; the production spine knowledge/status/forkgrade.spine has no row yet and gets its first point on the next real census over buildroot/runtime (nx_forkgrade is REGISTERED-DARK, its pinned census args unknown to this seat; a trend needs two censuses in time); referee grade and operator ACCEPT
2026-09-06	CE6	land	ce_number_envelope, ce_env_refuse and the CE_ENV_* codes landed in the gate base class nx_gate_verdict.nx (38,635 to 40,835 B, exactly one insertion by size arithmetic): a figure publishes as key=v n= lo= hi= null= only with n above 0 and lo at or below v at or below hi; every other shape prints REFUSED-BARE-NUMBER, REFUSED-NO-INTERVAL or REFUSED-INTERVAL-EXCLUDES-VALUE BY NAME in place of the value and returns its own code. nx_number_envelope_gate 12/12 GREEN, bite-proven (a bare percentage refused, the well-formed figure silent), built 29,482 B; staging through /api/build in flight. Owed: promote and register the gate, migrate this board's own emitted figures onto the envelope, referee grade, operator ACCEPT
2026-09-06	CE2	land	nx_build_firsttry_lib (16,617 B) is the ONE ruler: episodes classified by their FIRST BUILD row into first_pass, code_fail and box_refused (partition printed), 14-day rework is a same-target re-ship inside the window, a revert is a bytes= decrease, a pair lacking bytes= is revert_unobservable (counted, never guessed), an empty window reads UNOBSERVABLE never 0 permil, one spine row per MEASURED run at a path derived from the journal; nx_build_firsttry_gate 33/33 GREEN on six planted journals with hand-computed answers, the known revert moves rework AND reverts by exactly one, two bites proven, built 55,012 B. nx_organ_ship took six CAS edits (63,149 to 65,952 B, every predicted size exact): the import, os_build_refusal_class plus os_refusal_head plus os_bytes_of, a BUILD row journaled on an /api/build refusal (class plus a 160-byte reason head) and on WRONG-SRC, bytes= on the SHIPPED row, and the firsttry verb ce_build_firsttry. Compile-proof of the edited ship loop in flight; then contentdiff, promote, and firsttry over the production journal for the first spine row
2026-09-06	CE0	measure	2026-09-06 state after the three land rows: nx_compare_rank reads rungs=10 done=7 open=3 unmapped=0 (open: CE7 ce_fixloop_live #1 priority 300, CE3 ce_recurrence, CE9 sm_gauge_alarm); CE2, CE6 and CE8 flipped on symbol presence with their gates GREEN (33/33, 12/12, 12/12) but NONE is complete under the three-party law: referee grade and operator ACCEPT rows are still owed for CE1, CE2, CE4, CE5, CE6, CE8, CE10. Fleet refs gate 43/44 after the operator refs: my rows clean (pins 1515 of 1515 matched, unresolved marks 0 after the dcc cross-domain rows and the intelmine prose reword), the one RED tooth is declared-refs-cited-inline-every-one on three intelmine keys ud-ewt-train / ud-ewt-test / ud-ewt-license -- another seat's rows in flight, named here, not touched. Box: two consecutive nx_dstate samples show the same six nx_sov_build_run pids in D on md2_raid5 for 416 to 553 s, two of them this lane's (the nx_organ_ship compile-proof and the CE6 gate stage); they are wedged in the RAID storm, not dead, and are left to complete rather than re-issued
2026-09-06	CE7	retract	The rung's premise (the grader PROMOTED-UNREGISTERED) is STALE by measurement 2026-09-06: nx_catalog reads nx_autofix_auto (46,116 B source, 246,691 B promoted) REGISTERED and AUTHORISED, and nx_swebv_grade (52,655 B) REGISTERED and AUTHORISED, both REGISTERED-DARK (no MCP invocation on record). The registration half of CE7 is therefore already done; what remains is the one real episode (localise from the grader rows, propose, verify by the byte-exact judge, revert proven exact) with its RED-before and GREEN-after, and that episode needs a fresh compile-and-run as its sole judge, i.e. the build lane
2026-09-06	CE0	measure	THE STORM HAS A NAMED PRODUCER: nx_raidwatch 2026-09-06 reads md3 raid1 DEGRADED (1 of 2 members missing, NO REDUNDANCY, a second failure is data loss), md2 raid5 with two damaged members (dev-sata6p5 808 corrected reads, dev-sata8p5 88) and kernel io_error=5 unc=5 on 2 distinct sectors, span 1824 (a growing defect), verdict RED. Every admission flap and every nx_sov_build_run in D-state today sits on that array. This lane cannot fix hardware; it is surfaced to the operator and paced accordingly
2026-09-06	CE8	queue	The census's second tree is the LAPTOP tree staged on the NAS (usage nx_forkgrade <listfile> <dirA> <dirB>, dirB e.g. /tmp/lapfork; the organ's own header says moving bytes between the two trees is irreducibly laptop-side), so the production spine cannot be fed by a NAS clock row: its first two points belong to the tree-sync lane (nx_treediverge_beat already stages and compares the trees; one fg_trend census per sync is the adoption step). Not a build -- a declaration on that lane's board is owed
2026-09-06	CE6	land	nx_number_envelope_gate staged through /api/build at 34,312 B sha bf93a4e3 (src_path names it, src_stable=1) and its promote with expect_sha256 issued 2026-09-06; nx_organ_ship (CE2) compile-proof GREEN on the sovereign lane (122,520 B debug flavour) and its /api/build stage issued the same minute
2026-09-06	CE6	land	CE6 SHIPPED AND PROVEN ON THE SERVED ARTIFACT 2026-09-06: nx_number_envelope_gate PROMOTED 34,312 B sha bf93a4e3 (prev banked), REGISTERED (callable on the allowlist hot-read; toolreg reconcile issued for tools/list discovery), _offc installed from the promoted root, and the SERVED binary run through nx_job_run reads 12/12 GREEN with the bare-percentage bite proven on the live copy. Remaining for CE6: migrate this board's own emitted figures (gauge, coverage, adoption) onto ce_number_envelope so the emitter refuses a bare percentage in production, referee grade, operator ACCEPT
2026-09-06	CE9	queue	SCOPED 2026-09-06, not started (build lane wedged on the RAID): sm_gauge_alarm has two halves and the emitter half is the load-bearing one. (1) In the laptop-local meter runtime/_hdl_build/nx_seat_meter_beat.nx (56,340 B, built by the sanctioned wsl launcher, not the NAS lane): stamp the gauge on every beat (knowledge/status/seat_gauge.stamp with ts= and a scale field such as rows= so a fixture cannot forge it), and sm_gauge_alarm(stamp, max_age_s, now) classifying FRESH / STALE / ABSENT with the bound DERIVED from the beat cadence (one full missed beat of margin, the memvel precedent), truncate-writing knowledge/status/seat_gauge_alarm.status whose last line is verdict=; an abstaining axis in the gauge raises AXIS-BLIND by name. (2) In nx_swcompare_matrix (grow the emitter, never a side page): the GAUGE row reads that status and renders STALE or ABSENT in place of the number, never zero. (3) nx_seat_gauge_alarm_gate: a forced-stale fixture (stamp older than the bound) raises the alarm, a fresh one is silent, an absent stamp reads ABSENT, and the emitter run over a fixture tree prints STALE for the row. Owed after CE2 and CE8 ship
2026-09-06	CE2	land	CE2 SHIPPED AND PROVEN ON THE SERVED ARTIFACT 2026-09-06: nx_organ_ship compile-proof GREEN on the sovereign lane (122,520 B sha 19e6897c, debug flavour following the incumbent); /api/build REFUSED-and-queued twice on the D-state witness, so the stage went through nx_filecopy (a cap minted async through the mgmt API) into the promote slot; nx_contentdiff live-vs-staged GREEN lost_from_live=0 ONE-WAY-GAIN (86 runs gained: the firsttry literals and the REFUSED-ADMIT class); PROMOTED with expect_sha256, prev banked, _offc installed. nx_build_firsttry_gate PROMOTED 3fd7ff39 (55,012 B), REGISTERED, served copy 33/33 GREEN. nx_forkgrade_gate PROMOTED 4bae5923 (43,824 B), REGISTERED, served copy 12/12 GREEN against the served nx_forkgrade.elf -- CE8's gate is shipped too. FIRST PRODUCTION BEAT of the rule (nx_job_run nx_organ_ship firsttry; spine knowledge/status/build_firsttry.spine row ts=1788675542): rows=2572 rows_in_window=2091 targets=208 episodes=253 first_pass=253 code_fail=0 box_refused=0 retries=119 shipped=242 rework=96 rework_permil=396 reverts=0 revert_unobservable=96 verdict=MEASURED
2026-09-06	CE2	retract	The first beat's first_try_permil=1000 is a BARE PERCENTAGE, not a measurement: 119 retries with ZERO recorded failures means the journal never carried the negative class before today's ship-loop edit (it journaled BUILT rows only). The ruler now abstains by name -- retries with no failure class reads UNOBSERVABLE with neg_class_absent=1 -- source landed in nx_build_firsttry_lib, rebuild and re-promote owed. The rework rate (96 of 242 ships re-shipped inside 14 days, 396 permil) IS a measurement; reverts become observable as bytes= rows accumulate. Second beat owed by a daily clock row so two beats reconcile per the accept rule
2026-09-06	CE2	land	Second-beat plumbing and the abstention landed 2026-09-06: clock row firsttrybeat 86400 nx_organ_ship.elf firsttry declared (CLOCKJOB-PUT ok rows=174, desired state, reconciled at the clock's next window, storm-deferred like every clock row); nx_build_firsttry_lib grew BFT_R_NEG_CLASS_ABSENT -- a window with retries and not one recorded failure sets first_try UNOBSERVABLE and neg_class_absent=1, printed on the report line and carried on the spine row (16,617 to 17,444 B, four CAS edits, one 503 that had landed caught by the guard); nx_build_firsttry_gate grew fixture F6 (a retry with no failure row, the pre-fix journal shape) with five teeth and a bite over the bare-1000 refusal, F4 as the present-class control (12,863 to 13,238 B plus the teeth). Gate rebuilt on the sovereign lane; ship-loop rebuild in flight; stage, promote and a served re-run of firsttry follow so spine row 2 reads the abstention honestly
2026-09-06	CE2	land	The abstention is SHIPPED on the gate side 2026-09-06: nx_build_firsttry_gate rebuilt 64,768 B sha 637a0146 (debug flavour following the incumbent), staged via nx_filecopy, PROMOTED with expect_sha256, _offc installed, and the SERVED copy reads 39/39 GREEN -- the six F6 teeth pass (a retry with no failure row: one retry counted, first_try UNOBSERVABLE never a bare 1000, neg_class_absent=1, rework still MEASURED, F4 present-class control 0, bite over the bare-1000 refusal fires-on-bad silent-on-good) and its production-spine neg-control now reads the real 175-byte knowledge/status/build_firsttry.spine unchanged before and after the fixture runs. Ship loop rebuilt 123,056 B sha f24b730e; nx_contentdiff live-vs-staged reads RED at 2 permil with exactly ONE lost run named -- the old parenthetical (box refusals excluded: the box, not the code), superseded by the longer sentence the candidate GAINS beside the neg_class_absent= fields -- a reworded message, not a capability, so the promote proceeds under the calibrated guard with that run named here
2026-09-06	CE2	measure	SECOND PRODUCTION BEAT on the served ship loop (PROMOTED 123,056 B sha f24b730e under the calibrated guard, prev banked, _offc installed): spine row 1788676308 reads episodes=253 first_pass=253 code_fail=0 box_refused=0 retries=119 first_try_permil=-1 first_try_state=UNOBSERVABLE neg_class_absent=1 shipped=242 rework=96 rework_permil=396 reverts=0 revert_unobservable=96 verdict=MEASURED. Two beats on knowledge/status/build_firsttry.spine reconcile on every count against the same 2,091 in-window rows of organ_ship.jrnl; the accept rule's mechanical half (two beats reconciling, a single-sample window UNOBSERVABLE, a known revert raising rework by exactly one) is met on the production journal and the gate. The daily clock row carries the beat from here. Still owed for CE2: the first-try rate becomes a measurement only after refusal rows accumulate in the journal (the negative class now journaled), referee grade, operator ACCEPT
2026-09-06	CE7	queue	SCOPED 2026-09-06 after reading the loop's own header: nx_autofix_auto <model> [manifest] [ledger-override] discovers the failing function from a grader's per-function FNRES rows, hands the extracted line to a LOCAL 1.5B i8 maker (greedy, then one reframed retry, then AF_BON_N seeded samples), and lets a fresh compile-and-run grader be the only judge at every rung; it was proven on seeded candidates (candidate2 -> sgn, candidate3 -> dbl) in the testing lane. The one REAL episode the accept rule demands therefore needs three things this seat does not hold on the NAS today: (1) the local maker -- the dispatch ladder measured no local engine on the NAS, so the episode is a laptop-engine lane (the sanctioned wsl launcher) or waits for an engine on the box; (2) a real debt-board defect expressed as a failing FNRES row, i.e. a subject whose gate emits per-function results, chosen from the open board and named before the run; (3) the null controls beside it (empty patch, revert patch, echo) each scoring zero, plus the RED-before receipt banked. Registration is already done (both organs REGISTERED). Not started here: an autonomous fix loop on a degraded array with no engine would be a workaround, not the rung
2026-09-06	CE3	land	DESIGN LANDED AS SOURCE 2026-09-06, composed into the incumbent: nx_capsearch found no recurrence ruler but nx_debtmine (REGISTERED bugs-class miner: per-scope open/eaten, eat-rate, aging) already owns the class axis, so ce_recurrence extends it instead of a second ruler and the matrix organ path is repointed from runtime/nx_failclass.nx to runtime/_hdl_build/nx_debtmine.nx by declaration. nx_debt_recurrence_lib.nx (new): per scope the FIRST EATEN filing is the prior; every row filed after it is a recurrence (recur_total = history, non-decreasing by construction; recur_open = the live re-filings, which fall when eaten); the prior id and a bounded prefix of the recurring ids are NAMED beside their total; close time is UNRECORDED on this plane (ids are file-time) so the document says bound=UPPER; the spine (knowledge/status/debt_recurrence.spine, or a <source>.recurrence.spine sidecar for any non-live source) is written only by the recurrence verb and only when rows were measured, announced inside the JSON. nx_debt_recurrence_gate.nx (new, in-process): three planted planes -- F1 alpha eaten-then-open, beta open-open, gamma eaten-eaten, one malformed row; F2 = F1 plus a third alpha filing (non-decreasing across two beats); F3 = F1 with the re-filing eaten (the resolved class falls) -- ids named in the JSON, beta never counts, spine sidecar two rows, production spine unchanged, one bite. nx_debtmine edits: the import landed; the recurrence object inside mine and the recurrence verb (ce_recurrence) are the next two CAS edits, then build, gate, promote, and the first two beats on the live plane
2026-09-06	CE3	land	ce_recurrence SHIPPED: nx_debt_recurrence_lib is the ONE ruler (per scope the first EATEN filing is the prior and every later filing a recurrence; recur_total is non-decreasing history, recur_open is live and falls when eaten; ids named; close_time UNRECORDED so the bound is declared UPPER), composed into the incumbent nx_debtmine as a recurrence verb plus a recurrence object inside mine (import 23,590 to verb 27,139 B, four CAS edits, every predicted size exact). Gate nx_debt_recurrence_gate 24/24 GREEN and bite-proven (fires on the refiled class, silent on the never-resolved one), values emitted for independent adjudication; PROMOTED 56,209 B c8a8de62, REGISTERED callable=1, _offc refreshed. Miner PROMOTED 80,004 B 76f7295f after nx_contentdiff GREEN lost_from_live=0 gained=104 (strict superset). Spine knowledge/status/debt_recurrence.spine is written only by the recurrence verb on the production prefix and only when rows>0; the first two live beats follow this row.
2026-09-06	CE3	retract	beat 1 (spine row 1788692120: scopes=256 with_recurrence=68 recur_open=401 recur_total=630) was a FLOOR, not a total: the scope table was a 256-slot bound and it bound on the 4,353-row plane. The document printed scopes_capped=1 and the seat read past it one line later. Superseded by beat 2 below; the spine now carries capped= on every row so the retraction is machine-readable.
2026-09-06	CE3	measure	beat 2 on the promoted miner with the row-derived capacity (nx_debtmine 91,432 B 501df732, lib 15,123 B): rows=4353 malformed=66 scopes=2305 scopes_capped=0 scopes_with_recurrence=251 recur_open=889 recur_total=1528 id_bursts_adjacent=133 coverage_complete=1. The cap had hidden 889 permil of the scopes and 588 permil of the recurrence. Top recurring classes by open re-filings: stale-mirror 98 open of 117 (prior 1784654222), ivv-drift 41 of 41, gx-body 27 of 30, discoverability 25 of 26, search 23 of 29, atlas 22 of 37, control-plane 22 of 24. The ids repeating inside stale-mirror are an autofile burst (three rows 313 bytes apart sharing one epoch-second), distinct filings counted as such and declared by the burst field. close_time is UNRECORDED on the plane so every figure is an UPPER bound.
2026-09-06	CE3	lesson	A CAP THE DOCUMENT ANNOUNCES IS STILL READ AS A TOTAL BY THE NEXT LINE. The remedy was not a bigger number: capacity is derived from the rows (distinct scopes cannot exceed rows) so the bound cannot exist, and the gate proves it with an F4 plane of 300 scopes the old table would have cut (gate 33/33, capacity 603 over 602 rows). The same 256-slot table lived in the incumbent mine census (DM_SCOPES) and was fixed in the same edit: a fix that lives in one ruler and not its sibling is half a fix. Also caught: the recurrence object composed into the mine document made a whole-document absence tooth (T5) fail on the promoted binary; the tooth now names the census shape it tests.
2026-09-06	CE3	land	CE3 SHIPPED ON THE LIVE PLANE: nx_debtmine 91,496 B 393f80e8 PROMOTED (recurrence verb, recurrence object inside mine, row-derived scope capacity for BOTH rulers, T5 tooth scoped to the mine census) with _offc refreshed and selftest 10/10 GREEN on the promoted binary; nx_debt_recurrence_lib 15,123 B (drc_capacity, cap parameter, adjacent id-burst counter, capped= on every spine row); nx_debt_recurrence_gate 33/33 GREEN bite-proven (F4 300-scope plane counted in full), PROMOTED 68,888 B da4768c5, REGISTERED, cap minted 30 d, _offc refreshed. The live mine census now reads scopes=2305 scopes_capped=0 scopes_below_minsup=606 and names codewiki-debt 28 open, render-docs 25, atlas-consolidation 14, wiring-guard 13 -- scopes the capped table never interned. Spine knowledge/status/debt_recurrence.spine: row 1788692120 (capped, retracted) then beat 2 (capped=0); beat 3 follows so two comparable rows exist. Clock row debtrecurbeat 86400 declared. Shipped and gate-proven is NOT complete: referee grade and operator ACCEPT are owed before this rung reads done.
2026-09-06	CE3	measure	nx_compare_rank after the CE3 ship: rungs=10 done=8 open=2 (CE7 ce_fixloop_live priority 300 critical path; CE9 sm_gauge_alarm priority 100), unmapped=0, finish=0, rank artefact 883 B, plane comparerank-codeeffectiveness published. Clock row debtrecurbeat landed (CLOCKJOB-PUT rows=175, the verb and its args ride in the organ field). The recurrence gate reads REGISTERED-DARK on nx_catalog until the MCP surface discovers it on its next sweep (it was registered today; nx_job_run forks log the runner, not the organ).
2026-09-06	CE9	land	CE9 gauge ruler + gate SHIPPED NAS-native: nx_gauge_lib (7,661 B, THE ONE gauge-heartbeat ruler -- ts-first dead-man stamp, ABSENT>STALE>BLIND>FRESH classifier, future stamp and unpinned cadence both STALE, value shown ONLY when FRESH, max_age derived as two beats). nx_gauge_gate 14/14 GREEN bite-proven built+run on the NAS (withhold-value-fires-on-STALE-silent-on-FRESH BITE-PROVEN), PROMOTED 39,465 B 92648727, REGISTERED callable=1, _offc refreshed, cap minted 30 d. gauge_pass wired into the compare emitter nx_swcompare_lib (both generators, api.json + page); a <dom>.gauge row is re-judged every publish by the same ruler and the value key exists only on a FRESH row, so a missing key can never read as zero. Generator promote (to render live) deferred under the I/O storm; a codeeffectiveness.gauge data row waits for its stamp writer.
2026-09-06	CE9	retract	the beat-writer half was NOT shipped and its first cut was WRONG-LANED: sm_gauge_alarm was added to the NAS buildroot copy of nx_seat_meter_beat and /api/build failed on mp_hook_json undefined at two PRE-EXISTING call sites. nx_absent proved mp_hook_json absent from BOTH NAS trees (corpus_complete=1) -- nx_seat_meter_beat is a LAPTOP-LOCAL SessionStart organ, so /api/build was the wrong surface, not a divergence defect. All three NAS-copy edits reverted to the exact 56,340 B original. The stamp writer (tokens/rung is laptop-only) is a laptop-tree step: copy nx_gauge_lib into the laptop tree, wire sm_gauge_alarm, build in WSL. Until then the gauge reads ABSENT/STALE, which is the honest state the rung's accept rule wants.
2026-09-06	CE9	lesson	A COMPILE ERROR AT A PRE-EXISTING CALL SITE AFTER AN ADDITIVE EDIT MEANS THE FILE WAS ALREADY UNBUILDABLE ON THAT SURFACE -- check which tree owns the organ before choosing a build lane. And the accept rule is met by the gate, not the flip: a stale gauge that reads STALE not zero is proven by nx_gauge_gate, and the /compare cell reading ABSENT until a writer exists is honest, not incomplete-in-the-bad-direction.
2026-09-06	CE9	land	CE9 WRITER LANDED ON ITS NATIVE SURFACE (laptop), storm-independent. nx_seat_meter_beat is a laptop-local SessionStart organ (mp_hook_json lives in the LAPTOP runtime/_hdl_build/nx_memplane_lib.nx; nx_absent proved it absent from both NAS trees, so the NAS build farm is the wrong surface and its storm does not block this). The laptop source carries the gauge wiring (import nx_gauge_lib + sm_gauge_alarm + the call after the METERBEAT summary); built via the sanctioned wsl sovereign-ELF launcher (nx_sov_build_run nx_seat_meter_beat --build-only then --install), 82,970 B sha cc9a865a, installed to _offc/nx_seat_meter_beat.elf. PROVEN by a scan run: GAUGE-ALARM state=BLIND prev_age_s=-1 axes_known=1 axes_total=2 wrote=0 -- scan correctly sees one axis abstain (the remote meter log it does not read) and reports BLIND rather than acquitting, and writes nothing; a real SessionStart beat reads both axes and writes knowledge/status/seat_gauge.stamp. So the three CE9 layers now exist: ruler (nx_gauge_lib, NAS+laptop), gate (nx_gauge_gate 14/14 GREEN, NAS), writer (nx_seat_meter_beat --install, laptop live), emitter render (gauge_pass in nx_swcompare_lib, written).
2026-09-06	CE9	queue	two integration steps remain, both cleanly named: (1) generator rebuild+promote so gauge_pass renders live -- storm-blocked (procs_blocked 11 of 8), and inert until a .gauge data row exists; (2) the cross-surface stamp delivery -- the writer computes laptop-side (transcripts are laptop-only) while /compare renders NAS-side, so the gauge stamp must ride the seat meter's existing estate push to a NAS-readable path that a codeeffectiveness.gauge row points at. Neither is a hack; both ride mechanisms that already exist. Three-party ACCEPT (referee + operator) still owed for CE3 and CE9.
2026-09-06	CE9	land	GENERATOR PROMOTE DONE -- gauge_pass is LIVE in the compare emitter fleet-wide (the storm eased and /api/build granted). nx_swcompare_matrix 412,656 B 97b52ddb and nx_swcompare_sota 261,712 B c39dfdf8: both nx_contentdiff GREEN strict-superset (lost_from_live=0, gained the gauge strings gauge=/axes_known=/axes_total=/cadence_s=/age_s= and knowledge/status/seat_gauge.stamp), src_path confirms the target, src_stable=1. Promoted (prev banked), nishihost/_offc installed, AND buildroot/_offc refreshed via nx_restage so all THREE roots the regen forks agree (matrix 412,656 all three; sota 261,712 all three) -- the FORK-STALE class cannot fire. gauge_pass is additive (absent <dom>.gauge file => no section, returns 0), so it renders nothing on any page until a .gauge row exists, i.e. zero risk to the 109 live pages and confirmed by the strict-superset diff. CE9 layers now ALL live: ruler (nx_gauge_lib) + gate (nx_gauge_gate 14/14) + writer (nx_seat_meter_beat --install, laptop) + emitter render (gauge_pass, promoted). ONE step remains: cross-surface stamp delivery -- the laptop writer's stamp must reach a NAS-readable path a codeeffectiveness.gauge row points at (rides the seat meter's existing estate push; the clean form is a NAS-side stamp derived from the already-pushed meter data). Not a hack, a named follow-up. Three-party ACCEPT still owed.
2026-09-06	CE9	queue	CE9 LAST MILE PINNED AS A BUILD ORDER (the clean form, not a hack). The cross-surface stamp: nx_seat's `shift` verb (st_shift, buildroot/runtime/_hdl_build/nx_seat.nx ~204 KB) ALREADY runs NAS-side, computes tokens-per-rung, and writes shift.log + shift.jrnl -- so the gauge stamp is a bounded addition THERE: import nx_gauge_lib and call ga_stamp_write(GA_STAMP_PROD, now, gauge, axes_known, axes_total, cadence) at the end of st_shift, then a codeeffectiveness.gauge row points at knowledge/status/seat_gauge.stamp and gauge_pass renders it FRESH. NOT done this session, deliberately: it edits a 204 KB core organ, needs its axes-contract defined (nx_seat reads the pushed meter data, so axes_known/total means the coverage of that push, a different semantics from the laptop beat's local-scan+remote-log), and the box is under an intermittent RAID storm -- rushing a core-organ edit there is the workaround the standing orders forbid. The laptop writer already stamps locally (proven); the NAS-side stamp is the delivery. Build it on a calm box with the axes contract written first.
2026-09-06	CE9	queue	CROSS-BOARD THREAD: the operator's 2026-09-06 one-step-solve goal is declared as engineshift ES26 (al_steps on the incumbent nx_actlog) with its architecture twin ecosystem EC37 (the common-task registry's consolidation-debt ratchet) and a verbatim lesson row on EC18. Its gauge stamp knowledge/status/stepsolve.stamp will render through THIS board's gauge_pass -- the CE9 mechanism -- as a second gauge row: one render, many gauges. The operator also agreed (same day) that CE9's cross-surface stamp is the NAS-side form derived from the pushed meter data, a small deliberate build.
2026-09-06	CE4	queue	PUBLISH STORM-DEFERRED BY THE GOVERNOR, not by choice: two manual regens were refused -- first REFUSED-ALREADY-RUNNING (the compare beat held lease compare-regen, holder regen-1788654204, which started BEFORE the promote+restage and so forked the pre-fold _offc), then REFUSED-IO-STORM procs_blocked=7 ncpu=8 (the D-state witness at the storm line the build gate refuses on; the clock beat defers itself via clock_heavy.conf the same way). Every step ON the publish path is done and hash-verified (all three roots f7eaa106; CE4 matrix row repointed to cb_classify), so the UNSERVED fold and the LANDED watch publish on the first ADMITTED beat regen -- no seat action owed beyond re-issuing nishi_compare_regen once nx_build_admit reads GRANT. Do not launch a third regen into the storm.
2026-09-06	CE4	land	CE4 PUBLISH PATH COMPLETE ACROSS ALL THREE ROOTS. nx_swcompare_matrix promoted f7eaa106 (serving root); nx_restage copied the PROMOTED binary into buildroot/_offc (the regen's fork target); all three roots now hash f7eaa106 396848, verified by nx_filehash (the pre-fold 4eeb7688 392952 is gone). So the compare regen publishes the UNSERVED fold, not the old generator. CE4 matrix row repointed runtime/nx_catalog.nx -> runtime/nx_symdecl_lib.nx
2026-09-06	CE4	land	CE4 ADOPTION PROVEN END-TO-END, nx_sota_drift_gate 41/41 GREEN. Re-pointed gate PROMOTED (57952 B sha 8ca7be4e; the 6 CAPLOSS runs were all superseded -- old nx_swcompare_sota subject paths the gate no longer drives + the verdict note I extended, lost_srcpath=0, +127 runs gained, artifact GREW 42455->57952, so a forward change not a revert -- allow_capability_loss=yes named on that) and driven via nx_job_run against the freshly-built buildroot/_build/nx_swcompare_matrix.sov.elf (the fold, still unpromoted) on the /tmp fixture tree, ZERO live pages touched. NEW F-LEG ALL PASS: a shipped claim over a source-only organ reads UNSERVED (counted, named, html badge >UNSERVED</b>, json drift field), symbol-presence alone no longer blesses a shipped claim, a served organ stays CONSISTENT (C-again unserved=0, byte-identical), and unserved-fires-on-source-only-silent-on-served is BITE-PROVEN. This is the CE4 mechanism working on the surface that has the drift vocabulary. REMAINING (mechanical, proven-safe): promote nx_swcompare_matrix + nx_restage buildroot/_offc + nx_filehash all 3 roots; repoint CE4 matrix row runtime/nx_catalog.nx -> runtime/nx_symdecl_lib.nx
2026-09-06	CE4	land	CE4 EMITTER ADOPTION WRITTEN AND COMPILE-PROVEN across three files via the sovereign lane (source-landed, both _build artifacts compile clean; headline promote deferred -- build-admit read load1_centi=1634 against a 1643 ceiling, 16 running, and the matrix generator is the ONE assembler for ~106 /compare pages: promoting it into a near-ceiling storm is the hammer-the-NAS anti-pattern, not sota shipping). WHAT LANDED: (1) nx_symdecl_lib cb_serve_state_cwd -- the served-artifact reader, CWD-relative the same way sd_present resolves source (probes ../<base>.elf then ./<base>.elf, the two CWDs the generators run under), so the two rulers never judge different subjects for one path. (2) nx_swcompare_sota_lib: at the bound m==1/claims==1 branch it now calls cb_serve_state_cwd and a shipped claim whose organ is SOURCE-ONLY reads a NEW class SL_DRIFT_UNSERVED (its own name/class/badge/reason 'built nowhere -- over-reports what RUNS', counter b_unserved, census+html+json), additive: a served organ stays CONSISTENT so lang's 20 bindings are byte-identical. (3) nx_sota_drift_gate: kind-6 UNSERVED fixture (source planted, served twin planted for the consistent case, unserved twin unlinked), F-leg asserting unserved fires on source-only and is silent on the served consistent row, partition includes unserved, plus a `<subject_elf>` argv override (the nx_gate_bite subject precedent) so a staged build is provable before promote. Built shas: gate be2326a7, matrix d967f84e. TURNKEY SHIP, next low-load window: (a) /api/build+promote nx_sota_drift_gate, verify staged carries 'F(unserved)'; (b) nx_job_run nx_sota_drift_gate /tmp/o buildroot/_build/nx_swcompare_matrix.sov.elf -- proves UNSERVED end-to-end WITHOUT touching live pages; (c) if GREEN, /api/build+promote nx_swcompare_matrix + nx_restage buildroot/_offc + nx_filehash all 3 roots; (d) repoint CE4 matrix row runtime/nx_catalog.nx -> runtime/nx_symdecl_lib.nx
2026-09-05	CE4	queue	EMITTER ADOPTION MAPPED FROM SOURCE, turnkey (deferred under a MEASURED sustained storm: 4 consecutive edge-window 503s on READ calls 2026-09-05, so the multi-file generator edit is NOT made from guessed constants). (1) The drift classifier is INLINE and DUPLICATED per generator; the STALE-UNDER/UNBACKED/CONSISTENT vocabulary exists ONLY in nx_swcompare_sota (inline at the m==1 && claims==1 branch, ~line 285), while nx_swcompare_matrix flips WATCHING<->landed on bare symbol presence with NO drift classes. (2) codeeffectiveness is a .matrix, so CE4 on ITS OWN board needs the matrix generator taught the drift vocabulary FIRST (larger port); the CE4 MECHANISM lands cleanly on the sota generator today. PLAN, additive: add BD_DRIFT_UNSERVED to nx_swcompare_lib beside BD_DRIFT_STALE_UNDER (own bd_drift_name + bd_drift_class; matrix gen unaffected); in the sota classifier at m==1 && claims==1 call cb_serve_state(organ) (built+proven, nx_claim_binding_gate 12/12) and downgrade CONSISTENT->UNSERVED when SOURCE-ONLY, with a DISTINCT reason 'claimed shipped, built nowhere' -- NEVER reuse STALE_UNDER's 'graded No' text (names the wrong subject). Served organs stay CONSISTENT so lang's 20 bindings are byte-identical; add b_unserved counter + json key. VERIFY: extend nx_sota_drift_gate with a source-only-organ fixture asserting UNSERVED; ship sota via sovereign build + nx_restage + nx_filehash all 3 roots. THEN matrix-generator port, then repoint CE4 matrix row to runtime/nx_symdecl_lib.nx
2026-09-05	CE4	land	CE4 RULER BUILT AND GATE-PROVEN via the sovereign lane (governor-free, storm-immune -- admission GRANTed load1_centi=1287). Added the served-artifact classifier to nx_symdecl_lib.nx beside sd_present: cb_serve_state(organ) reads SERVED/SOURCE-ONLY/ABSENT by stating the serving-tree binary and the source, and cb_classify(claim_present, state) returns CONSISTENT / STALE-UNDER (verdict contradicts, publish WITH badge) / REFUSE (verdict denies, do not publish) -- the exact ACCEPT wording. This is the ruler CE5 named: SYMBOL PRESENCE CANNOT SEE A PROMOTE, and sd_present passed the McCabe SHIPPED-over-SOURCE-ONLY claim because mc_scan_file was in source the whole time. nx_claim_binding_gate GREEN 12/12, 3 bites BITE-PROVEN incl the McCabe end-to-end shape, built sha d9f4be7a in _build only. NOT DONE, honest: CE4 ACCEPT requires the EMITTER to re-read claims ON PUBLISH. NEXT = wire the nx_swcompare_sota drift loop (its sd_present(organ,sym,rl2) call site near off 17788) to also consult cb_serve_state for =<organ> bindings and render STALE-UNDER/REFUSE, then repoint the CE4 matrix row from runtime/nx_catalog.nx to runtime/nx_symdecl_lib.nx
2026-09-05	CE0	lesson	PROMOTE-PATH RULER VERIFIED LIVE-FIXED, one fewer open blocker than the record implied. Debt 1788635942 (nx_behaveprobe reported rc=0 evidence=STRONG for a subject the crash guard killed with SIGSEGV, clearing a crashing gate for promote) is correctly eaten: the live binary nishihost/nx_behaveprobe.elf 27879 B CONTAINS the emitted grade string SIGNALLED: an rc above 128 is 128+signal at off=25286, so the OWNER fix shipped, not only the DanceXR gate cursor fix that lane journaled. This board CE1/CE5 proof runs through that ruler; it now grades a crash as SIGNALLED and HARNESS, never STRONG. Verified by txtscan of the live elf, no build needed.
2026-09-05	CE1	measure	FIRST TRAJECTORY, two spine rows 101 minutes apart (1788643308 -> 1788649356): files 19505 -> 19511 (+6), funcs 112660 -> 112818 (+158), above_10 14293 -> 14309 (+16), permille 126 -> 126. Every delta sits in UNMARKED (17319 -> 17325 files, 107244 -> 107402 funcs, 14112 -> 14128 above); BUILDER, GENERATED and DERIVED are byte-for-byte unchanged. Sixteen new over-bar functions in an hour and a half of seat work is the number this board exists to watch, and this seat's own ce_author_health is a candidate among them; the next census names them by file through the worklist
2026-09-30	CE12	queue	2026-09-30 operator: true human-estate partnership and eventual first/third-party LLM comparison. First baseline token/resource costs, then accepted developer outcomes at task/workflow/project levels. This record adds acceptance criteria, not measured results.

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. Honest position, measured 2026-09-05. The estate has more effectiveness instrumentation than any field platform and its own headline gauge is RED. The engine-shift ratchet (nx_seat shift) reads Claude output-per-closed-rung at 5,892,596 tokens against 1,030,739 the prior 30-day window -- a 5.7x rise -- and it is RED by design because a rising cost of an outcome is a regression whatever the throughput. That is the number this board exists to drive down, and the field's own reality check is why we measure rather than assume: a randomized controlled trial found experienced developers 19 percent slower with AI while believing they were 20 percent faster [metr]. Where we are genuinely ahead is proof-carrying delivery: every promote runs a capability-loss diff that names each lost run [fitnessfn], the drift census rebuilds source and compares served bytes, and a mutation harness asks whether an oracle can fail at all [pitest] -- none of DX, Faros, GitClear or CodeScene refuses a deploy on a measured property. Where we are behind is the human-versus-AI authoring lens the operator named: GitClear measured across 623 million AI-era changes that block duplication rose 81 percent and refactored lines fell from 21 to 3.8 percent [gitclear26], and CodeScene's quantitative study found unhealthy code carries far more defects and costs materially more to change [codered22], but the estate has the authoring record and the health rulers and joins them nowhere -- a single-sample corpus that cannot see its own AI-authorship trajectory. And the discipline that makes a green mean something is thin: of 3,477 gate sources only 452 satisfy all three gate laws and 2,759 are hand-rolled and unmeasured. The board's own first finding is published rather than hidden: /compare/codequality asserts the cyclomatic-complexity keystone SHIPPED with its gate GREEN, while nx_catalog on the serving tree reads its binary SOURCE-ONLY -- a claim unbacked on the surface that runs, and the reason rung CE4 binds every effectiveness claim to a live verdict.

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

Every compare carries this layer. Declare knowledge/compare/codeeffectiveness.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 codeeffectiveness, 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).

Gauges — a heartbeat, never a bare number

Each row names a gauge stamp written by its measuring beat. This page re-judges the stamp on every publish with the same ruler the beat uses (nx_gauge_lib): FRESH shows the value; STALE withholds it (older than two beats); BLIND withholds it (an axis abstained); ABSENT means nothing has measured. A stale gauge never reads as zero.

gauge	state	reading	note
One-step solve: explained permil of common-task support	STALE	STALE age_s=1315528 max_age_s=172800 (the last reading is history, not a level: the gauge value is withheld)	Rendered here through CE9 gauge_pass (one render, many gauges); producer is engineshift ES26 nx_actlog steps over knowledge/status/actlog.jrnl (daily beat stepsolvebeat). gauge = declared classes over ALL recurring-signature support; axes = journal read + conf read; UNDECLARED is the worklist, never folded into one-step
Tokens per closed rung: claude output tokens per DONE workstream in the window	STALE	STALE age_s=1361595 max_age_s=1209600 (the last reading is history, not a level: the gauge value is withheld)	CE9 last mile: written NAS-side by nx_seat shift from the meter log and the ws journal the beat already pushes (axes = both files read; a window that closed nothing writes no stamp, so the reading ages to STALE, the axis-blind alarm). ES18 ratchet: this figure may not rise window over window; a token win with a referee-grade drop is a regression

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 21 capabilities: 10 exclusive to it and 11 shared with other boards. Shared rows on this page: 11 = home here 0 + home elsewhere 3 + ambiguous 1 + undecided 7 + refused 0 (partition 11); undecided = unmapped 0 + unproven 7 + unevidenced 0; homonyms 0; since the previous census: new 0, resolved 0. Census FRESH age_s=7160.

Capability	Organ	Boards grading it	Home	Rule	Evidence	Standing on this board
d_band_deploy	`nx_dora.nx`	swarm, devmgmt, codeeffectiveness	—	none	unproven	undecided
ch_frames	`nx_claude_harvest.nx`	devmgmt, codeeffectiveness	—	none	unproven	undecided
al_mine_json	`nx_actlog.nx`	devmgmt, codeeffectiveness	—	none	unproven	undecided
cd_show_lost	`nx_contentdiff.nx`	devmgmt, codequality, autodev, capregistry, codeeffectiveness	—	none	unproven	undecided
gb_try	`nx_gate_bite.nx`	devmgmt+, codequality, autodev*, codeeffectiveness	autodev	rung	plan	home elsewhere adopt: `@xrow autodev\|gb_try\|home by rung (fleet duplicate census)`
ps_judge	`nx_artifactdrift.nx`	devmgmt, capregistry, codeeffectiveness	—	none	unproven	undecided
st_boot	`nx_seat.nx`	devmgmt, engineshift, codeeffectiveness	—	none	unproven	undecided
gl_strip	`nx_gatelaw_gate.nx`	codequality, codeeffectiveness	—	none	unproven	undecided
fg_trend	`nx_forkgrade.nx`	codequality, codeeffectiveness	—	ambiguous	plan	ambiguous
fc_alert_score	`nx_failclass.nx`	failmodes*, autodev, codeeffectiveness	failmodes	rung	plan	home elsewhere adopt: `@xrow failmodes\|fc_alert_score\|home by rung (fleet duplicate census)`
ed_dispatch	`nx_engine_dispatch.nx`	engineshift*, codeeffectiveness, sovereignty	engineshift	rung	plan	home elsewhere adopt: `@xrow engineshift\|ed_dispatch\|home by rung (fleet duplicate census)`

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 1396265 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_mccabe_gate`	codeeffectiveness	plan-queue	rung=CE1
directive	`nx_headmark_lib`	codeeffectiveness	plan-queue	rung=CE1
directive	`nx_forkgrade_gate`	codeeffectiveness	plan-queue	rung=CE8
directive	`nx_container`	codeeffectiveness	plan-queue	rung=CE0
directive	`nx_container_run`	codeeffectiveness	plan-queue	rung=CE0
directive	`nx_container_verdict_is_valid`	codeeffectiveness	plan-queue	rung=CE0
directive	`nx_seat_gauge_alarm_gate`	codeeffectiveness	plan-queue	rung=CE9
directive	`nx_gauge_lib`	codeeffectiveness	plan-queue	rung=CE9
directive	`nx_claim_binding_gate`	codeeffectiveness	plan-queue	rung=CE4

rows shown 9|this board's directives 9|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.

[metr] Becker Klein Rush Barnes Rein and METR. Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity (2025). A randomized controlled trial in which 16 experienced open-source developers completing 246 real tasks in their own mature repositories were 19 percent SLOWER when allowed early-2025 AI tools, while both the developers and independent experts forecast a roughly 20 to 24 percent speedup -- the perception inverted the measurement. publisher · read in our library knowledge/fetched/rf_ea915a0a.raw · pin he4a5fe1b8b13fcd3aa4f6b768b96e0372f6644cae0d11722a3752f7f89cd0b0c · accessed 2026-09-05 · published-paperGrounds: GAUGE: why effectiveness is measured and not assumed
[gitclear26] GitClear. The Maintainability Gap - 2026 AI Code Quality Research. Over 623 million analyzed changes 2023 to 2026 show block duplication rising 81 percent to the highest level on record, refactored lines falling from 21 percent to 3.8 percent, and error-masking constructs up 47 percent. publisher · read in our library knowledge/fetched/rf_2da9d24b.raw · pin h5e28f3ac9e92e2d3d8b07f302c2696fd8312e8f5acb945933339c971141c12b5 · accessed 2026-09-05 · datasetGrounds: AUTHORING and BUILD and DUPLICATION: the AI-era duplication, refactor and churn trend
[faros26] Faros AI. AI Engineering Report 2026 -- are AI coding assistants really saving time, money and effort. Telemetry from about 22,000 developers across more than 4,000 teams: throughput is up, but AI adoption is associated with a 51.3 percent increase in average PR size and a 54 percent increase in bugs per developer (up from 9 percent in the prior dataset), and the probability of a production incident has more than tripled -- bugs, incidents and rework rising faster than the gains. publisher · read in our library knowledge/fetched/rf_cd93000d.raw · pin hd882676e2b45c50ab95b8c02ab8a20418936c4cb8a64d781cba2e43aa1f3b2da · accessed 2026-09-05 · datasetGrounds: AUTHORING and FEATURE-DEV and BUGRES: the field capture whose own numbers show AI throughput bought with bugs and review strain
[codered22] Tornhill Adam and Borg Markus. Code Red - The Business Impact of Code Quality - A Quantitative Study of 39 Proprietary Production Codebases. arXiv 2203.04374 2022. Code Health 1.0 to 10.0 over 30737 files; files in the unhealthiest band carry roughly 15 times more defects and cost about 124 percent more development time to change. publisher · read in our library knowledge/fetched/rf_593062bb.raw · pin h21559ccffd891c13641aadd33eeea1731172ec4d0de62e96731bde1435547262 · accessed 2026-09-05 · published-paperGrounds: AUTHORING: code health determines defect density and change cost
[dora-2025] DORA (Google Cloud research program), State of AI-assisted Software Development 2025: AI is an amplifier that magnifies existing organizational strengths and weaknesses and returns come from the underlying system; survey-based respondent self-report that structurally cannot see a development session. publisher · read in our library knowledge/fetched/rf_19135fe4.raw · pin h6bb53c243fc615849727e65b6a9aa33dbd6b0d5c4e63af5ce68137db0924b4d9 · accessed 2026-09-05 · published-paperGrounds: DEPLOY: the four keys and the AI-amplifier backdrop
[devlake-dora] Apache DevLake documentation, DORA metrics: computes DORA from exactly pull requests, deployments and incidents, with lead time measured commit-to-production, so everything before the first commit is invisible by construction. publisher · read in our library knowledge/fetched/rf_aeca14ce.raw · pin h621dd47e8d57767610d404157b81d24775d8d885e2b65351cbfa9c72c15152dd · accessed 2026-09-05 · vendor-docGrounds: DEPLOY: why a built-and-undeployed capability is invisible to the field clock
[dx-ai-framework] DX Research, Measuring AI code assistants and agents: three dimensions of utilization, impact and cost from AI-tool telemetry, git data and surveys, with the authors conceding agentic measurement must still evolve. publisher · read in our library knowledge/fetched/rf_c5f0df06.raw · pin hd1e598c967060c784dcf43b26b52f06fb874ee57fac130c09d2161617b2e3e42 · accessed 2026-09-05 · vendor-docGrounds: GAUGE: FinOps unit economics of the development loop
[harnessthesis] Zhang Wang Ge Xu Hamm and Reddy. Stop Comparing LLM Agents Without Disclosing the Harness (position paper, 2026). The Binding Constraint Thesis: harness configuration governs more performance variance than model choice, with measured model-ranking reversals, and it proposes a harness disclosure standard. arXiv 2605.23950. publisher · read in our library knowledge/fetched/cmp_autodev_harnessthesis.html · pin he6f975369dc5da48550dbf83894d8baad17eedf57200a9ee5481094df989aad3 · accessed 2026-08-27 · published-paperGrounds: RIGOR: why tooling configuration and its verdicts must be disclosed beside a number
[statprecipice] D'Oro et al. Computer Use at the Edge of the Statistical Precipice (2026). A 1 MB replay script that never observes the screen outperforms frontier models on static benchmarks, and the PRISM principles with Wilson score intervals and hierarchical bootstrap for the nested structure of agent benchmarks. arXiv 2605.08261. publisher · read in our library knowledge/fetched/cmp_autodev_statprecipice.html · pin hf1c4c6ed1d17b82a952ebcd3af187473c0b7afc840e861159499f1145e242e19 · accessed 2026-08-27 · published-paperGrounds: RIGOR: the null control and the interval every published number carries
[runornot] Lin et al. To Run or Not to Run (2026). 3,000 end-to-end attempts over 200 instances: prohibiting code execution narrows the resolve-rate gap by only 1.25 points while saving tokens and time, and the benefit concentrates on a subset of instances. arXiv 2606.26978. publisher · read in our library knowledge/fetched/cmp_autodev_runornot.html · pin hf05b212a46e782b0c126b8c9c40cafc28d95a124330956e48a6b2c51f7012ff4 · accessed 2026-08-27 · published-paperGrounds: FEATURE-DEV: execution budgeted per task class rather than defaulted
[swebench] Jimenez et al. SWE-bench Can Language Models Resolve Real-World GitHub Issues (arXiv 2310.06770). A patch is RESOLVED only when the FAIL_TO_PASS tests pass and the PASS_TO_PASS tests still pass, with the test suite as the sole judge. publisher · read in our library knowledge/fetched/cmp_autodev_swebench.html · pin hcd170cab3a56f891ee94070d5748d923685f2a84a91201d6336a221436fd0f9f · accessed 2026-08-27 · published-paperGrounds: BUGRES: the verify contract the sovereign fix loop adopts
[swebenchpro] Deng et al. SWE-Bench Pro (arXiv 2509.16941). 1,865 long-horizon problems from 41 actively maintained repositories, human-verified, patching multiple files and presented as a contamination-resistant testbed. publisher · read in our library knowledge/fetched/cmp_autodev_swebenchpro.html · pin h21f69a25e615c4223357fe653663aeb94f962732cbbb1277e827bd3014f9df53 · accessed 2026-08-27 · published-paperGrounds: BUGRES: the honest long-horizon multi-file bar
[fitnessfn] Thoughtworks Technology Radar. Architectural fitness function (Trial ring, November 2017). An architectural fitness function provides an objective integrity assessment of some architectural characteristic and may encompass existing verification criteria. publisher · read in our library knowledge/fetched/rf_397694c0.raw · pin hb0101000c0205262537de4c7b2ab35ae60f8c3a8c30202b6599819a0bbb7b986 · accessed 2026-09-05 · vendor-docGrounds: DEPLOY: the promote gate as a fitness function that refuses a deploy
[pitest] PIT Mutation Testing. Mutation score is the percentage of seeded faults killed by the suite, and line coverage does not check that tests can detect faults in the executed code. publisher · read in our library knowledge/fetched/rf_cad2084e.raw · pin hb0c7e0505c51c95b596060ef27babbce08e72d5aaf4557eb002c2637d2bf10cc · accessed 2026-09-05 · vendor-docGrounds: BUGRES: mutation score as the test-quality standard the bite harness mirrors
[metr-uplift-20260224] METR, February 2026 developer-productivity study update: participation and task selection plus concurrent agent work complicate causal estimates; the new signal is not treated as a reliable universal uplift estimate. publisher · read in our library knowledge/fetched/rf_7076254c.raw · pin h4fe0bc708ea08fb267555c61cf67e6aa0fcde6a098129b1619d5fefabd5a7d6b · accessed 2026-09-30 · research-reportGrounds: CE12: retain selection effects, task population, human active time and concurrency rather than treating tokens or elapsed time as productivity
[engthrive-2026] Microsoft Engineering Thrive: system-level assessment of speed, ease and quality alongside developer thriving in agentic workflows. publisher · read in our library knowledge/fetched/rf_f932ca9e.raw · pin hcecad473be201af48658fceadae2b7d6d9b03ac842feb6981a75f54b5d3ea7a3 · accessed 2026-09-30 · vendor-researchGrounds: CE12: compare accepted outcomes and human experience; do not rank individual contribution by activity counts

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