Nishi Family › Compare › Research Reach and Harvest

Nishi Compare · capabilities, evidence and delivery

Research Reach and Harvest

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

Nishi vs OpenAlex and CORE and Semantic Scholar and OpenAIRE -- can we reach, harvest and keep the whole research record, and can we prove which parts we cannot reach

Overview

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

Where we are. The estate can already FETCH anything it can name, over its own TLS, with fetch-time provenance binding every mirror to the url it claims, content-addressed identity for the bytes, and a tamper-evident append-only store to keep them in. What it cannot do is REACH THE RECORD AS A WHOLE: the autonomous frontier scanner is wired to exactly two sources, arXiv and HN Algolia, so everything outside those two is invisible to it and its misses read as absences rather than as blind spots. AND A CORRECTION TO THIS LINE'S FIRST DRAFT, WHICH CLAIMED THE ESTATE HAD NO SOURCE REGISTRY AND NO HARVEST PROTOCOL: THAT WAS FALSE ABOUT OUR OWN ESTATE AND nx_memfind REFUTED IT IN ONE CALL. nishi-library reached v0.0.102 in the Python and SQLite era, LIVE at nishifamily.com/research since 2026-04-29, carrying 2.2 million works, 446 thousand artifacts, 1.87 million citation edges, 71 thousand retractions, a FIFTY-PLUS SOURCE REGISTRY ACROSS NINE TIERS, a GENERIC OAI-PMH ADAPTER already covering DOAB, OAPEN, AAAI, AIIDE and NSF-PAR, resumptionToken and cursor persistence per adapter, DV2.0 hub-satellite-link with FTS5, and a render pipeline through Docling, GROBID and Marker. So the estate HAS solved most of this once, in a stack the sovereignty order forbids in the build path. THE REAL GAP IS THEREFORE A PORT AND A MEASUREMENT, NOT A GREENFIELD BUILD: the SOVEREIGN side has none of it, the frontier scanner still sees two sources, and no layer on either side measures REACH. That is why a fleet-wide sweep has been correctly refused since 2026-08-20 rather than run blind.

Where we need to go. Every source in the declared registry is either REACHED on a cadence or carries a NAMED blocker with the remedy that would close it, at a printed and obeyed per-host pace budget, expanding to the institutional tier through OAI-PMH registries rather than through a longer list of urls; and every document acquired is parsed with its structure intact, joined on a stable identifier, and compiled into a hybrid graph-and-vector memory by an agent whose writes cannot land until the existing claim critic has graded them.

The unit. 1 u = one measured session-leg. Local calibration: the source registry, this board and its evidence register landed inside one leg on 2026-09-06, including nine mirrored and pinned references and a full enumeration of the bar listing.

Where we are: 13 open rungs. Fetch, identity, provenance and store are present and are three measured exceeds. Registry is data as of today. Reach, harvest, cadence and the whole compile pipeline are open. Counts measured at emit below this line.

Cost to prove what we cannot reach: 1.5 u. RR1 and RR2 together: the registry reader plus the probe whose partition SUMS and whose UNPROBED is its own state, so no source can be silently absent from the count.

Cost to a safe fleet cadence: 2.5 u. RR3 the printed pace budget, then RR6 the staggered beat. This is the pair that unblocks deepresearch DN1, which has been held open since 2026-08-20 on exactly these two conditions.

Cost to the institutional tier: 2 u. RR4 OAI-PMH with resumption and RR5 registry expansion. These two turn 39 declared sources into thousands of reachable repositories without adding a single hand-written url.

Cost to the 2026 compile pipeline: 9 u. RR9 layout-aware parsing, RR11 hybrid graph and vector, RR12 the gated agentic compile, with RR7 and RR8 as their substrate.

Research bar. OpenAlex is measured on key-free REST over the open catalogue with polite-pool guidance. Theirs: the key-free index bar. Ours: RR1 and RR3 measured on this page; also the second index deepresearch DN1 requires.

Research bar. CORE v3 is measured on aggregated open-access corpus with full-text search and retrieval. Theirs: the full-text coverage bar. Ours: RR8 measured on this page.

Research bar. OAI-PMH v2.0 is measured on six verbs, resumption tokens, from-until selective harvest. Theirs: the harvest-protocol bar. Ours: RR4 and RR5 measured on this specification.

Operator research-source expansion 2026-09-10: REQUESTED / PARTIAL OBSERVATION. Add sources through existing registry and RR1/RR2 reach measurement, not a parallel miner. Root retrieved https://www.darpa.mil/work-with-us/opportunities today: official opportunities, topic/office filters and RSS; retain deadline, eligibility, retrieval and prerequisites. Funded challenge targets are not demonstrated SOTA. Root retrieved https://www.kaggle.com and official https://github.com/Kaggle/kaggle-api redirect to kaggle-cli: competition, dataset, model, notebook and discussion interfaces. Link problem -> metric -> solution -> reproduction -> estate applicability; retain dataset/license/version/split, leakage, public/private leaderboard, generalization and measured cost. Prize/popularity does not prove demand or validity. No submissions or new account connections authorized by this record.

Game and AI benchmark source leads 2026-09-10: UNVERIFIED LEADS. Operator names js13kGames, Ludum Dare Compo/Jam, Gamedev.js Jam, MIT Battlecode, CodeCombat AI League, CodinGame Clash of Code, GitHub Game Off, 7DFPS, Halite, CodeCup, SSCAIT, MicroRTS and Pommerman. Deduplicate and verify official current or historical rules, active status, source availability and each repository license before reuse. Universal claims that submissions must publish code are unverified. Connect compact game implementations, mechanics and AI controllers to existing gameengine and swarm owners with constraints, score, reproduction and cost; archive retired benchmarks as historical foundations, never rank frontier by popularity or age. Code visibility does not confer reuse/hosting permission. Beach remains primary.

Shared research-owner defects 2026-09-10: IDENTIFIED / REPAIR UNFINISHED. Root full-read native nx_infinigen_research_fetch.nx: nine hardcoded requests, mixed unpinned branches, fixed 2023/2024 date filters, OpenAlex per-page=15 without pagination, work-account contact and unconditional main success. Do not run unchanged. Root full-read nx_research_engine.nx: rf_fetch_bank accepts existing text without age/revalidation, also reuses raw indefinitely; failed rf_slot_acquire does not prevent fetch. Repair existing owners with explicit partial/error and pagination completeness, dated version/retrieval/qualification, configurable query lineage and enforced resource admission. Existing source is evidence of coverage/freshness defects, not complete current research. Root owns source repair after Beach priority.

Outcome-driven ingestion requirement 2026-09-10: REQUESTED / UNFINISHED. Competition outcomes and available code, plus Steam player reviews, must inform learning and first-byte design through source observation -> scoped claim -> failure or successful-technique taxonomy -> affected native capability -> proposed telemetry/check -> reproduction -> accepted/rejected result -> Compare rung and investment. Capture successful patterns as well as failures; review sentiment and code popularity are not direct outcome measurements. Record review/game version, context and uncertainty; respect supported interfaces and access rights. Instrument actual player outcomes and resource cost, then test whether a proposed technique improves them. Corpus size alone never completes a rung. Reuse researchreach, gameengine, telemetry and Compare owners; Beach remains the first concrete application.

Mod/plugin and embodied-agent source candidates 2026-09-10: UNVERIFIED LEADS. Operator supplies BuiltByBit Plugin Jam thread 693770, Spigot/Paper community jams, https://mod.curseforge.com/minecraft/modjam2026/ and https://mod.curseforge.com/minecraft-modjam-tos/, https://github.com/JamCoreModding, and MineRL docs/minerllabs/GitHub dzorlu/minerl. Verify current or historical official rules, actual artifacts and individual licenses. A public OR private judging repository requirement does not establish a public corpus; claims that every winner publishes code remain unverified. Mine mod/plugin lifecycle, compatibility, persistence, input and agent behavior for scoped techniques and reproducible regression oracles relevant to Beach/swarm. No contest entry, dataset training or automatic code reuse implied. Keep these in the existing source catalogue and outcome-learning pipeline.

Historical lineage and constrained-game leads 2026-09-10: UNVERIFIED LEADS. Operator adds Infiniminer, Java4K, Dungeon4K, Minecraft4K, Minicraft and TIGSource as Minecraft-history leads; distinguish influence from actual code dependency, licensed release from alleged leak, emergent player behavior from creator intent. No causal or code-reuse assertion accepted without evidence. Additional competition leads: 20SecondGameJam, PICO-8 jams, GBJam, AmigaJam and https://itch.io/jam/gmtk-jam-2026. Verify official rules before accepting claims about engines, mandatory GitHub publication, actual hardware or code-from-scratch. Extract size, latency, memory, creation-time and experience-length constraints as optional benchmark profiles, not estate-wide restrictions. Prioritize high capability and Beach-relevant implementation evidence; no separate history project or competition entry.

Latest recorded release

2026-10-03 · RR26

HUNT PAGING SHIPPED 2026-10-03: nx_frontier_scan a4dbd86d (189960 B, /api/promote with expect_sha256 after nx_contentdiff named its one lost run, the renamed receipt path; _offc twin installed) reads a huntpage row per source from knowledge/research_sources.conf (total marker, cursor/page/offset/none, page arg, next-token span, echo span; 10 rows landed), pages up to hunt_page_budget (THRESHOLD 10 in knowledge/frontier_scan.conf with its calibration), percent-encodes cursors, prints SOURCE-TOTAL fetched/total/pages/budget/capped/stop/paging/echo_permil/answer per source, classes a source OFF-QUERY when its records all miss the subject and it does not echo the query (counted unanswered, records excluded from totals), EMPTY-TOTAL when it declares zero and serves none, and HUNT-DONE now carries answered/off_query/no_items/unreachable with source_partition_sums, records_fetched, records_total_declared, capped_sources and population_complete; receipts are named /tmp/nx_frontier_scan/hunt_<domain>_<tag>.out. Offline twin huntfx reads <dir>/<key>.<page>.raw. nx_frontier_scan_gate 31eb04a1 (ship loop, loss approval 90dba7f9 = two superseded nx_syscalls arena messages, behavior review 9a824c6b) gained 12 hunt teeth on a four-source fixture (cursor to total, page budget binding with a present page 3 that must not be read, off-query neg-control, empty total): 45 of 45 GREEN against the promoted subject, 33 of 45 RED against the pre-change subject (the 12 hunt teeth are the ones that failed). The staged gate .new at 74320 B was a rebuild of the UNCHANGED gate source against the newer nx_syscalls (contentdiff: only the two arena messages lost), not mine; superseded by 31eb04a1. Zenodo join changed to AND (OR answered 85569, AND 26, rf_05694ba2.raw). nx_licclass_gate authorised (cap minted, nonce 1791045120) and exercised 64 of 64 GREEN.

Release history and work log

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

12 of 30 capability source symbols found|4 of them authored stronger-claims|18 open|coverage 400/1000|adoption 8 full / 4 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 researchreach 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

Ladder to SOTA

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

best-in-class (RT1), dated by RB1 = 2026-09 UNATTESTED. Reach and harvest the whole declared research record -- 39 declared sources plus the institutional OAI-PMH tier reached through registries -- where every source is either REACHED on a cadence or carries a NAMED blocker and the remedy that would close it, under a per-host pace budget the run prints before it issues its first request. EXCEED CANDIDATE STATED PLAINLY: none of the four rivals publishes a per-source reachability verdict at all, so proving what we cannot reach is a lead nobody currently holds. ref openalex_api

frontier (RT2), dated by RB2 = 2026-09 UNATTESTED. Compile acquired documents into a hybrid graph-and-vector memory by a staged agent whose writes are gated by the claim critic the estate already owns, so that nothing enters memory ungraded. OUR HEAD START IS THE CRITIC AND IT IS UNUSUAL: nxr_assess already grades every numeric claim against its cited source and gates publish, and cvx_answer_verify already corroborates across independent origins. The frontier papers describe the governance we would otherwise have to invent. ref agentkg2026

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Sovereign fetch, identity, provenance and store (RR0)	crg_prov_state	LANDED before this board existed. Fetch over own TLS, sha256 identity for the bytes, fetch-time provenance binding mirror to url in three states, tamper-evident append-only segments. Verified on disk 2026-09-06.	Organ	0 u
Source registry reader (RR1) after RR0	sr_load	Parses knowledge/research_sources.conf into typed rows and REFUSES a row whose protocol is not one of the six declared kinds, whose pace is unparseable, or whose kind is unknown. Gate proves a malformed row is refused BY NAME rather than skipped, because a registry that silently drops a row under-reports coverage in the flattering direction.	Organ	0.5 u
Reachability probe with the blocker named (RR2) after RR1	sr_probe	Probes every registry row and emits a partition that SUMS to the row count over REACHABLE, RATE-LIMITED, TLS-FAIL, AUTH-REQUIRED, ROBOTS-DENIED, NOT-FOUND and UNPROBED. UNPROBED is its own state so a source nobody probed can never read as one that passed. Each non-reachable row carries its remedy. Gate is bite-proven in both directions and carries a NEGATIVE CONTROL that must stay REACHABLE, because a probe that refused everything would pass every absence test.	Organ	1 u
Per-source pace budget printed and obeyed (RR3) after RR1	sr_pace	The run PRINTS sum of planned requests before issuing the first one and REFUSES to start when that exceeds the declared ceiling. Budget is per host, not global, because a shared budget starves the slow hosts first. Gate proves the printed count equals the requests actually issued, and that a ceiling breach refuses rather than truncates.	Organ	1 u
OAI-PMH harvest with resumption (RR4) after RR1,RR3	oai_list_records	ListRecords and ListIdentifiers over the six verbs with resumption tokens followed to exhaustion and a from-until selective window. Gate proves a multi-page response is followed to its end and that a run stopped early reports PARTIAL rather than success, because a harvester that stops at page one acquires a prefix and reports a whole.	Organ	1.5 u
Registry-of-registries expansion (RR5) after RR4	oai_registry_expand	Walks OpenDOAR and ROAR, admits each discovered endpoint as a derived source under the default pace, and writes it to knowledge/status/oai_endpoints.derived WITH the registry that produced it. Gate proves a derived endpoint is never written back into the hand-authored registry and that removing a registry removes exactly its own children.	Organ	1 u
Staggered fleet cadence (RR6) after RR3,RR2	rh_beat	Every /compare domain revalidated against the record on a beat, STAGGERED across domains so outbound load is flat rather than a daily spike, under RR3's budget. Done when a full pass states its own request count, that count matches what was issued, and the beat writes a truncate-written verdict log whose LAST line is canonical. THIS RUNG IS THE ONE THAT CLOSES deepresearch DN1, whose two blocking conditions are RR3 and a second key-free index, both supplied here.	Organ	1.5 u
Cross-source dedup on a stable identifier (RR7) after RR4	xr_doi_join	The same work harvested from a preprint server, an institutional repository and a publisher collapses to one record joined on DOI, with the alternatives kept as sources rather than discarded. Gate proves a planted three-way duplicate collapses to one and that a record with no DOI is bucketed as UNJOINABLE rather than dropped.	Organ	1 u
Open-access full text acquired (RR8) after RR7	oa_fulltext	A DOI resolves to its legal open-access copy and the document is acquired, content-addressed and stored. Gate proves a closed-access DOI yields a NAMED refusal and never a fabricated body, because a pipeline that invents a full text is worse than one that admits it has none.	Organ	1.5 u
Layout-aware parsing (RR9) after RR8	lp_blocks	A PDF becomes a typed block tree in which a table round-trips to its cells and an equation survives as its own region, rather than the flattened text stream nx_pdf_text produces today. Gate carries an ANTI-VACUITY tooth: a plain text extractor must FAIL it, because a checker that only asks whether text came out is passed by the tool we already have.	Organ	3 u
Code graphing for acquired repositories (RR10) after RR2	cg_ast_edges	An acquired repository becomes an AST-level call graph, generalising what nx_funcmine and nx_capgraph_edges already do for our own tree. Arm rather than contender: it triangulates the record against its implementations and does not compete for the coverage number.	Organ	2 u
Hybrid graph and vector retrieval (RR11) after RR7,RR9	kgh_query	One query answered against a knowledge graph and a vector index together, with the graph carrying the strict relations and the vectors the semantic reach. Accept rule DECLARED BEFORE THE RUN: must EXCEED the lexical arm on the same held-out questions on a pre-declared ruler, or it stays UNWIRED and the number is published anyway.	Organ	3 u
Gated agentic compile (RR12) after RR11	ag_stage	Staged gather, write and criticise, where the writing stage CANNOT commit to the store until the critic has graded it. Gate proves a planted false claim never lands, and the critic is the one the estate already owns rather than a second one built beside it.	Organ	3 u
The drift comparator gets its contracted entry point (RR13)	rd_same	CORRECTION RUNG, not a new capability. nx_refdrift's four-state comparator is real and runs inline in main, so the deepresearch row claiming symbol rd_same publishes as UNGROUNDED while the capability it names genuinely works. Extract the comparator to the contracted name, prove neutrality by nx_behaveprobe against the live binary, and the sibling row grounds itself. Done-rule: nx_domain_admit deepresearch reports ungrounded 0.	Organ	0.5 u

Rung	Role	Toward	Stands on	Why
RR0	substrate	RT1	-	Fetch, identity, provenance and store are what every reach and harvest rung stands on; without content-addressed identity the dedup rung has nothing stable to join on.
RR1	substrate	RT1	RR0	The registry is the population every coverage and reachability number is measured over. A count without a declared denominator is not a coverage claim.
RR2	contender	RT1	RR1	Its done-rule IS the target's number: every source reached or its blocker named. This is the row the board was admitted for.
RR3	substrate	RT1	RR1	The budget is what makes reaching the whole record safe to do repeatedly rather than once.
RR4	contender	RT1	RR1,RR3	OAI-PMH is the protocol that makes the institutional tier reachable at all, so it competes directly on the coverage number.
RR5	contender	RT1	RR4	Registry expansion is how the denominator grows from 39 to thousands without hand-written urls.
RR6	contender	RT1	RR3,RR2	Cadence is the difference between having reached the record once and staying current with it, which is the operator's stated requirement.
RR7	substrate	RT2	RR4	Dedup is the join the compile pipeline needs before a graph can have one node per work rather than three.
RR8	contender	RT1	RR7	Full text acquired rather than merely indexed is the axis CORE leads on and the one our coverage number is weakest on.
RR9	contender	RT2	RR8	Structure-preserving parsing is stage one of the frontier pipeline; a flattened text stream cannot feed a graph that needs typed regions.
RR10	arm	RT2	RR2	A sovereign arm that triangulates papers against their implementations. It does not compete for the coverage number and is scored separately so it cannot flatter it.
RR11	contender	RT2	RR7,RR9	The hybrid join IS the frontier architecture named by the 2026-09 bar.
RR12	contender	RT2	RR11	The gated compile is the governance half of the frontier bar, and the half where our existing critic is already ahead.
RR13	substrate	RT1	-	A board whose sibling carries an ungrounded exceed claim cannot be trusted about its own grounding. Correcting it is substrate for every claim this board makes.

Milestones

Milestone	Rungs	Cumulative
M1 · We can prove what we cannot reach	RR1,RR2,RR13	2.5 u
M2 · The whole record, on a safe cadence	RR3,RR4,RR5,RR6	6.5 u
M3 · Acquired and compiled	RR7,RR8,RR9,RR11,RR12	13 u

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Port the legacy source registry and OAI-PMH adapter to NishiLang (RR14) after RR4,RR5	sr_port_parity	THE ESTATE ALREADY SOLVED MOST OF THIS ONCE, IN A STACK THE SOVEREIGNTY ORDER FORBIDS IN THE BUILD PATH. nishi-library v0.0.102 carries a fifty-plus source registry across nine tiers and a GENERIC OAI-PMH adapter with resumptionToken cursors already covering DOAB, OAPEN, AAAI, AIIDE and NSF-PAR, in Python over SQLite. Done-rule is PARITY MEASURED AND NOT ASSERTED: the sovereign harvester reaches the same declared sources and returns the same record count for one named tier, proven side by side, before the legacy path is retired. Retire by nx_retire_path so it is reversible, never by deletion, and never before parity is shown -- a legacy stack that works is not litter.	Organ	3 u

Rung	Role	Toward	Stands on	Why
RR14	substrate	RT1	RR4,RR5	Coverage the estate already has in a forbidden stack is coverage the sovereign side does not have. The port is what turns a working legacy number into a sovereign one, and until it lands every coverage claim on this board must say which stack it was measured on.

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
A research paper retains complete human and machine evidence (RR15) after RR8,RR9	_ABSENT_:paper_dual_evidence_accept	SOURCE-BOUND PLANNING CONTRACT 2026-09-24. Extend existing fetch, library, layout parser and document emitter owners. Store exact source identity and access/rights state beside typed text, tables, equations, figures, captions and reference links with source-region correspondence. Produce readable human rendering and machine records from the same retained evidence, with unsupported regions and uncertainty explicit. Verify a representative multicolumn paper including a table, equation and figure; round-trip table cells and references, preserve original bytes, and show a flattened-text negative control cannot claim complete extraction. Incremental versions retain predecessors and changed claims; a failed fetch never becomes an empty successful paper. Comparisons with reference parsers are independent measured work, not supplied-name superiority. Effort unestimated.	Workflow	unestimated u

Rung	Role	Toward	Stands on	Why
RR15	substrate	RT2	RR8,RR9	Complete region-bound paper evidence is the shared input for structured parsing, graph compilation and reproducible reports.

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Venue-tag scans as frontier source rows (RR16) after RR1	_ABSENT_:fs_venue_rows	The arXiv API indexes the comment field, where authors write the venue: co:SIGGRAPH 2026, co:SIGGRAPH Asia 2026, co:Eurographics 2026, co:CVPR 2026, co:ECCV 2026, co:NeurIPS 2026 and co:IEEE VR 2026 become DATED source rows in knowledge/frontier_sources.conf with the seed ANDed in, so nx_frontier_scan attest asks every venue what the bar's subject says this month. Landed as rows 2026-09-27, each proven reachable the same day; the rung closes when attest prints a per-venue NEWER, SAME or OLDER line on a live board. Done-rule: one attest run on graphics prints the venue rows beside the arXiv and HN rows.	Organ	1 u
Static venue indices and the Eurographics digital library (RR17) after RR1,RR4	_ABSENT_:fs_static_venue_rows	Ke-Sen Huang's per-venue indices (SIGGRAPH Asia 2026, Eurographics 2026, EGSR, HPG, I3D, SCA) join the SIGGRAPH 2026 row as static month-tagged sources, and the Eurographics Digital Library's OAI-PMH endpoint joins research_sources.conf as a harvestable publisher; the extractor is the existing generic dt-B grammar. Done-rule: attest reads each index and the SIGGRAPH Asia row grows from 7 to the proceedings count by December.	Organ	1 u
Venue completeness by DBLP (RR18) after RR7	_ABSENT_:sr_dblp_population	A venue index read at a cap is a PREFIX; DBLP's publication API lists every paper of a venue and year and is the population count a mirrored index is reconciled against, so coverage is a fraction with a denominator rather than a count. Done-rule: the researcher prints mirrored over DBLP-listed per venue with the partition summing.	Organ	2 u
Labs, standards bodies and governments on the registry (RR19) after RR1,RR2	_ABSENT_:sr_lab_rows	NVIDIA Research, Meta Reality Labs, Disney Research, MPI Perceiving Systems, Google, Microsoft, Adobe, Epic and Apple publication records; Khronos and MPEG registries; NIST, DARPA, CORDIS and NIH RePORTER programme records: each a research_sources.conf row under a declared kind and pace, reached on the fleet cadence with the srcreach partition naming any blocker. Landed as rows 2026-09-27 under the publisher kind because sr_load refuses an unknown kind. Done-rule: nx_srcreach probe reads every row REACHABLE or names its blocker, and a lab kind is admitted to the registry reader rather than borrowing publisher.	Organ	1.5 u
Login walls and script shells named as blockers with the sovereign browser as the door (RR20) after RR2	_ABSENT_:sr_shell_blocker	Measured 2026-09-27: the MetaHuman news site redirects to an Epic login (a login wall) while the developer documentation site serves the same release notes as 259454 bytes of HTML; earlier the MetaHuman encyclopedia page returned status-0 fragments and the Unreal release-notes page an Angular shell. A vendor row therefore carries an ALTERNATE WITNESS url, and the probe classifies AUTH-REQUIRED and SCRIPT-SHELL separately with the sovereign browser (nx_browser, graphics GR53's CDP witness) as the door for shells. Done-rule: every vendor bar on the boards cites bytes the estate mirrored, never a search summary.	Organ	2 u
Page to the month boundary (RR21) after RR4	_ABSENT_:ad_page_month	A newest-first listing fetched at max_results is a PREFIX (CVPR 2026 human read 120 at its cap; the splat listing read 30 of 771). The digester pages with start= until the first entry older than the bar month, so a month's population is complete by construction and nx_arxiv_census reconciles it against totalResults. Done-rule: every barscan row prints entries read equal to the month's population or names the cap it stopped at.	Organ	1.5 u
Homonym subject classifier before newest (RR22) after RR21	fs_cls_admit	Debt 1789839847: attest took the newest listing item as the bar month while the item was the KV-cache InfiniGen homonym, and the 2026-09-27 wet scan's newest entry is a quantum-gravity paper on the phrase wet hair. BUILT 2026-09-27 in the attester (runtime/nx_frontier_scan.nx, fs_cls_admit called inside fs_newest_month before any max is taken): the axis is the listing's OWN taxonomy, arXiv category terms matched any-of against the classes a board declares as data in knowledge/subject_classes.conf, because subject tokens cannot separate a homonym that shares the query token (every entry on both listings contains it). ON-SUBJECT admitted, OFF-SUBJECT named on its own row and excluded from the max, UNCLASSIFIED admitted and counted, an undeclared board says undeclared and keeps its max; a listing whose every classified item is OFF attests nothing rather than a newer month. The summary line carries on= off= unclassified= and off_subject_newer=. Done-rule: the two homonyms above classify OFF on a live run and a control paper classifies ON.	Organ	2 u
Full text behind every bar claim (RR23) after RR8,RR9	_ABSENT_:ad_fulltext_claim	A bar written from an abstract is a claim about an abstract. The open-access full text (RR8, unpaywall) is read through nx_pdf_text and the sentence the bar rests on is quoted with its byte offset in the mirrored file, so a bar row can be refuted by a reader without a fetch. Done-rule: the sotabar rows landed 2026-09-27 on graphics, gameengine, skintwin and softbody each gain a quoted sentence with its offset, or name the paywall that prevents it.	Organ	3 u

Rung	Role	Toward	Stands on	Why
RR16	contender	RT1	RR1	venue tags make the reach per-venue, the RT1 claim
RR17	contender	RT1	RR1,RR4	static venue indices and an OAI publisher widen the declared record
RR18	arm	RT1	RR7	DBLP is the population oracle the reach fraction is measured against
RR19	contender	RT1	RR1,RR2	labs, standards and governments are sources the registry did not name
RR20	contender	RT1	RR2	a login wall or script shell is a named blocker with its door
RR21	substrate	RT2	RR4	paging to the month boundary makes the compile complete
RR22	substrate	RT2	RR21	subject classification gates what enters memory
RR23	contender	RT2	RR8,RR9	a quoted sentence with its offset is the graded claim RT2 demands

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
PubMed as a dated frontier source for clinical bars (RR24) after RR16,RR19	_ABSENT_:fs_pubmed_attest	MEASURED 2026-09-28: the arXiv listing for aesthetic-surgery 3D simulation holds 16 entries back to 2006 while one PubMed E-utilities search returns 28 records in the last twelve months, fetched and read by the researcher (twiningest refs twin_pubmed_search_202609 and twin_pubmed_abstracts_202609). The registry already carries the pubmed row, but frontier_sources.conf has no E-utilities grammar, so the attester cannot date a clinical bar and those bars are hand-attested. Grow the attester: esearch sorted by publication date, efetch for the records, the month read from the online publication date (one record carried a December 2026 issue date over a January 2026 epub), and the same homonym classifier by MeSH heading. Done-rule: nx_frontier_scan attest twiningest dates SB-TI-CLINIC from PubMed bytes on a live run and names any record it could not date.	Organ	1.5 u

Rung	Role	Toward	Stands on	Why
RR24	contender	RT1	RR16,RR19	reach into the biomedical record, measured on a live attest of a clinical bar

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Open-format full text first (RR25)	_ABSENT_:ft_open_fulltext	OPERATOR 2026-09-28: no proprietary formats anywhere in the estate where an open form exists, and arXiv's PDF-to-HTML conversion or better. A paper's full text is fetched in the order of its most structured OPEN form: arXiv's HTML (LaTeXML) first, then the arXiv e-print source (LaTeX, unpacked by the estate's own gunzip and untar), then PMC's JATS XML for the biomedical record, and PDF only last through RR9. Headings, equations (one copy, the TeX annotation never duplicated beside its rendering) and table cells survive as typed text. Done-rule: on a declared set of recent papers the open route returns complete text for every paper that publishes an open form, the mangled-word rate (split words, lost ligatures) is published beside the PDF route's on the same papers, and a paper with no open form falls through to RR9 by name rather than returning nothing	Organ	2 u

Rung	Role	Toward	Stands on	Why
RR25	contender	RT1	-	its done-rule is complete text and a published mangled-word rate against the PDF route on the same papers

Rung	Closes with	Definition of done (pre-declared)	Executor	Est.
Dataset and media hunt from anywhere, licence-classed (RR26) after RR1,RR2	fs_hunt	OPERATOR 2026-10-03: build our capabilities in the nishi estate researcher to get from anywhere, under the licence rule (acquire only when the licence puts no royalty and no obligation on the product or delivery path; restricted data only as math and only where the licence permits; files stored by licence class). One verb in the incumbent frontier scanner, nx_frontier_scan hunt <domain> <tag> <subject words>, over the SAME fetch, per-host pace and item-extent machinery, against dataset and media repositories declared as hunt rows in knowledge/research_sources.conf, each naming where in the hit's own record the licence lives [rr26_datacite_dois_202610] [rr26_europepmc_core_202610]. Every hit's licence is classed by nx_licclass_lib against knowledge/license_classes.conf (ACQUIRE, CALIBRATE-ONLY, ACQUIRE-PRIVATE-REFERENCE, UNKNOWN-NEEDS-LICENCE-READ, REFUSED with the clause named) and written to a ranked worklist with the evidence and the deciding row. Done-rule: the OpenOximetry precedent is refused mechanically from the repository's own licence field [rr26_physionet_catalogue_202610], a live run names a permitted calibration set the seat did not know, UNKNOWN is never acquired, and every partition sums	Organ	2 u

Rung	Role	Toward	Stands on	Why
RR26	contender	RT1	RR1,RR2	reach into the dataset and media record with a licence verdict per hit, measured on a live hunt whose partition sums

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

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

Ladder verdict, counted
Class	Count	Note
targets	2	sotatarget rows on the plan
best in class	1	targets whose class is best-in-class, by code
frontier	1	targets whose class is frontier, by code
undated	0	targets naming no bar that exists
bad class	0	targets whose class is neither word
duplicate	0	target ids an earlier row already carries
rungs	27	rung rows on the plan
placed	27	rungs carrying a rungrole row
unplaced	0	rungs with no role toward any target
bad role	0	rungrole rows naming a role the ladder does not know
bad target	0	rungrole rows naming a target no sotatarget row declares
rungs, summed	27	RECONCILED: the parts sum to the total 27
substrate	9	what the target's own methods consume
arm	2	a sovereign arm that triangulates without competing for the number
contender	16	its done-rule is a number against the target
superseded	0	kept as capability, retired as a claim
placed, summed	27	RECONCILED: the parts sum to the total 27

Bar freshness, counted
Class	Count	Note
bars	4	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	4	nobody looked this month
malformed	0	a sotabar row missing a field
duplicate	0	bar ids an earlier row already carries
bars, summed	4	RECONCILED: the parts sum to the total 4
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 27. 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 d8b8ac3e4e651a6d0281dd0d4848f24f3b05f0c44703de04268839a622423a90. Target rows 2; role rows 27. Existing risks and release worklog retain their own scope; no node completion is inferred.

Runtime observation coverage: 0 of 27 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.

RR0 — Sovereign fetch, identity, provenance and store
Prerequisites: None declared; this does not establish execution eligibility.
Contract: crg_prov_state
Acceptance: LANDED before this board existed. Fetch over own TLS, sha256 identity for the bytes, fetch-time provenance binding mirror to url in three states, tamper-evident append-only segments. Verified on disk 2026-09-06.
Authored effort: 0. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
RR1 — Source registry reader
Prerequisites: RR0 (acceptance unverified)
Contract: sr_load
Acceptance: Parses knowledge/research_sources.conf into typed rows and REFUSES a row whose protocol is not one of the six declared kinds, whose pace is unparseable, or whose kind is unknown. Gate proves a malformed row is refused BY NAME rather than skipped, because a registry that silently drops a row under-reports coverage in the flattering direction.
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
RR2 — Reachability probe with the blocker named
Prerequisites: RR1 (acceptance unverified)
Contract: sr_probe
Acceptance: Probes every registry row and emits a partition that SUMS to the row count over REACHABLE, RATE-LIMITED, TLS-FAIL, AUTH-REQUIRED, ROBOTS-DENIED, NOT-FOUND and UNPROBED. UNPROBED is its own state so a source nobody probed can never read as one that passed. Each non-reachable row carries its remedy. Gate is bite-proven in both directions and carries a NEGATIVE CONTROL that must stay REACHABLE, because a probe that refused everything would pass every absence test.
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.
RR3 — Per-source pace budget printed and obeyed
Prerequisites: RR1 (acceptance unverified)
Contract: sr_pace
Acceptance: The run PRINTS sum of planned requests before issuing the first one and REFUSES to start when that exceeds the declared ceiling. Budget is per host, not global, because a shared budget starves the slow hosts first. Gate proves the printed count equals the requests actually issued, and that a ceiling breach refuses rather than truncates.
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.
RR4 — OAI-PMH harvest with resumption
Prerequisites: RR1 (acceptance unverified), RR3 (acceptance unverified)
Contract: oai_list_records
Acceptance: ListRecords and ListIdentifiers over the six verbs with resumption tokens followed to exhaustion and a from-until selective window. Gate proves a multi-page response is followed to its end and that a run stopped early reports PARTIAL rather than success, because a harvester that stops at page one acquires a prefix and reports a whole.
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.
RR5 — Registry-of-registries expansion
Prerequisites: RR4 (acceptance unverified)
Contract: oai_registry_expand
Acceptance: Walks OpenDOAR and ROAR, admits each discovered endpoint as a derived source under the default pace, and writes it to knowledge/status/oai_endpoints.derived WITH the registry that produced it. Gate proves a derived endpoint is never written back into the hand-authored registry and that removing a registry removes exactly its own children.
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.
RR6 — Staggered fleet cadence
Prerequisites: RR3 (acceptance unverified), RR2 (acceptance unverified)
Contract: rh_beat
Acceptance: Every /compare domain revalidated against the record on a beat, STAGGERED across domains so outbound load is flat rather than a daily spike, under RR3's budget. Done when a full pass states its own request count, that count matches what was issued, and the beat writes a truncate-written verdict log whose LAST line is canonical. THIS RUNG IS THE ONE THAT CLOSES deepresearch DN1, whose two blocking conditions are RR3 and a second key-free index, both supplied here.
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.
RR7 — Cross-source dedup on a stable identifier
Prerequisites: RR4 (acceptance unverified)
Contract: xr_doi_join
Acceptance: The same work harvested from a preprint server, an institutional repository and a publisher collapses to one record joined on DOI, with the alternatives kept as sources rather than discarded. Gate proves a planted three-way duplicate collapses to one and that a record with no DOI is bucketed as UNJOINABLE rather than dropped.
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.
RR8 — Open-access full text acquired
Prerequisites: RR7 (acceptance unverified)
Contract: oa_fulltext
Acceptance: A DOI resolves to its legal open-access copy and the document is acquired, content-addressed and stored. Gate proves a closed-access DOI yields a NAMED refusal and never a fabricated body, because a pipeline that invents a full text is worse than one that admits it has none.
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.
RR9 — Layout-aware parsing
Prerequisites: RR8 (acceptance unverified)
Contract: lp_blocks
Acceptance: A PDF becomes a typed block tree in which a table round-trips to its cells and an equation survives as its own region, rather than the flattened text stream nx_pdf_text produces today. Gate carries an ANTI-VACUITY tooth: a plain text extractor must FAIL it, because a checker that only asks whether text came out is passed by the tool we already have.
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
RR10 — Code graphing for acquired repositories
Prerequisites: RR2 (acceptance unverified)
Contract: cg_ast_edges
Acceptance: An acquired repository becomes an AST-level call graph, generalising what nx_funcmine and nx_capgraph_edges already do for our own tree. Arm rather than contender: it triangulates the record against its implementations and does not compete for the coverage number.
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.
RR11 — Hybrid graph and vector retrieval
Prerequisites: RR7 (acceptance unverified), RR9 (acceptance unverified)
Contract: kgh_query
Acceptance: One query answered against a knowledge graph and a vector index together, with the graph carrying the strict relations and the vectors the semantic reach. Accept rule DECLARED BEFORE THE RUN: must EXCEED the lexical arm on the same held-out questions on a pre-declared ruler, or it stays UNWIRED and the number is published anyway.
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
RR12 — Gated agentic compile
Prerequisites: RR11 (acceptance unverified)
Contract: ag_stage
Acceptance: Staged gather, write and criticise, where the writing stage CANNOT commit to the store until the critic has graded it. Gate proves a planted false claim never lands, and the critic is the one the estate already owns rather than a second one built beside it.
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
RR13 — The drift comparator gets its contracted entry point
Prerequisites: None declared; this does not establish execution eligibility.
Contract: rd_same
Acceptance: CORRECTION RUNG, not a new capability. nx_refdrift's four-state comparator is real and runs inline in main, so the deepresearch row claiming symbol rd_same publishes as UNGROUNDED while the capability it names genuinely works. Extract the comparator to the contracted name, prove neutrality by nx_behaveprobe against the live binary, and the sibling row grounds itself. Done-rule: nx_domain_admit deepresearch reports ungrounded 0.
Authored effort: 0.5. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
RR14 — Port the legacy source registry and OAI-PMH adapter to NishiLang
Prerequisites: RR4 (acceptance unverified), RR5 (acceptance unverified)
Contract: sr_port_parity
Acceptance: THE ESTATE ALREADY SOLVED MOST OF THIS ONCE, IN A STACK THE SOVEREIGNTY ORDER FORBIDS IN THE BUILD PATH. nishi-library v0.0.102 carries a fifty-plus source registry across nine tiers and a GENERIC OAI-PMH adapter with resumptionToken cursors already covering DOAB, OAPEN, AAAI, AIIDE and NSF-PAR, in Python over SQLite. Done-rule is PARITY MEASURED AND NOT ASSERTED: the sovereign harvester reaches the same declared sources and returns the same record count for one named tier, proven side by side, before the legacy path is retired. Retire by nx_retire_path so it is reversible, never by deletion, and never before parity is shown -- a legacy stack that works is not litter.
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
RR15 — A research paper retains complete human and machine evidence
Prerequisites: RR8 (acceptance unverified), RR9 (acceptance unverified)
Contract: _ABSENT_:paper_dual_evidence_accept
Acceptance: SOURCE-BOUND PLANNING CONTRACT 2026-09-24. Extend existing fetch, library, layout parser and document emitter owners. Store exact source identity and access/rights state beside typed text, tables, equations, figures, captions and reference links with source-region correspondence. Produce readable human rendering and machine records from the same retained evidence, with unsupported regions and uncertainty explicit. Verify a representative multicolumn paper including a table, equation and figure; round-trip table cells and references, preserve original bytes, and show a flattened-text negative control cannot claim complete extraction. Incremental versions retain predecessors and changed claims; a failed fetch never becomes an empty successful paper. Comparisons with reference parsers are independent measured work, not supplied-name superiority. Effort unestimated.
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.
RR16 — Venue-tag scans as frontier source rows
Prerequisites: RR1 (acceptance unverified)
Contract: _ABSENT_:fs_venue_rows
Acceptance: The arXiv API indexes the comment field, where authors write the venue: co:SIGGRAPH 2026, co:SIGGRAPH Asia 2026, co:Eurographics 2026, co:CVPR 2026, co:ECCV 2026, co:NeurIPS 2026 and co:IEEE VR 2026 become DATED source rows in knowledge/frontier_sources.conf with the seed ANDed in, so nx_frontier_scan attest asks every venue what the bar's subject says this month. Landed as rows 2026-09-27, each proven reachable the same day; the rung closes when attest prints a per-venue NEWER, SAME or OLDER line on a live board. Done-rule: one attest run on graphics prints the venue rows beside the arXiv and HN rows.
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.
RR17 — Static venue indices and the Eurographics digital library
Prerequisites: RR1 (acceptance unverified), RR4 (acceptance unverified)
Contract: _ABSENT_:fs_static_venue_rows
Acceptance: Ke-Sen Huang's per-venue indices (SIGGRAPH Asia 2026, Eurographics 2026, EGSR, HPG, I3D, SCA) join the SIGGRAPH 2026 row as static month-tagged sources, and the Eurographics Digital Library's OAI-PMH endpoint joins research_sources.conf as a harvestable publisher; the extractor is the existing generic dt-B grammar. Done-rule: attest reads each index and the SIGGRAPH Asia row grows from 7 to the proceedings count by December.
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.
RR18 — Venue completeness by DBLP
Prerequisites: RR7 (acceptance unverified)
Contract: _ABSENT_:sr_dblp_population
Acceptance: A venue index read at a cap is a PREFIX; DBLP's publication API lists every paper of a venue and year and is the population count a mirrored index is reconciled against, so coverage is a fraction with a denominator rather than a count. Done-rule: the researcher prints mirrored over DBLP-listed per venue with the partition summing.
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.
RR19 — Labs, standards bodies and governments on the registry
Prerequisites: RR1 (acceptance unverified), RR2 (acceptance unverified)
Contract: _ABSENT_:sr_lab_rows
Acceptance: NVIDIA Research, Meta Reality Labs, Disney Research, MPI Perceiving Systems, Google, Microsoft, Adobe, Epic and Apple publication records; Khronos and MPEG registries; NIST, DARPA, CORDIS and NIH RePORTER programme records: each a research_sources.conf row under a declared kind and pace, reached on the fleet cadence with the srcreach partition naming any blocker. Landed as rows 2026-09-27 under the publisher kind because sr_load refuses an unknown kind. Done-rule: nx_srcreach probe reads every row REACHABLE or names its blocker, and a lab kind is admitted to the registry reader rather than borrowing publisher.
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.
RR20 — Login walls and script shells named as blockers with the sovereign browser as the door
Prerequisites: RR2 (acceptance unverified)
Contract: _ABSENT_:sr_shell_blocker
Acceptance: Measured 2026-09-27: the MetaHuman news site redirects to an Epic login (a login wall) while the developer documentation site serves the same release notes as 259454 bytes of HTML; earlier the MetaHuman encyclopedia page returned status-0 fragments and the Unreal release-notes page an Angular shell. A vendor row therefore carries an ALTERNATE WITNESS url, and the probe classifies AUTH-REQUIRED and SCRIPT-SHELL separately with the sovereign browser (nx_browser, graphics GR53's CDP witness) as the door for shells. Done-rule: every vendor bar on the boards cites bytes the estate mirrored, never a search summary.
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.
RR21 — Page to the month boundary
Prerequisites: RR4 (acceptance unverified)
Contract: _ABSENT_:ad_page_month
Acceptance: A newest-first listing fetched at max_results is a PREFIX (CVPR 2026 human read 120 at its cap; the splat listing read 30 of 771). The digester pages with start= until the first entry older than the bar month, so a month's population is complete by construction and nx_arxiv_census reconciles it against totalResults. Done-rule: every barscan row prints entries read equal to the month's population or names the cap it stopped at.
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.
RR22 — Homonym subject classifier before newest
Prerequisites: RR21 (acceptance unverified)
Contract: fs_cls_admit
Acceptance: Debt 1789839847: attest took the newest listing item as the bar month while the item was the KV-cache InfiniGen homonym, and the 2026-09-27 wet scan's newest entry is a quantum-gravity paper on the phrase wet hair. BUILT 2026-09-27 in the attester (runtime/nx_frontier_scan.nx, fs_cls_admit called inside fs_newest_month before any max is taken): the axis is the listing's OWN taxonomy, arXiv category terms matched any-of against the classes a board declares as data in knowledge/subject_classes.conf, because subject tokens cannot separate a homonym that shares the query token (every entry on both listings contains it). ON-SUBJECT admitted, OFF-SUBJECT named on its own row and excluded from the max, UNCLASSIFIED admitted and counted, an undeclared board says undeclared and keeps its max; a listing whose every classified item is OFF attests nothing rather than a newer month. The summary line carries on= off= unclassified= and off_subject_newer=. Done-rule: the two homonyms above classify OFF on a live run and a control paper classifies ON.
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.
RR23 — Full text behind every bar claim
Prerequisites: RR8 (acceptance unverified), RR9 (acceptance unverified)
Contract: _ABSENT_:ad_fulltext_claim
Acceptance: A bar written from an abstract is a claim about an abstract. The open-access full text (RR8, unpaywall) is read through nx_pdf_text and the sentence the bar rests on is quoted with its byte offset in the mirrored file, so a bar row can be refuted by a reader without a fetch. Done-rule: the sotabar rows landed 2026-09-27 on graphics, gameengine, skintwin and softbody each gain a quoted sentence with its offset, or name the paywall that prevents it.
Authored effort: 3. Executor kind: Organ. Responsible, accountable and verifier not established. Inspect retained worklog.
Runtime observation: UNKNOWN; outcome UNKNOWN; observed epoch -1. Declared owner: . Receipt , SHA-256 . Identity-bound observation is not accepted completion. Qualification and deployment remain unverified by this projection.
RR24 — PubMed as a dated frontier source for clinical bars
Prerequisites: RR16 (acceptance unverified), RR19 (acceptance unverified)
Contract: _ABSENT_:fs_pubmed_attest
Acceptance: MEASURED 2026-09-28: the arXiv listing for aesthetic-surgery 3D simulation holds 16 entries back to 2006 while one PubMed E-utilities search returns 28 records in the last twelve months, fetched and read by the researcher (twiningest refs twin_pubmed_search_202609 and twin_pubmed_abstracts_202609). The registry already carries the pubmed row, but frontier_sources.conf has no E-utilities grammar, so the attester cannot date a clinical bar and those bars are hand-attested. Grow the attester: esearch sorted by publication date, efetch for the records, the month read from the online publication date (one record carried a December 2026 issue date over a January 2026 epub), and the same homonym classifier by MeSH heading. Done-rule: nx_frontier_scan attest twiningest dates SB-TI-CLINIC from PubMed bytes on a live run and names any record it could not date.
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.
RR25 — Open-format full text first
Prerequisites: None declared; this does not establish execution eligibility.
Contract: _ABSENT_:ft_open_fulltext
Acceptance: OPERATOR 2026-09-28: no proprietary formats anywhere in the estate where an open form exists, and arXiv's PDF-to-HTML conversion or better. A paper's full text is fetched in the order of its most structured OPEN form: arXiv's HTML (LaTeXML) first, then the arXiv e-print source (LaTeX, unpacked by the estate's own gunzip and untar), then PMC's JATS XML for the biomedical record, and PDF only last through RR9. Headings, equations (one copy, the TeX annotation never duplicated beside its rendering) and table cells survive as typed text. Done-rule: on a declared set of recent papers the open route returns complete text for every paper that publishes an open form, the mangled-word rate (split words, lost ligatures) is published beside the PDF route's on the same papers, and a paper with no open form falls through to RR9 by name rather than returning nothing
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.
RR26 — Dataset and media hunt from anywhere, licence-classed
Prerequisites: RR1 (acceptance unverified), RR2 (acceptance unverified)
Contract: fs_hunt
Acceptance: OPERATOR 2026-10-03: build our capabilities in the nishi estate researcher to get from anywhere, under the licence rule (acquire only when the licence puts no royalty and no obligation on the product or delivery path; restricted data only as math and only where the licence permits; files stored by licence class). One verb in the incumbent frontier scanner, nx_frontier_scan hunt <domain> <tag> <subject words>, over the SAME fetch, per-host pace and item-extent machinery, against dataset and media repositories declared as hunt rows in knowledge/research_sources.conf, each naming where in the hit's own record the licence lives [@rr26_datacite_dois_202610] [@rr26_europepmc_core_202610]. Every hit's licence is classed by nx_licclass_lib against knowledge/license_classes.conf (ACQUIRE, CALIBRATE-ONLY, ACQUIRE-PRIVATE-REFERENCE, UNKNOWN-NEEDS-LICENCE-READ, REFUSED with the clause named) and written to a ranked worklist with the evidence and the deciding row. Done-rule: the OpenOximetry precedent is refused mechanically from the repository's own licence field [@rr26_physionet_catalogue_202610], a live run names a permitted calibration set the seat did not know, UNKNOWN is never acquired, and every partition sums
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.

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 researchreach 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.

RR0 Sovereign fetch, identity, provenance and store · cell PRESENT · gate NO-GATE-DECLARED

Contract	`crg_prov_state` · executor Organ · 0 u · stands on nothing
Definition of done	LANDED before this board existed. Fetch over own TLS, sha256 identity for the bytes, fetch-time provenance binding mirror to url in three states, tamper-evident append-only segments. Verified on disk 2026-09-06.
Matrix cell	Fetch-time provenance binding a mirror to the url it claims · `runtime/nx_compare_refs_gate.nx` · PRESENT 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	no log row names this rung yet

RR1 Source registry reader · cell LANDED · gate NO-GATE-DECLARED

Contract	`sr_load` · executor Organ · 0.5 u · stands on RR0
Definition of done	Parses knowledge/research_sources.conf into typed rows and REFUSES a row whose protocol is not one of the six declared kinds, whose pace is unparseable, or whose kind is unknown. Gate proves a malformed row is refused BY NAME rather than skipped, because a registry that silently drops a row under-reports coverage in the flattering direction.
Matrix cell	Source registry as data rather than hardcoded urls · `runtime/nx_srcreg_lib.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_srcreg_lib`; declare `nx_srcreg_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 lesson AN APPEND TO regen.list DOUBLE-APPLIED AFTER A 503, DESPITE FOLLOWING THE SPACED-READS DISCIPLINE EXACTLY. The call returned 503 Outcome Unknown at stage read-after-accept. Two spaced size reads both returned 1052, so the append was re-issued, and the re-issue reported total=1080 -- which is 1052 plus 14 plus 14, proving the first call had landed all along and both reads were stale. THE READ PATH LAGS THE WRITE PATH FOR EVERY VERB, SO NO NUMBER OF READS IS EVIDENCE THAT A WRITE DID NOT LAND. Only the writer's own reported total settled it, and it settled it instantly. For an append, predict the resulting byte count and read it back from the WRITE, never from a separate size call. 2026-09-06 land RR1 AND RR3 SHIPPED AS SOURCE AND BITE-PROVEN ON THE LAPTOP BUILD FARM, BECAUSE THE NAS COMPILER WAS REFUSING ALL DAY. nx_srcreg_lib.nx (10788 B, sr_load plus the closed kind and protocol vocabularies and sr_classify) and nx_srcreach.nx (9552 B, sr_pace plus the load and plan verbs) both landed in buildroot/runtime. THE LOCAL LANE IS NOT A CONCESSION, IT IS THE DOCUMENTED REMEDY: nx_sov_build_run --build-only on the laptop is governor-free, it compiled first try to 29605 bytes through nx_cc into nxasm_x86 with no gcc anywhere, and the laptop measured load1_centi=19 against the NAS's 1918. PROVEN IN BOTH DIRECTIONS ON A FIXTURE ASSEMBLED AT RUNTIME: four planted defects each FIRED and each was NAMED by its key -- BAD-KIND, BAD-PROTO, BAD-PACE, BAD-FIELDS -- with the partition summing exactly, rows=6 ok=2 bad_fields=1 bad_kind=1 bad_proto=1 bad_pace=1 sum=6 verdict=REFUSED exit=1. THE POSITIVE CONTROL IS THE HALF THAT MATTERS MOST and it is why this is not a guard that refuses everything: a clean three-row registry read rows=3 ok=3 sum=3 verdict=OK exit=0. The budget was proven both ways too: 3 sources at 10 requests each is 30 planned, under a ceiling of 100 it passes exit 0, over a ceiling of 20 it prints REFUSED-OVER-CEILING and exits 1 saying Nothing was fetched. The arithmetic is independently checkable from the printed values: longest_source_wall_ms 45000 is 9 waits times the 5000 ms source, sequential_wall_ms 81000 is 9 times 3000 plus 1000 plus 5000. 2026-09-06 queue THE NAS BUILD IS QUEUED, NOT SKIPPED. nx_build_admit read QUEUE on three different conjuncts across the session -- procs_blocked 12 of 8 as an I/O storm, then procs_running 14 of 8 as CPU saturation -- so the build was handed to /api/build, which absorbs a QUEUE verdict by launching nx_buildq rather than failing. The organ is source-landed and locally proven; when headroom appears the NAS artifact follows, and the RR1 and RR3 watch cells flip themselves on the next emit because the contracts name the real symbols sr_load and sr_pace. NOTHING HERE NEEDS A SEAT TO REMEMBER IT. 2026-09-06 measure THE READER CORRECTED ITS OWN BOARD ON ITS FIRST LIVE RUN. Every prose claim on this domain said 39 declared sources, hand-counted while the registry was written. nx_srcreach load knowledge/research_sources.conf reports rows=40 ok=40 bad_fields=0 bad_kind=0 bad_proto=0 bad_pace=0 sum=40 verdict=OK on 10177 bytes. THE ORGAN IS RIGHT AND THE PROSE WAS NOT, and that is precisely the argument for RR1 existing at all: a population figure that nobody re-measures is a claim with an expiry date nobody recorded. Corrected to 40 wherever it was published. 2026-09-06 land THE LIAR GATE, DONE PROPERLY, AND A PRECISION ERROR IN MY OWN CLAIM CORRECTED FIRST. I had described nx_filehash_gate as bite-proven when what it carries is an INTERNAL gv_bite tooth (fires-on-bad, silent-on-good). That is not the same proof as nx_gate_bite, the estate's MUTATION tester, and conflating them is exactly the kind of overclaim this board exists to refuse. nx_srcreach_gate was therefore built and put through the real thing: nx_gate_bite over runtime/nx_srcreg_lib.nx reports baseline green=1, a COMPILING mutant KILLED at cmp site 0, restore verified BYTE-IDENTICAL to pristine at 11347 bytes, twin RESTORED with mutant_live=0, valid_mutants=1 killed=1 not_reached=0, VERDICT GREEN -- non-vacuity PROVEN rather than asserted. The gate itself is 31/31 GREEN with every planted class caught by name, a positive control proving it is not a refuse-everything guard, both closed vocabularies proven to refuse an unknown token rather than default to the permissive value, and the budget asserted against hand-computed constants through gv_check_eq so the number tested and the number printed are one expression. 2026-09-06 lesson MUTATING THE ESTATE'S IDENTITY RULER WAS DELIBERATELY NOT DONE, AND THE REASON IS THE POINT. nx_filehash sits under every content pin in every refs register on this estate, and nx_gate_bite works by replacing the subject and rebuilding. The custody report from the run above shows the harness restores correctly -- twins_checked=1 restored=1 mutant_live=0 -- but one successful restore on a low-stakes subject is not evidence for a load-bearing one during an I/O storm, and a failed restore there breaks every pin check at once. So nx_filehash_gate stands on its internal bite tooth plus two published KATs it carries itself, and that limit is STATED rather than papered over. ★★★★★A PROOF YOU DECLINED TO RUN MUST BE NAMED AS DECLINED, NEVER QUIETLY OMITTED -- the difference between an honest gap and an overclaim is whether the reader is told. 2026-09-06 land THE BOARD IS NOW SOTA-CLASS, WHICH IN THIS ESTATE IS A MECHANISM AND NOT AN ADJECTIVE. researchreach.sota landed at 12326 bytes, making it the twelfth sota-class domain against eleven before. EVERY GRADE ROW CARRIES A MEASURED BINDING that the one symbol ruler re-checks on EVERY publish, and the rule that makes it worth having is that a Yes, Best or Part over an ABSENT symbol is the UNBACKED class and REFUSES TO PUBLISH. A claim on that page therefore cannot outlive the capability behind it, and no seat can grade this board green by writing prose. Eight bindings, every one verified present on disk before its row was written and every one verified IN THE ORGAN THAT NAMES IT rather than by a grep across the tree: cid_of, ss_commit, crg_prov_state, rd_same, fs_scan_source, sr_load, sr_pace, sr_probe. THE ABSENT CAPABILITIES ARE GRADED No AND BOUND TO NOTHING ON PURPOSE -- OAI-PMH, registry expansion, DOI dedup, full-text acquisition, layout parsing, hybrid retrieval, agentic compile -- because a No over an absent symbol is the only honest pairing and binding one to an aspirational name is precisely how a watch cell flips on a word instead of a capability. 2026-09-06 measure THE SOTA PAGE STATES THE BEHIND CELL IN NUMBERS RATHER THAN IN A FOOTNOTE. Coverage of the scholarly record reads 40 declared sources and ZERO harvested against roughly 250M, 300M, 200M and 180M records for OpenAlex, CORE, Semantic Scholar and OpenAIRE -- rival figures quoted as ORDER-OF-MAGNITUDE READS of their own published descriptions and explicitly NOT re-measured by us, so they are scale and not audited counts. The honest comparison is not 40 against 250 million: it is that we have declared a denominator and they have a corpus. Five rows are graded Best and each is a governance property no aggregator offers -- content-addressed identity, a tamper-evident store, fetch-time provenance, drift over time, and board-driven frontier noticing -- so the page claims a real lead on PROVENANCE while conceding COVERAGE outright. 2026-09-06 lesson THE APPEND ABOVE DOUBLE-APPLIED AFTER TWO 503s AND THE DEDUPLICATION IS THIS ROW. Both Outcome-Unknown calls had landed. I checked before re-issuing, using a PAYLOAD TOKEN rather than a size -- the method this estate prescribes precisely because size lags -- and the token count read 0 while the text was already there twice by the time the next read ran. SECOND OCCURRENCE TODAY, and it tightens the rule rather than restating it: FOR AN APPEND, ONLY THE WRITE'S OWN total= FIELD IS AUTHORITATIVE. Not a size, not a token count, not two spaced reads, and not any number of verbs on the reader side -- every one of them shares the lagging read path. The one instrument that does not is the writer's own receipt, and when a 503 destroys that receipt there is no sound pre-check available at all: the only safe shapes are an idempotent verb, a caller-supplied nonce the writer echoes, or an anchor-destroying edit whose re-issue self-adjudicates. 2026-09-15 land SOTA BOARD EXPANDED 2026-09-15 FROM FOUR RIVALS TO SEVEN (Crossref, Europe PMC, OpenCitations) so the field pass's own liar-kill bars hold: 8 columns, 20 rows, 3 quantitative rows, every Best evidenced. The two Best rows the pass had named UNEVIDENCED (content-addressed identity, append-only store) carried a symbol verification but no date or gate token; both now name the 2026-09-06 verification and the store's promoted gates. Seven refs added, every mirror pinned by sha256, every live count read from the reply and never from prose.

RR2 Reachability probe with the blocker named · cell LANDED · gate NO-GATE-DECLARED

Contract	`sr_probe` · executor Organ · 1 u · stands on RR1
Definition of done	Probes every registry row and emits a partition that SUMS to the row count over REACHABLE, RATE-LIMITED, TLS-FAIL, AUTH-REQUIRED, ROBOTS-DENIED, NOT-FOUND and UNPROBED. UNPROBED is its own state so a source nobody probed can never read as one that passed. Each non-reachable row carries its remedy. Gate is bite-proven in both directions and carries a NEGATIVE CONTROL that must stay REACHABLE, because a probe that refused everything would pass every absence test.
Matrix cell	Reachability measured per source with the blocker NAMED · `runtime/nx_srcreach.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_srcreach`; declare `nx_srcreach TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 measure TWO OF FOUR RIVAL DOCUMENTATION MIRRORS WERE NON-CONTENT PAGES RETURNING A CLEAN STATUS 200. graph.openaire.eu/docs/apis is an Apache directory index of 2813 bytes and api.semanticscholar.org/api-docs is a ReDoc JavaScript shell of 1752 bytes whose real specification loads at view time from graph/v1/swagger.json. Both saved as successful bodies. This is the stub class the estate has measured before, and it happened on the very board being built to measure reachability, which is the argument for RR2 stated better than any specification could: A HOST THAT ANSWERS IS NOT A HOST THAT DELIVERED WHAT YOU ASKED FOR, and a reachability probe that only checks status is a probe that would have called both of these fine. 2026-09-06 lesson THIS BOARD'S FIRST CITATION SET FAILED THIS BOARD'S OWN THESIS, AND THE ESTATE'S GATE NAMED IT. nx_compare_refs_gate reported OPEN-REDIRECT-RATCHET-ROSE-NEW OPEN-REDIRECT researchreach.refs openalex_api MOVED: the mirror was fetched with hops=1 and its canonical read CANON-HEAD-TRUNCATED, so it is NOT acquitted as a capture of the url the row cites. A register that only checks that a mirror EXISTS and that its pin matches would have passed it, because both were true. THE FETCH SUCCEEDED, THE PIN WAS CORRECT, AND THE CITATION WAS STILL NOT PROVEN TO BE OF THE PAGE CITED. This is the exact reason RR2's done-rule refuses to treat an HTTP 200 as reach: a host that answers is not a host that delivered what you asked for. 2026-09-06 lesson sr_probe WAS DELIBERATELY NOT DEFINED, AND THE ORGAN SAYS SO IN ITS OWN HEADER. Defining the symbol would flip the RR2 watch cell on the published board with no capability behind it, and a completion signal that keys on a name rewards writing the name. RR2 needs the network leg and stays an OPEN contract until it can actually probe. The two rungs that could be honestly finished were finished; the one that could not was left visibly unfinished. 2026-09-06 land RR2 SHIPPED AND THE FIRST REAL REACHABILITY CENSUS RAN. nx_srcreach PROMOTED 795d0912 at 50392 B, REGISTERED, CAPPED, LIVE. Live over all 40 declared sources, paced per host, every probe into ONE fixed scratch name: probed=40 reachable=19 rate_limited=0 auth_required=3 not_found=6 other_status=12 transport_fail=0 unprobed=0 sum=40 rows=40. THE PARTITION SUMS TO THE POPULATION and every non-reachable row printed its key and the remedy that would close it. 2026-09-06 retract DO NOT READ 19 OF 40 AS 21 BROKEN SOURCES -- TWO OF THE THREE THINGS THAT CENSUS MEASURED ARE DEFECTS IN OUR DATA AND OUR PROBE, NOT IN THE HOSTS. ONE: the registry's base-url field is a BASE and the probe treats it as a TARGET. api.github.com answers 200 bare while huggingface.co/api returns 404 because the real call needs a sub-path, so six NOT-FOUND rows and much of the OTHER-STATUS class are the probe asking a url nobody serves rather than a source we cannot reach. The registry owes a per-row PROBE PATH column and until it has one this census measures url shape as much as reach. TWO: codeberg came back OTHER-STATUS and transport_fail is ZERO across all forty, so the reported codeberg-fails-our-TLS class DID NOT REPRODUCE in this run; it is recorded as unconfirmed here rather than repeated. ★★★★★★A CENSUS THAT SUMS IS NOT THE SAME AS A CENSUS THAT MEANS WHAT ITS LABEL SAYS, and the first duty of a new instrument's first run is to say which of its findings are about the subject and which are about itself. 2026-09-06 lesson THREE DEFECTS IN THE ORGAN, FOUND BY RUNNING IT AND NAMED RATHER THAN LEFT FOR THE NEXT READER. ONE, the request ceiling doubles as the probe limit, so there is no way to say probe five of forty -- they must become separate arguments. TWO, the budget and the probe disagree on the wire: at one request per source the budget correctly prints wall 0 because a single request needs no inter-request wait, while sr_probe sleeps pace_ms after EVERY request, so the model and the behaviour describe different things. THREE, the per-row lines print only after the whole sweep returns, so an interrupted census yields nothing -- the all-or-nothing shape this estate has already paid for elsewhere. Each is a rung, not a rewrite. 2026-09-06 land ALL THREE SELF-REPORTED DEFECTS FIXED AND VERIFIED LIVE, PLUS THE PROBE-PATH FIELD THE FIRST CENSUS ASKED FOR. nx_srcreach PROMOTED 0f3b4759 at 50872 B. ONE, the probe limit is now its own argument: probe conf 1 100 5 reports probed=5 max_probe=5 against ceiling=100, which was impossible when one number served both questions. TWO, the budget models the behaviour that exists rather than the one a formula suggests: reqs x pace, not (reqs-1) x pace, because sr_probe sleeps after EVERY request including the last to a host -- at one request per source the old model printed wall 0 while the run slept 140000 ms, and it now prints longest_source_wall_ms=10000 sequential_wall_ms=140000, which is the cost that will actually be paid. THREE, every row prints AS IT IS DECIDED rather than after the sweep returns, so an interrupted census keeps what it found. FOUR, the optional tenth field probe_path is honoured, and it is PROVEN by outcome and not by inspection: huggingface read NOT-FOUND against its bare base, and with /models?limit=1 the same probe reads REACHABLE -- probed=2 reachable=2 not_found=0. THE DIAGNOSIS THAT THE BASE-URL WAS THE DEFECT RATHER THAN THE SOURCE IS THEREFORE CONFIRMED BY EXPERIMENT, not asserted. The partition still sums to the population in every run. 2026-09-15 measure LIVE COUNTS READ FROM THE SOURCES' OWN APIs THROUGH THE SOVEREIGN FETCHER WITH NO KEY: OpenAIRE numFound 389982852 (2233 B), Crossref total-results 186608495 (2472 B), Europe PMC hitCount 5596567 for one term (1156 B), OpenCitations citation-count 200 (37 B), all 2026-09-15; OpenAlex meta.count 325977999 from the 2026-09-06 mirror, which corrects the ~250M cell this board had quoted from prose. CORE and Semantic Scholar stay order-of-magnitude reads and the page says so. 2026-09-15 lesson TWO CANDIDATE COLUMNS WERE STUBS WITH A CLEAN 200 AND THE FETCH-TIME READ IS WHAT REFUSED THEM: BASE served an Anubis proof-of-work challenge (7843 B) and Unpaywall a 1421 B Vue shell. A column graded from either would have been eight cells of fiction under a perfectly valid pin. Neither is cited; the mirrors stay on disk as the record of the refusal. 2026-10-01 queue ATTEST VERDICT ACQUITS ON FULL ABSTENTION, measured 2026-10-01 and confirmed in source: the ATTEST-DONE summary in buildroot/runtime/nx_frontier_scan.nx sets the verdict from stale_bars alone (verdict=ATTESTED, exit 0, whenever stale_bars is 0), so the twiningest attest that morning read ATTESTED with appended 0 and unattestable 11 of 11 bars, every arXiv own-ref leg ATTEST-UNATTESTABLE reason=source-unreachable. nx_barfresh was not fooled (it reads barscan rows and the board stayed UNATTESTED until a seat read the listings), but any caller that branches on the organ's exit or verdict line is. Remedy, not built in this lane: a third state, verdict=UNOBSERVABLE with exit 3 when appended plus already is 0 and unattestable is above 0, PARTIAL when only some bars are unattestable, and a planted all-unreachable fixture as the gate tooth. Same morning, the reach fact behind it: export.arxiv.org answered HTTP 429 to the NAS on every attempt for hours and arxiv.org search dropped the connection intermittently, while the OpenAlex works API (filter primary_location.source.id S4306400194, the arXiv source, with title_and_abstract.search and sort by publication_date descending) answered every request; it indexes arXiv about a week behind, enough for a month-granular bar, and it read every arXiv bar this seat attested on aesthetictwin and twiningest (mirrors rf_0fd48a64, rf_1c19978b, rf_0423ce9b and nine more under knowledge/fetched/). Candidate: OpenAlex as a second key-free dated source row for arXiv-hosted bars, announced as borrowed, with its indexing lag printed beside the newest month.

RR3 Per-source pace budget printed and obeyed · cell LANDED · gate NO-GATE-DECLARED

Contract	`sr_pace` · executor Organ · 1 u · stands on RR1
Definition of done	The run PRINTS sum of planned requests before issuing the first one and REFUSES to start when that exceeds the declared ceiling. Budget is per host, not global, because a shared budget starves the slow hosts first. Gate proves the printed count equals the requests actually issued, and that a ceiling breach refuses rather than truncates.
Matrix cell	Per-source pace budget printed and obeyed · `runtime/nx_srcreach.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_srcreach`; declare `nx_srcreach TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 measure THE CADENCE QUESTION NOW HAS A NUMBER, COMPUTED BEFORE ANYTHING IS FETCHED. Live on the real registry: 40 sources at 3 requests each is 120 planned_requests, longest_source_wall_ms 20000 set by the one host measured to throttle us, sequential_wall_ms 280000. Against a ceiling of 100 it printed REFUSED-OVER-CEILING and exited 1 with Nothing was fetched. The arithmetic is checkable from the printed values alone: the declared paces sum to 140000 ms and two waits per source is 280000, and 2 x the 10000 ms semanticscholar pace is the 20000 floor. THAT IS THE ANSWER TO not smashing our nas resources -- the cost is stated and refused in advance rather than discovered from the array afterwards.

RR4 OAI-PMH harvest with resumption · cell WATCHING · gate NO-GATE-DECLARED

Contract	`oai_list_records` · executor Organ · 1.5 u · stands on RR1, RR3
Definition of done	ListRecords and ListIdentifiers over the six verbs with resumption tokens followed to exhaustion and a from-until selective window. Gate proves a multi-page response is followed to its end and that a run stopped early reports PARTIAL rather than success, because a harvester that stops at page one acquires a prefix and reports a whole.
Matrix cell	OAI-PMH incremental harvest · `runtime/nx_oaipmh_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_oaipmh_lib`; declare `nx_oaipmh_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

RR5 Registry-of-registries expansion · cell WATCHING · gate NO-GATE-DECLARED

Contract	`oai_registry_expand` · executor Organ · 1 u · stands on RR4
Definition of done	Walks OpenDOAR and ROAR, admits each discovered endpoint as a derived source under the default pace, and writes it to knowledge/status/oai_endpoints.derived WITH the registry that produced it. Gate proves a derived endpoint is never written back into the hand-authored registry and that removing a registry removes exactly its own children.
Matrix cell	Registry-of-registries expansion to the institutional tier · `runtime/nx_oai_registry.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_oai_registry`; declare `nx_oai_registry TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

RR6 Staggered fleet cadence · cell WATCHING · gate NO-GATE-DECLARED

Contract	`rh_beat` · executor Organ · 1.5 u · stands on RR3, RR2
Definition of done	Every /compare domain revalidated against the record on a beat, STAGGERED across domains so outbound load is flat rather than a daily spike, under RR3's budget. Done when a full pass states its own request count, that count matches what was issued, and the beat writes a truncate-written verdict log whose LAST line is canonical. THIS RUNG IS THE ONE THAT CLOSES deepresearch DN1, whose two blocking conditions are RR3 and a second key-free index, both supplied here.
Matrix cell	Fleet cadence: every compare domain revalidated against the record on a beat · `runtime/nx_research_harvest.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_research_harvest`; declare `nx_research_harvest TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 queue THE CADENCE RUNG IS NOT NEW WORK, IT IS A BLOCKED SIBLING GETTING ITS BLOCKERS. deepresearch DN1 fs_sweep_all has been open since 2026-08-20 with a done-rule naming exactly two preconditions: a second key-free scholarly index admitted as a source row, and a per-source pace budget the run prints and obeys. openalex is now a registry row and RR3 is the budget. When RR6 lands, DN1's done-rule is satisfiable and the two rungs should be reconciled rather than both left open, because one capability carried as two open contracts on two boards is how a backlog grows without work being missed.

RR7 Cross-source dedup on a stable identifier · cell WATCHING · gate NO-GATE-DECLARED

Contract	`xr_doi_join` · executor Organ · 1 u · stands on RR4
Definition of done	The same work harvested from a preprint server, an institutional repository and a publisher collapses to one record joined on DOI, with the alternatives kept as sources rather than discarded. Gate proves a planted three-way duplicate collapses to one and that a record with no DOI is bucketed as UNJOINABLE rather than dropped.
Matrix cell	Cross-source deduplication on a stable identifier · `runtime/nx_xref_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_xref_lib`; declare `nx_xref_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

RR8 Open-access full text acquired · cell WATCHING · gate NO-GATE-DECLARED

Contract	`oa_fulltext` · executor Organ · 1.5 u · stands on RR7
Definition of done	A DOI resolves to its legal open-access copy and the document is acquired, content-addressed and stored. Gate proves a closed-access DOI yields a NAMED refusal and never a fabricated body, because a pipeline that invents a full text is worse than one that admits it has none.
Matrix cell	Open-access full text actually acquired, not merely indexed · `runtime/nx_oafull_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_oafull_lib`; declare `nx_oafull_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-23 queue INTAKE 1790173548 routed by nx_intake_route (nishi-review-via-codex, epoch 1790173579) from operator-via-codex: Build a sovereign legal-open-access research resolver, callable by MCP and API. Given DOI, title or URL, discover Unpaywall, author, institutional and publisher OA versions without bypassing controls; -- sponsor intake-1790173548 plus2 2026-09-23 queue INTAKE 1790173820 routed by nx_intake_route (nishi-review-via-codex, epoch 1790173910) from operator-via-codex: Create a Nishi sovereign research-publication and corpus pipeline. It must preserve original bytes and provenance, validate DOI or arXiv identity/title/authors/version/date/license, retain reading ord -- sponsor intake-1790173820 plus2

RR9 Layout-aware parsing · cell WATCHING · gate NO-GATE-DECLARED

Contract	`lp_blocks` · executor Organ · 3 u · stands on RR8
Definition of done	A PDF becomes a typed block tree in which a table round-trips to its cells and an equation survives as its own region, rather than the flattened text stream nx_pdf_text produces today. Gate carries an ANTI-VACUITY tooth: a plain text extractor must FAIL it, because a checker that only asks whether text came out is passed by the tool we already have.
Matrix cell	Layout-aware parsing: a PDF becomes structured text with its maths and tables intact · `runtime/nx_layout_parse.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_layout_parse`; declare `nx_layout_parse TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-28 measure MEASURED 2026-09-28 ON A MODERN ARXIV PDF: nx_pdf_text over Large-scale Codec Avatars (2604.02320v2, 32064633 B, knowledge/fetched/cmp_twin_lca_2604_02320v2.pdf) wrote about 66 KB of body text with kerning-split words (multi-vie wstudio, achie vehigh) and ligatures lost as replacement bytes, then about 11.3 MB of binary stream content: nx_txtscan ctl read ctl_runs 961750 ctl_bytes 2018613 verdict CORRUPT. The organ's own LIMITS line names the causes (no object streams, no xref streams, no glyph-position layout). RR23 cannot land on a reader that cannot read the paper, and the twiningest board's studio camera counts stay UNREAD for the same reason. 2026-09-29 measure THE OPEN FORM BEAT THE PDF ON THE SAME PAPER, 2026-09-28: Large-scale Codec Avatars (2604.02320) read through nx_pdf_text gave about 66 KB of text with split words and lost ligatures followed by about 11.3 MB of undecoded stream bytes (verdict CORRUPT); the same paper's arXiv HTML read through nx_dr_extract gave 73 KB of clean text and recovered the capture facts the PDF route lost (200 calibrated synchronised 4K cameras, about 5,000 frames per subject, 2,737 studio participants). The HTML carries each equation twice (the rendering and its TeX annotation), which RR25 must collapse to one copy. RR9 stays the fallback for papers with no open form.

RR10 Code graphing for acquired repositories · cell WATCHING · gate NO-GATE-DECLARED

Contract	`cg_ast_edges` · executor Organ · 2 u · stands on RR2
Definition of done	An acquired repository becomes an AST-level call graph, generalising what nx_funcmine and nx_capgraph_edges already do for our own tree. Arm rather than contender: it triangulates the record against its implementations and does not compete for the coverage number.
Matrix cell	Code graphing: a repository becomes an AST-level call graph · `runtime/nx_codegraph.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_codegraph`; declare `nx_codegraph TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

RR11 Hybrid graph and vector retrieval · cell WATCHING · gate NO-GATE-DECLARED

Contract	`kgh_query` · executor Organ · 3 u · stands on RR7, RR9
Definition of done	One query answered against a knowledge graph and a vector index together, with the graph carrying the strict relations and the vectors the semantic reach. Accept rule DECLARED BEFORE THE RUN: must EXCEED the lexical arm on the same held-out questions on a pre-declared ruler, or it stays UNWIRED and the number is published anyway.
Matrix cell	Hybrid retrieval: a knowledge graph and a vector index answering one query · `runtime/nx_kg_hybrid.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_kg_hybrid`; declare `nx_kg_hybrid TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

RR12 Gated agentic compile · cell WATCHING · gate NO-GATE-DECLARED

Contract	`ag_stage` · executor Organ · 3 u · stands on RR11
Definition of done	Staged gather, write and criticise, where the writing stage CANNOT commit to the store until the critic has graded it. Gate proves a planted false claim never lands, and the critic is the one the estate already owns rather than a second one built beside it.
Matrix cell	Agentic multi-stage compile with a critic that can refuse · `runtime/nx_agent_compile.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_agent_compile`; declare `nx_agent_compile TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

RR13 The drift comparator gets its contracted entry point · cell LANDED · gate NO-GATE-DECLARED

Contract	`rd_same` · executor Organ · 0.5 u · stands on nothing
Definition of done	CORRECTION RUNG, not a new capability. nx_refdrift's four-state comparator is real and runs inline in main, so the deepresearch row claiming symbol rd_same publishes as UNGROUNDED while the capability it names genuinely works. Extract the comparator to the contracted name, prove neutrality by nx_behaveprobe against the live binary, and the sibling row grounds itself. Done-rule: nx_domain_admit deepresearch reports ungrounded 0.
Matrix cell	Citation still served: does the source still serve the bytes we pinned · `runtime/nx_refdrift.nx` · LANDED decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_refdrift`; declare `nx_refdrift TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-06 measure A PUBLISHED EXCEED CLAIM WAS FOUND UNGROUNDED, AND THE COINCIDENCE THAT NEARLY HID IT IS THE FINDING. nx_domain_admit reported UNGROUNDED symbol rd_same not word-boundary-present in runtime/nx_refdrift.nx. A grep over buildroot/runtime with corpus_complete=1 found func rd_same in exactly one file: nx_rebuild_drain.nx, an unrelated organ sharing the rd_ prefix. A verification that greps for the symbol name alone would have CONFIRMED it and closed the finding. Only the complete outline of nx_refdrift.nx settled it: 64 declarations, functions rd_slurp rd_put2 rd_ymd rd_basename rd_sha rd_recite_row and main, and no comparator among them. THE CAPABILITY IS REAL AND RUNS INLINE IN MAIN; ITS CONTRACTED ENTRY POINT WAS NEVER WRITTEN. Verify a symbol IN THE ORGAN THAT NAMES IT, never in the tree. 2026-09-06 land RR13 CLOSED: rd_same EXISTS AND IS LIVE (nx_refdrift promoted 4a87c99c, 65360 B). The discovery that made it correct is that the byte comparator was never missing -- rp_same in nx_rowparse_lib already was it, so writing a second would have been the duplicate-ruler defect. What had no name was the FOUR-STATE DECISION wrapping it. Extracted without widening: the caller still performs both fetches and both slurps under exactly its prior guards. NEUTRALITY PROVEN RATHER THAN ASSERTED -- nx_contentdiff live vs candidate reports runs_scanned=260 lost_from_live=0 lost_permil=0 symmetry=ONE-WAY-GAIN verdict=GREEN, a strict superset. nx_domain_admit now reports ungrounded=0 on deepresearch, and that board's row names the real symbol again with its exceed left at 0 for the owning lane to re-assert. ★★★★★THE MISSING ENTRY POINT WAS NOT THE MISSING CAPABILITY: when a contracted symbol is absent, ask whether the primitive already exists under another name before writing a second one.

RR14 Port the legacy source registry and OAI-PMH adapter to NishiLang · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW

Contract	`sr_port_parity` · executor Organ · 3 u · stands on RR4, RR5
Definition of done	THE ESTATE ALREADY SOLVED MOST OF THIS ONCE, IN A STACK THE SOVEREIGNTY ORDER FORBIDS IN THE BUILD PATH. nishi-library v0.0.102 carries a fifty-plus source registry across nine tiers and a GENERIC OAI-PMH adapter with resumptionToken cursors already covering DOAB, OAPEN, AAAI, AIIDE and NSF-PAR, in Python over SQLite. Done-rule is PARITY MEASURED AND NOT ASSERTED: the sovereign harvester reaches the same declared sources and returns the same record count for one named tier, proven side by side, before the legacy path is retired. Retire by nx_retire_path so it is reversible, never by deletion, and never before parity is shown -- a legacy stack that works is not litter.
Matrix cell	NO-MATRIX-ROW — no matrix row carries `sr_port_parity`; 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	2026-09-06 retract A CLAIM IN THIS PLAN'S OWN FIRST DRAFT WAS FALSE ABOUT OUR OWN ESTATE, AND ONE nx_memfind CALL REFUTED IT. The pos line asserted there is no declared source registry and no harvest protocol. nishi-library reached v0.0.102 with a fifty-plus source registry across nine tiers, a generic OAI-PMH adapter with resumptionToken cursors, DV2.0 hub-satellite-link with FTS5, 2.2 million works and 1.87 million citation edges, LIVE since 2026-04-29. THE DISCRIMINATING QUESTION WAS NOT WHETHER THE CAPABILITY EXISTS BUT WHICH STACK IT EXISTS IN, and asking the memory plane BEFORE writing the claim costs one call while publishing it costs the next reader the whole investigation. The capability search this estate mandates covers ORGANS; it does not cover retired-stack capability, and nx_memfind is the instrument that does.

RR15 A research paper retains complete human and machine evidence · cell NO-MATRIX-ROW · gate NO-MATRIX-ROW

Contract	`_ABSENT_:paper_dual_evidence_accept` · executor Workflow · unestimated u · stands on RR8, RR9
Definition of done	SOURCE-BOUND PLANNING CONTRACT 2026-09-24. Extend existing fetch, library, layout parser and document emitter owners. Store exact source identity and access/rights state beside typed text, tables, equations, figures, captions and reference links with source-region correspondence. Produce readable human rendering and machine records from the same retained evidence, with unsupported regions and uncertainty explicit. Verify a representative multicolumn paper including a table, equation and figure; round-trip table cells and references, preserve original bytes, and show a flattened-text negative control cannot claim complete extraction. Incremental versions retain predecessors and changed claims; a failed fetch never becomes an empty successful paper. Comparisons with reference parsers are independent measured work, not supplied-name superiority. Effort unestimated.
Matrix cell	NO-MATRIX-ROW — no matrix row carries `paper_dual_evidence_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	no log row names this rung yet

RR16 Venue-tag scans as frontier source rows · cell WATCHING · gate NO-GATE-DECLARED

Contract	`_ABSENT_:fs_venue_rows` · executor Organ · 1 u · stands on RR1
Definition of done	The arXiv API indexes the comment field, where authors write the venue: co:SIGGRAPH 2026, co:SIGGRAPH Asia 2026, co:Eurographics 2026, co:CVPR 2026, co:ECCV 2026, co:NeurIPS 2026 and co:IEEE VR 2026 become DATED source rows in knowledge/frontier_sources.conf with the seed ANDed in, so nx_frontier_scan attest asks every venue what the bar's subject says this month. Landed as rows 2026-09-27, each proven reachable the same day; the rung closes when attest prints a per-venue NEWER, SAME or OLDER line on a live board. Done-rule: one attest run on graphics prints the venue rows beside the arXiv and HN rows.
Matrix cell	Venue tags as frontier source rows · `runtime/nx_frontier_scan.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_frontier_scan`; declare `nx_frontier_scan TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-27 land Venue-tag rows and static venue rows landed in knowledge/frontier_sources.conf 2026-09-27 after each url was fetched through nx_research_fetch with status 200 and a provenance row; the attest run that consumes them is the next leg. 2026-09-27 measure FIRST ATTEST OVER GRAPHICS WITH THE VENUE ROWS, 2026-09-27 (run artifact /tmp/fs_attest_graphics.txt, 109 lines at the time of this row): the venue-tag rows answered for the first bars (co:SIGGRAPH 2026 returned an item for procedural generation, the Eurographics 2026 static index 4 items dated 2026-05, the SIGGRAPH Asia 2026 static index 1 item) and then 30 of the first 40 venue-tag queries returned UNREACHABLE, printed as UNOBSERVABLE and never as zero. Seven venue rows times seven bars is 49 arXiv requests in one run with no per-host pace between them, which is arXiv rate limiting honestly reported, not a dead source. 2026-09-27 queue The per-source pace budget (RR3) must govern nx_frontier_scan's fan-out per HOST across all rows (one arXiv request per pace interval, whatever row it serves), or the venue rows coalesce into ONE OR-query per seed; until one of those lands, attest reads the venue rows as UNOBSERVABLE after the first bars and no bar is misdated by it. 2026-09-27 land THE FAN-OUT IS PACED (2026-09-27, same session as the venue rows): nx_frontier_scan gained fs_pace_wait at its ONE fetch site, fs_pace_for (the pace for a url's host read from knowledge/research_sources.conf field pace_ms, the registry RR3 already governs; the largest declared pace for a host the registry does not name; PACE-UNKNOWN announced and no wait when the registry is unreadable), fs_url_host and fs_num_at, with six static module slots (90504 -> 96160 bytes). Sovereign build 126144 bytes sha256 be8a1fb3ba6ce17a050ac07708f3806777a34a57bd94d3ea9182da2aec51ace9; nx_contentdiff lost ONE run, ct_pick, an unreferenced symbol name the organ never calls (0 occurrences in its source), gained the registry path and the newer bar-ruler refusal strings the live binary predated; nx_behaveprobe seeds graphics IDENTICAL (8852 bytes both); PROMOTED with prev banked and _offc refreshed. First measured wait in the paced attest: PACE host=export.arxiv.org pace_ms=3000 elapsed_ms=303 waited_ms=2697, and the venue row that read UNREACHABLE unpaced now answers (co:SIGGRAPH Asia 2026 for the GI bar: Gaussian Light Transport 2609.11430, NEWER than the 2026-07 bar). Scope stated: consecutive requests to one host are paced, which is the measured burst shape; a per-host table for interleaved bursts is the next rung only if one is measured. 2026-09-27 measure LIKE FOR LIKE, SAME BOARD, SAME BARS, SAME ROWS, SAME AFTERNOON: the unpaced attest over graphics read 22 of its first 33 venue-tag queries UNREACHABLE (30 of the first 40); the paced attest read 0 of its first 33 UNREACHABLE, every wait announced (PACE host=export.arxiv.org pace_ms=3000, waits of 2486 to 2704 ms after 296 to 514 ms gaps), and the rows that had read UNOBSERVABLE now carry items: co:SIGGRAPH Asia 2026 for the GI bar (Gaussian Light Transport 2609.11430, 2026-09, NEWER than the 2026-07 bar), co:SIGGRAPH 2026 for real-time rendering (4 items, newest 2026-08), co:CVPR 2026 (8 items, newest 2026-07), co:ECCV 2026 (8 items, newest 2026-08). The number the fix is judged by is the unreachable count at equal query count, 22 to 0; wall time rose by the sum of the announced waits, which is the price the registry's own pace declares. 2026-09-27 measure paced attest run complete 2026-09-27 (nx_frontier_scan attest graphics, promoted be8a1fb3): bars=7 fetches=112 venue-tag fan-out rows=49 verdict=UNREACHABLE rows=0 newer_sources=8 stale_bars=1; every consecutive export.arxiv.org request printed PACE pace_ms=3000 waited_ms 2575 to 2718; the matched unpaced run read 30 of its first 40 venue-tag rows UNREACHABLE. The paced reach surfaced graphics SB-RTGI STALE (Gaussian Light Transport 2609.11430 under the SIGGRAPH Asia 2026 tag) that the throttled run could not raise 2026-09-27 lesson an UNREACHABLE fan-out row is an abstention that acquits: nx_barfresh cannot read STALE from a listing the scanner never received, so a throttled researcher keeps every bar looking fresh. Pacing is therefore a freshness instrument, not a courtesy to the host; count UNREACHABLE rows beside stale_bars on every attest 2026-09-27 land the fleet refs referee nx_compare_refs_gate names its offenders now (promoted 2026-09-27, 165376 B sha256 3e93bce3, prev banked 140080 B sha256 242e51a8): dupkey= and missing= were counters with no worklist row in crg_scan, so a duplicate key (softbody arxiv_softbody_202609, this lane's own inherited row) was found only because its UNCITED sibling named it; the scan prints DUP-KEY file= key= and MIRROR-MISSING file= key= mirror= beside the counts. Content ruler: lost 1 (sys_openat_append, the base-class symbol the matched-vintage control nx_vasc_gate lost through the same lane today), gained the fio and framed-append closures the served binary predated; promoted under allow_capability_loss with that run named. Acceptance is the next fleet run printing the one missing mirror by register and key 2026-09-27 lesson a referee that counts a defect class without naming a row makes every fix a search; the two counters in the estate's own citation referee had been printed for weeks with nobody able to act on them. When a counter is added, add its row in the same edit 2026-09-27 measure acceptance of the promoted refs referee, fleet run 2026-09-27 (38/44, verdict unchanged, offenders now named): MIRROR-MISSING file=buildroot/knowledge/compare/motion-rigmodels-20260919.refs key=rigmodels_viewer_20260919 mirror=knowledge/compare/motion-rigmodels-acceptance-20260919.json (a 09-19 motion row citing a mirror outside knowledge/fetched -- for that row's owner); DUP-KEY fires on no live register (dupkey=0) and both rows fire on the gate's own scan-defects fixture bad.refs key=gamma, so the bite came free from the fixture the gate already planted 2026-09-28 land HOST-BORROWED GRAMMAR 2026-09-28: a bar whose ref url is a query the source templates do not spell (abs: instead of all:, a parenthesised OR, a venue tag) read ATTEST-UNATTESTABLE no-dated-source-row-knows-this-listing although its host is a dated source whose item and date grammar are exactly what that listing answers in; nx_frontier_scan now lends the first host-matching row's grammar when no template prefix matches and announces ATTEST-GRAMMAR borrowed-by-host (four anchored CAS edits; build 134968 B sha256 a691fd96; contentdiff one-way gain of the two announcement strings; behaveprobe seeds IDENTICAL; promoted; gate 32/32; _offc refreshed). Live: skintwin 7 bars, 5 attested by the organ (the NIST and MetaHuman html refs correctly not), barscans appended with on-subject months; SB-FABRIC re-dated to 2026-03 and SB-WET to 2025-04 because the classifier now excludes the June cond-mat yarn paper and the December quantum-gravity homonym the 09-27 hand rows had carried 2026-09-28 lesson an attester that matches a citation by template PREFIX can only attest the queries it was taught to spell; the grammar a listing answers in is a property of its HOST, so match the host and lend the grammar, and say so on the row. Until this landed every custom-query arXiv bar in the estate would have read UNATTESTED on the first of October 2026-09-28 land the citation referee no longer convicts a binary mirror on an HTML axis (2026-09-28): a redirected PDF or ZIP-based office mirror read CANON-HEAD-TRUNCATED (no head closes in a PDF) and joined the open-redirect ratchet forever; the Jensen 1999 mirror raised that ratchet 20 -> 21 the hour it landed. nx_compare_refs_gate now recognises the PDF and ZIP magics in crg_is_binary_doc, crg_canon_of_mirror returns a fourth negative class, the row is censused as CANON-NOT-APPLICABLE-BINARY in its own partition bucket (pv[21], summed into the redirected partition, never open), the summary prints not_applicable_binary=, and a neg-control fixture assembled byte by byte proves the class fires (promoted 167336 B sha256 c14ed0da, contentdiff one-way gain, prev 3e93bce3 banked). Its pin is the identity proof a canonical would have been 2026-09-28 measure publisher blocked 2026-09-28: the scoped regen refused every domain with FORK-STALE (buildroot/_offc/nx_swcompare_matrix.elf af966543 1283352 B against the promoted 603de9cd 1341712 B shipped by another seat at 1790603620 per organ_ship.jrnl); the fork was refreshed by copying the promoted serving-root binary (nx_filecopy, 1341712 B verified byte-for-byte), the documented remedy for a buildroot/_offc generator left behind by a promote. The ship loop's OFFC-TWIN stage refreshes nishihost/_offc, not buildroot/_offc: the third root the publisher forks, still owed a stage of its own 2026-09-28 measure fleet refs referee on the promoted binary 2026-09-28 (167376 B sha256 fddf4815): 39/45, up from 38/44 before the binary class; the Jensen 1999 mirror reads CANON-NOT-APPLICABLE-BINARY, not_applicable_binary=3 (the swarm verma2015.pdf and the automation osworld results .xlsx were the same misreading), canon_unknown 19 -> 16, the open-redirect set 21 -> 18 with both identity teeth (open equals moved plus canon-unknown; worklist rows equal the open count) PASSING again after the first ship had broken them by excluding the class from the worklist but not from the count (the OPEN count lived in a second expression, redirected minus benign). Residue, all other owners: 1 missing mirror (motion-rigmodels 09-19), 6 pins on beat-rewritten library files, 45 uncited keys in nine domains, 2 provenance mismatches (datavis, loadgov), the open-redirect ratchet at 18 against a baseline of 14, the moved ratchet without an attribution snapshot 2026-09-28 land THE THIRD ROOT HAS ITS STAGE (2026-09-28): nx_offc_install <target> promoted now also refreshes buildroot/_offc/<target>.elf from the PROMOTED root binary whenever that mirror already exists (oi_refresh_buildroot, the one atomic install oi_install_from_to parameterised by destination; a mirror is maintained, never created), and announces buildroot_offc=REFRESHED or ABSENT or FAILED on the same line; the ship loop's OFFC-TWIN stage forks this installer, so every future generator promote refreshes the publisher's fork without a hand copy. Promoted 37616 B sha256 c53a1539 under allow_capability_loss for the one rewritten success sentence (the served binary carried no source paths; the candidate gains them); witnessed live: buildroot_offc=ABSENT on nx_kubelka_gate, REFRESHED on nx_swcompare_matrix; the FAILED state is named, not witnessed 2026-09-28 land THE CITATION REFEREE'S WORKLIST HAS REPEATABLE REMEDIES (2026-09-28, operator: repeatable capabilities, never hand edits): nx_compare_cite grew mark (UNCITED, label prefix or rung:ID or auto from the grounds field), markmany (a whole board's uncited keys in one call, all-or-none, one publish per touched file), labels, repin (PIN-MISMATCH: the row's own url re-fetched through the researcher into an immutable knowledge/fetched mirror, a redirect refused), reprov (PROV-MISMATCH: the url the fetch journal holds), canon and recanon (OPEN-REDIRECT: re-fetched at the declared canonical with zero hops); one row matcher shared with the legacy path; gate nx_compare_cite_gate 11 to 24 teeth with planted refusals, promoted 3e870fc3 / 9f5878ab; the gate caught the first markmany parser on its live run (a loop-exit sentinel in the cursor, the estate's own law) and the fix shipped inside the hour. Live on the fleet referee's rows: 2 repins, 2 reprovs, 1 recanon, 45 UNCITED marks across 9 boards, the motion-competitor side register merged into motion.refs and retired 2026-09-29 measure REACH GAP 2026-09-28: https://civitai.red/ answered the researcher 403 and nothing was saved (RF-DEFINITIVE, not retried); the sibling hubs answered 200 and are pinned on twiningest (the civitai.com public API listing and the civitaiarchive.com index). A 403 from a hub the operator named is a wall to classify (login, region or bot challenge), never evidence that the site is absent. 2026-10-01 queue BARFRESH 2026-10-01 (seat residue-fleet, T0 1790856940): STALE before and after, 4 of 4 bars UNATTESTED; this seat wrote NO barscan here and names two ABSTENTIONS. RB1 (openalex_api): help.openalex.org re-fetched over the sovereign fetcher (mirror knowledge/fetched/cmp_researchreach_openalex_help_202610.html, 20201 B, sha256 b542cdca68571d1a, changed since the 6 Sep pin) carries no dated mark at all (zero occurrences of the substring 202), so a month cannot be attested from the bar's own ref; the bar needs a dated ref (a release or changelog listing) before any barscan is honest. SB-VENUEREACH-2609 (kesen_sig2026): the SIGGRAPH 2026 index re-fetched byte-identical (288392 B, sha256 7e6c166c5ae0fca3), a static July 2026 listing whose newest month is 2026-07 while the bar is dated 2026-09 by its measurement day; attesting newest=2026-07 would not match the bar, and re-measuring its eight sources needs arXiv fetches this seat does not make. RB2 (agentkg2026) and RB3 (foxden2026) are arxiv.org abstract pages, left for the arXiv attest runner.

RR17 Static venue indices and the Eurographics digital library · cell WATCHING · gate NO-GATE-DECLARED

Contract	`_ABSENT_:fs_static_venue_rows` · executor Organ · 1 u · stands on RR1, RR4
Definition of done	Ke-Sen Huang's per-venue indices (SIGGRAPH Asia 2026, Eurographics 2026, EGSR, HPG, I3D, SCA) join the SIGGRAPH 2026 row as static month-tagged sources, and the Eurographics Digital Library's OAI-PMH endpoint joins research_sources.conf as a harvestable publisher; the extractor is the existing generic dt-B grammar. Done-rule: attest reads each index and the SIGGRAPH Asia row grows from 7 to the proceedings count by December.
Matrix cell	Static venue indices and the Eurographics digital library · `runtime/nx_frontier_scan.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_frontier_scan`; declare `nx_frontier_scan TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-27 lesson A STATIC INDEX ROW DATED TO ITS PROCEEDINGS MONTH READS NEWER THAN EVERY CURRENT BAR: the siga2026 row was admitted with month=2026-12 (the December proceedings) while its items are September's conditional acceptances, so attest printed NEWER 2026-12 against September bars. Re-dated to 2026-09, the month the list is served; a second field for the proceedings month is a grammar the conf does not yet carry.

RR18 Venue completeness by DBLP · cell WATCHING · gate NO-GATE-DECLARED

Contract	`_ABSENT_:sr_dblp_population` · executor Organ · 2 u · stands on RR7
Definition of done	A venue index read at a cap is a PREFIX; DBLP's publication API lists every paper of a venue and year and is the population count a mirrored index is reconciled against, so coverage is a fraction with a denominator rather than a count. Done-rule: the researcher prints mirrored over DBLP-listed per venue with the partition summing.
Matrix cell	Venue completeness by DBLP · `runtime/nx_srcreach.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_srcreach`; declare `nx_srcreach TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

RR19 Labs, standards bodies and governments on the registry · cell WATCHING · gate NO-GATE-DECLARED

Contract	`_ABSENT_:sr_lab_rows` · executor Organ · 1.5 u · stands on RR1, RR2
Definition of done	NVIDIA Research, Meta Reality Labs, Disney Research, MPI Perceiving Systems, Google, Microsoft, Adobe, Epic and Apple publication records; Khronos and MPEG registries; NIST, DARPA, CORDIS and NIH RePORTER programme records: each a research_sources.conf row under a declared kind and pace, reached on the fleet cadence with the srcreach partition naming any blocker. Landed as rows 2026-09-27 under the publisher kind because sr_load refuses an unknown kind. Done-rule: nx_srcreach probe reads every row REACHABLE or names its blocker, and a lab kind is admitted to the registry reader rather than borrowing publisher.
Matrix cell	Labs, standards bodies and governments on the registry · `runtime/nx_srcreach.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_srcreach`; declare `nx_srcreach TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-27 measure nx_srcreach load before the landing: rows=40 ok=40 verdict=OK; after the landing the same verb is the gate that the new rows parse.

RR20 Login walls and script shells named as blockers with the sovereign browser as the door · cell WATCHING · gate NO-GATE-DECLARED

Contract	`_ABSENT_:sr_shell_blocker` · executor Organ · 2 u · stands on RR2
Definition of done	Measured 2026-09-27: the MetaHuman news site redirects to an Epic login (a login wall) while the developer documentation site serves the same release notes as 259454 bytes of HTML; earlier the MetaHuman encyclopedia page returned status-0 fragments and the Unreal release-notes page an Angular shell. A vendor row therefore carries an ALTERNATE WITNESS url, and the probe classifies AUTH-REQUIRED and SCRIPT-SHELL separately with the sovereign browser (nx_browser, graphics GR53's CDP witness) as the door for shells. Done-rule: every vendor bar on the boards cites bytes the estate mirrored, never a search summary.
Matrix cell	Login walls and script shells named as blockers with the sovereign browser as the door · `runtime/nx_srcreach.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_srcreach`; declare `nx_srcreach TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-27 lesson A SEARCH-ENGINE SUMMARY IS NOT A WITNESS AND THE ESTATE CAUGHT ONE: a search result asserted an NVIDIA hybrid real-time subsurface talk at SIGGRAPH 2026; the mirrored Advances course page and NVIDIA's events page (1010241 bytes) carry the token subsurface zero times, so no bar was raised on it. The researcher's mirror-then-count discipline is what makes the oracle triage safe.

RR21 Page to the month boundary · cell WATCHING · gate NO-GATE-DECLARED

Contract	`_ABSENT_:ad_page_month` · executor Organ · 1.5 u · stands on RR4
Definition of done	A newest-first listing fetched at max_results is a PREFIX (CVPR 2026 human read 120 at its cap; the splat listing read 30 of 771). The digester pages with start= until the first entry older than the bar month, so a month's population is complete by construction and nx_arxiv_census reconciles it against totalResults. Done-rule: every barscan row prints entries read equal to the month's population or names the cap it stopped at.
Matrix cell	Page to the month boundary · `runtime/nx_arxiv_digest_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_arxiv_digest_lib`; declare `nx_arxiv_digest_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-28 measure THE DIGESTER CUTS INSIDE A CODEPOINT: nx_arxiv_digest truncates each abstract head at abstract_bytes without backing off to a UTF-8 boundary, so a 600-byte digest of the body-shape listing is refused by nx_fs as malformed at byte 12385 (inside a math abstract carrying multi-byte symbols) and the estate's own reader cannot read the estate's own digest. Remedy: back the cut off over continuation bytes inside ad_digest, with a gate tooth on a fixture whose cap lands mid-codepoint. 2026-09-29 land SHIPPED 2026-09-28: nx_arxiv_digest's abstract cap no longer cuts inside a UTF-8 codepoint (ad_utf8_floor backs off over continuation bytes to the last complete codepoint); nx_arxiv_digest_gate 44/44 GREEN with the floor's known answers, a cut inside and after a two-byte codepoint and a neg-control that cuts by plain bytes; gate bc66b1e1 and CLI 7d4598f2 promoted; verified on the real body-shape listing, whose 43,680-byte digest now reads end to end.

RR22 Homonym subject classifier before newest · cell PRESENT · gate NO-GATE-DECLARED

Contract	`fs_cls_admit` · executor Organ · 2 u · stands on RR21
Definition of done	Debt 1789839847: attest took the newest listing item as the bar month while the item was the KV-cache InfiniGen homonym, and the 2026-09-27 wet scan's newest entry is a quantum-gravity paper on the phrase wet hair. BUILT 2026-09-27 in the attester (runtime/nx_frontier_scan.nx, fs_cls_admit called inside fs_newest_month before any max is taken): the axis is the listing's OWN taxonomy, arXiv category terms matched any-of against the classes a board declares as data in knowledge/subject_classes.conf, because subject tokens cannot separate a homonym that shares the query token (every entry on both listings contains it). ON-SUBJECT admitted, OFF-SUBJECT named on its own row and excluded from the max, UNCLASSIFIED admitted and counted, an undeclared board says undeclared and keeps its max; a listing whose every classified item is OFF attests nothing rather than a newer month. The summary line carries on= off= unclassified= and off_subject_newer=. Done-rule: the two homonyms above classify OFF on a live run and a control paper classifies ON.
Matrix cell	Homonym subject classifier before newest · `runtime/nx_frontier_scan.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_frontier_scan`; declare `nx_frontier_scan TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-27 measure the two homonyms named by debt 1789839847, read from the mirrored bytes through nx_arxiv_digest 2026-09-27: 2609.16215 (the KV-cache placement paper in the all:infinigen listing) carries category cs.AI, its siblings 2604.05012 cs.AR,cs.AI and 2406.19707 cs.LG,cs.DC; 2512.11025 (the phrase wet hair) carries hep-th,gr-qc,hep-ph. The controls: SceneForge 2605.14399 cs.CV,cs.GR and Infinigen-Articulated 2505.10755 cs.RO,cs.GR on the same listing; 2401.15628 (wet porous materials rendering) cs.GR on the wet listing. Subject tokens cannot separate a homonym that shares the query token (every entry above contains infinigen or wet hair), the listing's own taxonomy separates all of them, so the classifier's axis is category terms against a per-board set declared as data in knowledge/subject_classes.conf (9 boards seeded; any-of, not primary-only, because Infinigen-Articulated is primary cs.RO) 2026-09-27 land classifier BUILT AND PROMOTED 2026-09-27: nx_frontier_scan 133344 B sha256 ff32fccd (prev 126144 B sha256 be8a1fb3 banked in the prev slot), six anchored CAS edits (fs_cls_load, fs_cls_has, fs_cls_admit, fs_w_class; fs_newest_month admits only ON or UNCLASSIFIED items into its max and keeps the any-item max as off_subject_newer; the ATTEST branch refuses a homonym-only listing by name; fs_attest_domain prints CLASSES domain= allowed= or UNDECLARED), knowledge/subject_classes.conf seeded for 9 boards; contentdiff GREEN one-way gain (lost 0, gained 26 runs), behaveprobe seeds procgen IDENTICAL 4872 B, nx_frontier_scan_gate 27/27 GREEN, _offc refreshed. Rung symbol and matrix row repointed from the digester's ad_subject_classify to the attester's fs_cls_admit (the DARK-CONTRACT law: name the real entry point) 2026-09-27 measure DONE-RULE MET 2026-09-27 on live bytes with the promoted scanner: (1) procgen SB-PROCGEN's own listing read on=6 off=2, the KV-cache homonyms 2609.16215 (cs.AI) and 2604.05012 (cs.AR,cs.AI) named OFF-SUBJECT, the September month carried by AgriGen 2609.25725 (cs.RO,cs.GR) and GAE 2609.24981 (cs.CV); (2) the mirrored wet listing knowledge/fetched/cmp_skintwin_arxiv_wet_202609.xml through monthprobe cls=cs.GR,cs.CV,physics.optics,physics.med-ph,eess.IV read items=11 dated=11 newest=2025-04 on=2 off=9 unclassified=0 off_subject_newer=2025-12, the quantum-gravity Wet Hair paper 2512.11025 (hep-th,gr-qc,hep-ph) named OFF and the wet porous rendering paper 2401.15628 (cs.GR) ON; the skintwin attest fan-out named 8 OFF rows and the procgen attest 4. The bite is permanent in the referee: nx_frontier_scan_gate gained five classprobe teeth (a planted cs.AI homonym as the newest fixture item, an on-subject May item, an undated on-subject item, a dated item with no category; newest=2026-06 not 2026-07, OFF row named, off_subject_newer visible, and the neg-control: undeclared classes take the homonym month and say undeclared), 27/27 -> 32/32 GREEN; scanner re-promoted 133824 B sha256 0c4996d4 (monthprobe takes cls= on its own argv so the referee shares no production conf), gate promoted 73880 B sha256 63fceb5e with the same three base-class symbols the day's matched-vintage control lost. Debt 1789839847 is answered by mechanism and left open for its owner's adjudication 2026-09-27 lesson classify by the listing's own taxonomy before taking a max, and declare the allowed classes as data per board: a token axis cannot separate a homonym that shares the query token, a threshold axis is a magic number, a taxonomy axis is derivable and the calibration (which arXiv terms a board lives in) is a conf row with its witnesses. The any-of rule is itself calibrated: a primary-only rule would have excluded a real procedural generator (Infinigen-Articulated, primary cs.RO) 2026-10-01 queue A CATEGORY CLASSIFIER CANNOT SEPARATE THIS HOMONYM CLASS: the aesthetictwin face-anatomy listing arxiv_faceanat_202609, read 2026-10-01 through OpenAlex, had September and August rows that are homonyms on the word face, some inside cs.CV itself (SignDino sign language, ophthalmic pretraining, sheep facial pain) beside off-category ones (DITTO teleoperation, dental CBCT, a medical-imaging privacy audit); the newest on-subject first submissions are TongueReenact 2607.28039 (2026-07-30) and GNM Head 2607.23687 (first submitted 2026-07-26, v3 2026-09-11). The seat re-dated SB-FACEANAT-2609 and MB8 to 2026-07. aesthetictwin has no cls row in knowledge/subject_classes.conf, and a category row would still admit the cs.CV homonyms, so the remedy is the ref query (require a head-model, skull or musculoskeletal term) plus a check of which Atom date the arXiv row reads (published versus updated, since a v3 must not read as a new entry). Until then the next attest of these refs can write newest 2026-09 and convict both re-dated bars STALE by homonym.

RR23 Full text behind every bar claim · cell WATCHING · gate NO-GATE-DECLARED

Contract	`_ABSENT_:ad_fulltext_claim` · executor Organ · 3 u · stands on RR8, RR9
Definition of done	A bar written from an abstract is a claim about an abstract. The open-access full text (RR8, unpaywall) is read through nx_pdf_text and the sentence the bar rests on is quoted with its byte offset in the mirrored file, so a bar row can be refuted by a reader without a fetch. Done-rule: the sotabar rows landed 2026-09-27 on graphics, gameengine, skintwin and softbody each gain a quoted sentence with its offset, or name the paywall that prevents it.
Matrix cell	Full text behind every bar claim · `runtime/nx_arxiv_digest_lib.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_arxiv_digest_lib`; declare `nx_arxiv_digest_lib TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

RR24 PubMed as a dated frontier source for clinical bars · cell WATCHING · gate NO-GATE-DECLARED

Contract	`_ABSENT_:fs_pubmed_attest` · executor Organ · 1.5 u · stands on RR16, RR19
Definition of done	MEASURED 2026-09-28: the arXiv listing for aesthetic-surgery 3D simulation holds 16 entries back to 2006 while one PubMed E-utilities search returns 28 records in the last twelve months, fetched and read by the researcher (twiningest refs twin_pubmed_search_202609 and twin_pubmed_abstracts_202609). The registry already carries the pubmed row, but frontier_sources.conf has no E-utilities grammar, so the attester cannot date a clinical bar and those bars are hand-attested. Grow the attester: esearch sorted by publication date, efetch for the records, the month read from the online publication date (one record carried a December 2026 issue date over a January 2026 epub), and the same homonym classifier by MeSH heading. Done-rule: nx_frontier_scan attest twiningest dates SB-TI-CLINIC from PubMed bytes on a live run and names any record it could not date.
Matrix cell	PubMed as a dated frontier source for clinical bars · `runtime/nx_frontier_scan.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_frontier_scan`; declare `nx_frontier_scan TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-09-29 queue The Europe PMC dated source row landed in knowledge/frontier_sources.conf (src europepmc, ISO firstPublicationDate), appended the same session. The attest proof is owed on a board whose clinical bar cites a Europe PMC query, and the mirrored Europe PMC listing's first record (an organoid hair-regeneration paper) is off-subject for aesthetic 3D imaging, so a biomedical subject classifier must stand in front of its newest month before that month can be trusted (RR22's arXiv categories do not apply to PubMed records). 2026-10-01 queue A CLINICAL BAR NOW WAITS ON THIS RUNG: aesthetictwin SB-SURGOUTCOME-2609 (predicting the surgical outcome from a 3D face or body twin, landed 2026-10-01) cites the Europe PMC listing surg_europepmc_listing_202610 (knowledge/fetched/rf_570050b4.raw), because PubMed HTML answered nx_research_fetch with a bot wall while NIH E-utilities efetch text and the Europe PMC REST search both answered; twiningest SB-TI-CLINIC-2608 was read the same way. Until fs_pubmed_attest exists every reading of those two bars is a seat reading. First live subjects for its done-rule: attest aesthetictwin prints SB-SURGOUTCOME-2609 ATTEST newest 2026-09 from the Europe PMC row, and attest twiningest prints SB-TI-CLINIC-2608 newest 2026-08.

RR25 Open-format full text first · cell WATCHING · gate NO-GATE-DECLARED

Contract	`_ABSENT_:ft_open_fulltext` · executor Organ · 2 u · stands on nothing
Definition of done	OPERATOR 2026-09-28: no proprietary formats anywhere in the estate where an open form exists, and arXiv's PDF-to-HTML conversion or better. A paper's full text is fetched in the order of its most structured OPEN form: arXiv's HTML (LaTeXML) first, then the arXiv e-print source (LaTeX, unpacked by the estate's own gunzip and untar), then PMC's JATS XML for the biomedical record, and PDF only last through RR9. Headings, equations (one copy, the TeX annotation never duplicated beside its rendering) and table cells survive as typed text. Done-rule: on a declared set of recent papers the open route returns complete text for every paper that publishes an open form, the mangled-word rate (split words, lost ligatures) is published beside the PDF route's on the same papers, and a paper with no open form falls through to RR9 by name rather than returning nothing
Matrix cell	FULLTEXT: open-format full text first -- arXiv HTML, then the LaTeX source, then JATS XML, PDF last · `runtime/nx_research_fulltext.nx` · WATCHING decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_research_fulltext`; declare `nx_research_fulltext TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	no log row names this rung yet

RR26 Dataset and media hunt from anywhere, licence-classed · cell PRESENT · gate NO-GATE-DECLARED

Contract	`fs_hunt` · executor Organ · 2 u · stands on RR1, RR2
Definition of done	OPERATOR 2026-10-03: build our capabilities in the nishi estate researcher to get from anywhere, under the licence rule (acquire only when the licence puts no royalty and no obligation on the product or delivery path; restricted data only as math and only where the licence permits; files stored by licence class). One verb in the incumbent frontier scanner, nx_frontier_scan hunt <domain> <tag> <subject words>, over the SAME fetch, per-host pace and item-extent machinery, against dataset and media repositories declared as hunt rows in knowledge/research_sources.conf, each naming where in the hit's own record the licence lives [rr26_datacite_dois_202610] [rr26_europepmc_core_202610]. Every hit's licence is classed by nx_licclass_lib against knowledge/license_classes.conf (ACQUIRE, CALIBRATE-ONLY, ACQUIRE-PRIVATE-REFERENCE, UNKNOWN-NEEDS-LICENCE-READ, REFUSED with the clause named) and written to a ranked worklist with the evidence and the deciding row. Done-rule: the OpenOximetry precedent is refused mechanically from the repository's own licence field [rr26_physionet_catalogue_202610], a live run names a permitted calibration set the seat did not know, UNKNOWN is never acquired, and every partition sums
Matrix cell	Dataset and media hunt from any repository, licence-classed · `runtime/nx_frontier_scan.nx` · PRESENT decl
Gate	NO-GATE-DECLARED — `knowledge/organ_gate.conf` carries no row for `nx_frontier_scan`; declare `nx_frontier_scan TAB <gate>` and the ship loop proves it and the verdict lands here.
Work log	2026-10-03 land SHIPPED 2026-10-03: nx_licclass_lib (new, buildroot/runtime/nx_licclass_lib.nx) with its rule as data in knowledge/license_classes.conf (48 pattern rows strictest first, 9 verdict rows, the OpenOximetry licence text quoted as the RESTRICTED-RESEARCH-ONLY grounds); nx_licclass_gate 35/35 GREEN, promoted 4780550c and registered, known answers on the live conf for Zenodo, Europe PMC, DataCite, Crossref (JSON-escaped url folded), PhysioNet licence name AND licence text REFUSED, an all-rights-reserved page PRIVATE-REFERENCE, empty and custom-terms evidence UNKNOWN, and three fail-closed neg-controls each beside a positive control; the hunt verb in nx_frontier_scan promoted 91c96e37 through the ship loop (gate GREEN, behaviour probe IDENTICAL on seeds, two lost runs adjudicated as the shared syscall shim's reworded mmap refusal and approved by evidence aa32be91); 10 hunt rows and 3 src rows (physionet, wikimedia-commons, datagov) admitted to knowledge/research_sources.conf, nx_srcreach load 65 of 65 OK with hunt rows ignored by the src reader; hunt_cover_permil 500 added to knowledge/frontier_scan.conf with its derivation 2026-10-03 measure FIRST LIVE HUNTS 2026-10-03 (skintwin): reflectance (skin reflectance spectral) 10 of 10 sources reachable, 1102 records, 198 admitted, 897 irrelevant, 7 unpaired, ACQUIRE 114, CALIBRATE-ONLY 1, PRIVATE-REFERENCE 17, UNKNOWN 65, REFUSED 1, degraded 0, partition_sums 1; colour (skin color colorimetry) 949 records, 79 admitted, ACQUIRE 20, PRIVATE-REFERENCE 9, UNKNOWN 45, REFUSED 5 -- the four openox-repo versions (pattern restricted health data license, from PhysioNet's license.name) and ENCoDE (credentialed). Three method classes agree on the precedent: the organ's verdict from the API field, the gate's verdict from the licence TEXT mirror, and the seat's hand adjudication of the same day (skintwin.plan log 1791040900). The hunt named the International Skin Spectra Archive (figshare 28228571, CC BY 4.0 on the figshare record and the DataCite rightsList), mirrored for skintwin 2026-10-03 lesson A LICENCE TEXT CAN NAME A STRICTER AND A LOOSER CLASS AT ONCE, AND THE ORDER OF THE RULE ROWS DECIDES WHICH DUTY SURVIVES: the CC BY 4.0 legal code carries public domain and CC0 once, in its closing notice that dedicates the licence text itself to the public domain, so with the public-domain rows first a CC BY text classed PD-CC0 and its attribution duty vanished. Found by reading the mirrored text before trusting the first gate GREEN; attribution rows now run first and a gate tooth holds the real text. Second lesson, same hour: Zenodo answered HTTP 400 to size=100 from an anonymous caller (UNREACHABLE on the first run, never an empty result); the row now asks size 25 2026-10-03 queue OPEN: (1) a licence-class store does not exist yet (knowledge/licstore/ absent): acquired bytes sit in knowledge/fetched with the store class on the worklist row and the refs row; owed, a store index per class and the move. (2) The hunt shares one fetch receipt path (/tmp/nx_frontier_scan/hunt.out), so two concurrent hunts race on it; derive the path from domain and tag. (3) Relevance is token coverage and cannot tell human skin from lizard or frog skin (both ranked ACQUIRE on the reflectance run): a subject classifier for the biomedical record is owed, as RR22 is for arXiv. (4) The data.gov CKAN API answered 404 (api/3 and api/action) and the GSA mirror 503; its HTML page carries no licence, so every data.gov hit reads UNKNOWN by construction. (5) A figshare COLLECTION record carries no licence field (University of Minho hyperspectral faces, c.7163569): the hunt must follow to the member articles. (6) UNKNOWN rows (65 on the reflectance run, the NIST human skin reflectance set among them) need a licence-page read lane: fetch the rightsUri, class the text, append the verdict 2026-10-03 land IDEA LEVEL SHIPPED 2026-10-03 (operator, verbatim: "we also have idea level that is just used for idea but doesnt require attribution"): a USE LEVEL orthogonal to the licence class, as data in knowledge/license_classes.conf (sha256 e4ad5d8e, 9377 B): U rows level 2026-10-03 queue OPEN for the idea level: (1) nx_frontier_scan hunt does not yet print the IDEA-ONLY verdict beside the acquisition verdict on its worklist rows (another lane owns that file today); (2) CONTRACT-BOUND detection is phrase-based and cannot see a click-through gate a page enforces without naming it; a page whose access flow requires acceptance must be read by hand until a flow probe exists; (3) CONTRACTED is allowed at IDEA-ONLY on the reading that our own two-party contract grants use -- a contract that restricts idea use would need its own B row 2026-10-03 retract THE 1102 RECORDS OF THE FIRST LIVE HUNT (log 1791043531) WERE A SAMPLE, NOT A POPULATION: three verifiers measured on 2026-10-03 that every hunt row fetched ONE page of a fixed size and never read the source total (Europe PMC rf_4208d023.raw hitCount 308 with a nextCursorMark, 100 read; Zenodo rf_522c046e.raw total 85569, 25 read; DataCite 42 read whole), HUNT-DONE printed no per-source total or capped flag, data.gov answered an off-query popular-datasets page and was counted reachable, and OSF admitted 0 with no item count. The ranked worklist of that run stands as proposals; its record count is withdrawn. The discriminating experiment is the paged re-run below. 2026-10-03 land HUNT PAGING SHIPPED 2026-10-03: nx_frontier_scan a4dbd86d (189960 B, /api/promote with expect_sha256 after nx_contentdiff named its one lost run, the renamed receipt path; _offc twin installed) reads a huntpage row per source from knowledge/research_sources.conf (total marker, cursor/page/offset/none, page arg, next-token span, echo span; 10 rows landed), pages up to hunt_page_budget (THRESHOLD 10 in knowledge/frontier_scan.conf with its calibration), percent-encodes cursors, prints SOURCE-TOTAL fetched/total/pages/budget/capped/stop/paging/echo_permil/answer per source, classes a source OFF-QUERY when its records all miss the subject and it does not echo the query (counted unanswered, records excluded from totals), EMPTY-TOTAL when it declares zero and serves none, and HUNT-DONE now carries answered/off_query/no_items/unreachable with source_partition_sums, records_fetched, records_total_declared, capped_sources and population_complete; receipts are named /tmp/nx_frontier_scan/hunt_<domain>_<tag>.out. Offline twin huntfx reads <dir>/<key>.<page>.raw. nx_frontier_scan_gate 31eb04a1 (ship loop, loss approval 90dba7f9 = two superseded nx_syscalls arena messages, behavior review 9a824c6b) gained 12 hunt teeth on a four-source fixture (cursor to total, page budget binding with a present page 3 that must not be read, off-query neg-control, empty total): 45 of 45 GREEN against the promoted subject, 33 of 45 RED against the pre-change subject (the 12 hunt teeth are the ones that failed). The staged gate .new at 74320 B was a rebuild of the UNCHANGED gate source against the newer nx_syscalls (contentdiff: only the two arena messages lost), not mine; superseded by 31eb04a1. Zenodo join changed to AND (OR answered 85569, AND 26, rf_05694ba2.raw). nx_licclass_gate authorised (cap minted, nonce 1791045120) and exercised 64 of 64 GREEN. 2026-10-03 measure PAGED RE-RUN 2026-10-03 skintwin reflectance (skin reflectance spectral), receipt /tmp/rr26_hunt_skintwin_reflectance.out: sources 10 = answered 9 + off_query 1 (datagov, 20 records served, none relevant, no echo) + no_items 0 + unreachable 0, partition sums. Per source fetched/total: europepmc 314/308 over 4 cursor pages (echo 1000), zenodo 26/26 (2 pages), datacite 42/42, figshare 26/26, osf 0/0 EMPTY-TOTAL, dryad 2/2, crossref-peerj 123/123 (2 offset pages), physionet 731 static whole catalogue, wikimedia-commons 500 over 10 pages CAPPED (publishes no total). records_fetched 1764, records_total_declared 527, admitted 378, acquire 126, private_reference 24, unknown 227, refused 1, population_complete 0 (wikimedia alone). Three method classes: the gate fixture of known answers (45 of 45), the totals checked against independent mirrors fetched before the change (Europe PMC 308, DataCite 42, Zenodo AND 26 from a separate probe), and the sums reconciled by hand (527 = 308+26+42+26+0+2+123; 1764 = 314+26+42+26+0+2+731+500+123). 2026-10-03 lesson ONE PAGE READ WHOLE IS STILL A SAMPLE OF ITS SOURCE: reading every item of the response bounded the count by the response, never by the source, and the header comment said nothing was capped in silence while every source was capped. The source total is DATA beside the row, and a source that serves records none of which match and does not echo the query answered a different question, so it is unanswered, never reachable. And a gate that forks the live subject cannot prove a new verb before promote: the subject was promoted on the content ruler first, then the gate shipped through the loop against it. 2026-10-03 queue OPEN: (1) europepmc counts 314 items over hitCount 308: the item delimiter matches six nested objects, so fetched can exceed total; name it as delimiter_overcount on SOURCE-TOTAL and tighten the delimiter; (2) wikimedia-commons is capped at 10 pages with no total (CirrusSearch ORs the words); find an AND form or a totalhits field; (3) the idea-level verdict is still not printed on hunt worklist rows (queue 1791045001); (4) the ship loop cannot prove an end-to-end gate against a staged subject (it forks the served binary), so a new verb needs promote-then-gate; a staged-subject mode for the loop is owed. 2026-10-03 retract the paged reflectance hunt published records_fetched=1764 and Europe PMC fetched=314; an independent count of one isOpenAccess per result across the four mirrored pages (rf_4208d023 100, rf_46fb4c34 100, rf_1ca6152d 100, rf_f527f9f5 8) gives exactly 308 Europe PMC records, so the true totals are 308 and 1758. The 6 extra are phantom delimiter matches that landed in unpaired. Owed: tighten the Europe PMC item delimiter, a gate tooth that fetched never exceeds a declared total, per-source lines written to the durable worklist (not only /tmp), mirror= naming every page file, and a human-subject relevance classifier (off-subject records still rank ACQUIRE).

dossiers=27 cells_joined=25 no_matrix_row=2 (sum 27) gates_recorded=0 gates_unrecorded=0 no_gate_declared=25 (sum 25 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. 24 feature rows mapped of 30 matrix rows; 6 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/researchreach.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 300–900, distinctiveness 0–1000 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 507 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 300–900 · distinctiveness 0–1000 · breadth 0–500 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 0 (a fact of the data: two players that close are that close) · data spread 284 permil of the plot

Nishi OpenAlex CORE Semantic Scholar OpenAIRE
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 4 (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	Unconventional	440	714	236	809	0 since 2026-10-03	7	13	2026-09-15
OpenAlex	Mainstream	720	233	138	666	0 since 2026-10-03	5	13	2026-09-15
CORE	Aspirational	640	400	166	800	0 since 2026-10-03	5	13	2026-09-15
Semantic Scholar	Peripheral	480	166	83	666	0 since 2026-10-03	3	13	2026-09-15
OpenAIRE	Aspirational	720	400	180	866	0 since 2026-10-03	5	13	2026-09-15

Day	Matrix rows reviewed	Players recorded
2026-09-15	18	5
2026-09-16	18	5
2026-09-17	18	5
2026-09-18	13	5
2026-09-19	13	5
2026-09-20	13	5
2026-09-21	13	5
2026-09-22	13	5
2026-09-23	13	5
2026-09-27	21	5
2026-09-28	20	5
2026-10-03	24	5
2026-10-08	24	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 120, unassessed 0. Authored assessments are not verified outcomes.

Capability	Nishi	OpenAlex	CORE	Semantic Scholar	OpenAIRE
Source registry as data rather than hardcoded urls Measured: `sr_load` exists in `runtime/nx_srcreg_lib.nx`, verified at emit. LANDED AS DATA 2026-09-06 and OPEN AS A READER. knowledge/research_sources.conf carries 40 declared sources across 7 kinds and 6 protocols with a per-host pace and a licence field -- 40 MEASURED BY THE READER ITSELF (nx_srcreach load reports rows=40 ok=40 sum=40), correcting a hand-count of 39 this board published first: the organ was right and the prose was not, which is the whole argument for a reader that counts, and it DECLARES INTENT ONLY -- endpoint, protocol, pace, licence -- because a status column in a declaration file is a constant wearing the shape of a measurement. Every aggregator is configuration-driven internally and none of the four publishes that configuration as a readable contract, except OpenAIRE whose own published API documentation index names a dspace-eprints-api for the institutional-repository tier and is coded 2 on that narrow basis [openaire_docs]. The contract here is the READER, not the file: sr_load parses the registry and refuses a row whose protocol is not one of the six declared kinds Adoption: `LIB-WIRED importers=2 nonval=1` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons behind OpenAIRE (leads)	●	●	●	●	◉
OAI-PMH incremental harvest Open — watching `runtime/nx_oaipmh_lib.nx` : `oai_list_records`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. THE ONE PROTOCOL THAT TURNS A LIST OF SITES INTO A HARVESTABLE RECORD. OAI-PMH v2.0 defines six verbs over HTTP with resumption tokens and a from-until selective-harvest window [oaipmh_spec], which is what makes INCREMENTAL harvest possible rather than re-querying an API forever. CORE and OpenAIRE are both built on harvesting repositories and are coded 2, OpenAIRE on the dspace-eprints-api its own documentation index names [openaire_docs]; OpenAlex ingests from Crossref, PubMed and repositories and exposes a REST catalogue rather than a harvest endpoint, proven by a live unauthenticated query returning 200 [openalex_live]; Semantic Scholar publishes Graph, Recommendations and Datasets APIs and no harvest verb in its mirrored surface [s2_api]. Done-rule names resumption explicitly because a harvester that stops at the first page silently acquires a prefix and reports success dossierwatching oai_list_records Pros none measured yetCons behind OpenAlex, CORE (leads), OpenAIRE (leads); open contract; expected source declaration not established	○	●	◉	○	◉
Registry-of-registries expansion to the institutional tier Open — watching `runtime/nx_oai_registry.nx` : `oai_registry_expand`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. THE OPERATOR'S OWN EXPANSION MECHANISM, ENCODED AS DATA INSTEAD OF AS A LIST. OpenDOAR and ROAR each enumerate institutional repositories that publish an OAI-PMH endpoint, so reach grows by ADMITTING A REGISTRY rather than by pasting a thousand urls, and the two registry rows already sit in the conf. A derived endpoint is never hand-written back into the registry -- it is written to knowledge/status/oai_endpoints.derived with the registry that produced it, so removing a registry removes exactly its own children and the provenance of every reachable repository stays a row a person can read. OpenAIRE aggregates this tier already and its own documentation index names a dspace-eprints-api for it [openaire_docs] dossierwatching oai_registry_expand Pros none measured yetCons behind OpenAlex, CORE (leads), OpenAIRE (leads); open contract; expected source declaration not established	○	●	◉	○	◉
Reachability measured per source with the blocker NAMED Measured: `sr_probe` exists in `runtime/nx_srcreach.nx`, verified at emit. THE ROW THIS BOARD WAS ADMITTED FOR. Operator 2026-09-06: we need to measure and make sure we never have any unreachable data. NONE of the four rivals publishes a per-source reachability verdict, so all four are coded 0 as a measured absence of a comparable surface and not as a fabricated audit. Done-rule: a partition over the whole registry that SUMS -- REACHABLE, RATE-LIMITED, TLS-FAIL, AUTH-REQUIRED, ROBOTS-DENIED, NOT-FOUND, UNPROBED -- where UNPROBED is its own state because a source nobody probed must never read as one that passed, and where each non-reachable row carries the remedy that would close it. TWO BLOCKERS ARE ALREADY MEASURED AND CARRIED IN THE REGISTRY RATHER THAN DROPPED FROM IT: codeberg fails our sovereign TLS chain and semanticscholar rate-limits us, both reported by the sibling probe lane on 2026-09-06 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 OpenAlex, CORE, Semantic Scholar, OpenAIRECons 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)	●	○	○	○	○
Per-source pace budget printed and obeyed Measured: `sr_pace` exists in `runtime/nx_srcreach.nx`, verified at emit. Operator 2026-09-06: a regular cadence for all domains without smashing our nas resources. The budget is PER HOST and not global because a shared budget starves the slow hosts first, and the pace values are derived where they can be and DECLARED AS DEFAULTS where they cannot -- 3000 ms is ALSO a CONSERVATIVE DEFAULT with UNGROUNDED provenance -- this row first called it arXiv's published courtesy rate adopted verbatim, and the arXiv API User's Manual mirrored at 148526 bytes contains the token seconds ZERO times and rate limit ZERO times, so that derivation is withdrawn on the same evidence standard as the OpenAlex one [arxiv_apimanual]; 1000 ms for OpenAlex is a CONSERVATIVE DEFAULT AND NOT A DERIVATION, because this row first claimed it was adopted from published polite-pool guidance and the mirrored documentation surface [openalex_api] contains no such guidance, so the derivation is WITHDRAWN rather than left standing while the key-free access OpenAlex genuinely does provide is proven instead by a live unauthenticated query returning 200 [openalex_live]; 5000 ms is the default for a host that has published no rate, and 10000 ms is reserved for a host MEASURED to throttle us. Done-rule: the run PRINTS sum of requests before it issues the first one and refuses to start if that number exceeds the declared ceiling, because a sweep that cannot state its own request count is a load incident waiting to happen 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 OpenAlex, CORE, Semantic Scholar, OpenAIRECons 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)	●	○	○	○	○
Fleet cadence
every compare domain revalidated against the record on a beat Open — watching `runtime/nx_research_harvest.nx` : `rh_beat`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. Operator 2026-09-06: going out and validating we are staying on top of current research and trends and functions. Rivals coded 0 for the category reason that none of them maintains a capability board to revalidate. THIS IS THE SAME CONTRACT deepresearch DN1 fs_sweep_all HAS HELD OPEN SINCE 2026-08-20 AND IT IS NAMED HERE RATHER THAN DUPLICATED: that rung is blocked on exactly two things, a second key-free scholarly index and a per-source pace budget, and this board supplies both -- openalex is admitted as a registry row and sr_pace is the budget. The cadence must be STAGGERED across domains rather than run as one fleet pass, so that outbound load is flat instead of a daily spike dossierwatching rh_beat Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Cross-source deduplication on a stable identifier Open — watching `runtime/nx_xref_lib.nx` : `xr_doi_join`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. All four rivals dedupe on DOI and it is the core of what an aggregator is, so all four are coded 2 and this row is a genuine behind. The join key is why crossref and datacite sit in the registry as sources in their own right: without a resolvable identifier, the same paper harvested from arXiv, from an institutional repository and from a publisher is three records, and coverage measured over undeduplicated records is inflated by construction rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them dossierwatching xr_doi_join Pros none measured yetCons behind OpenAlex (leads), CORE (leads), Semantic Scholar (leads), OpenAIRE (leads); open contract; expected source declaration not established	○	◉	◉	◉	◉
Open-access full text actually acquired, not merely indexed Open — watching `runtime/nx_oafull_lib.nx` : `oa_fulltext`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. CORE is coded 2 because full text is the axis it is built on and its v3 API exposes full-text search and download over the aggregated corpus [core_api]. The unpaywall row in the registry is what turns a metadata hit into a legally readable document by mapping a DOI to its open-access copy. Metadata coverage without full text is the trap this row exists to name: a record you can count but not read contributes nothing to a reader or a reranker dossierwatching oa_fulltext Pros none measured yetCons behind OpenAlex, CORE (leads), Semantic Scholar, OpenAIRE; open contract; expected source declaration not established	○	●	◉	●	●
Layout-aware parsing: a PDF becomes structured text with its maths and tables intact Open — watching `runtime/nx_layout_parse.nx` : `lp_blocks`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. PIPELINE STAGE ONE FROM THE OPERATOR'S ARCHITECTURE FEEDBACK. CORE is coded 3 for extracted full text without a published claim of structural fidelity; the other three index metadata and abstracts. The estate has nx_pdf_text today, which recovers a text stream and not a structure, so equations and tables arrive as flattened prose and a reader cannot tell a caption from a cell. The field moved here in 2026: the Institutional Books enriched-text pipeline reports denoising and deduplication over historical corpora at scale, and a comparative evaluation of digitization pipelines was published the same month [arxiv_csdl_202609]. Done-rule: a block tree with typed regions where a table round-trips to its cells and an equation survives as its own region dossierwatching lp_blocks Pros none measured yetCons behind CORE; open contract; expected source declaration not established	○	○	◐	○	○
Code graphing
a repository becomes an AST-level call graph Open — watching `runtime/nx_codegraph.nx` : `cg_ast_edges`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. PIPELINE STAGE TWO. All four rivals are coded 0 as a measured category absence: they aggregate papers and none of them parses repositories, which is precisely the white space that makes the forge rows in the registry worth harvesting at all. The estate already parses ITS OWN tree with nx_funcmine and nx_capgraph_edges and owns a full compiler front end in nx_cc, so this is a reuse and generalisation rung rather than a new capability from nothing, and it belongs to the coding board for our own sources and to this board for acquired third-party repositories dossierwatching cg_ast_edges Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Hybrid retrieval: a knowledge graph and a vector index answering one query Open — watching `runtime/nx_kg_hybrid.nx` : `kgh_query`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. PIPELINE STAGE THREE. OpenAIRE is coded 2 because its published surface is explicitly a research GRAPH with a v4 graph API [openaire_api]; OpenAlex exposes an entity graph over works, authors and institutions, visible directly in the authorship and institution entities of a live works response [openalex_live]; Semantic Scholar publishes an Academic Graph API [s2_api]; CORE is coded 0 for a corpus-and-search surface with no graph claim in its mirrored docs [core_api]. THE 2026 FRONTIER STATES THE ARCHITECTURE DIRECTLY: LLMs interpret, embeddings organize, graphs emerge, as agent-driven compilation of scientific knowledge [agentkg2026]. Our substrate for both halves already exists separately -- a vector index and a segment store -- and the contract is the JOIN, not either leg dossierwatching kgh_query Pros none measured yetCons behind OpenAlex, Semantic Scholar, OpenAIRE (leads); open contract; expected source declaration not established	○	●	○	●	◉
Agentic multi-stage compile with a critic that can refuse Open — watching `runtime/nx_agent_compile.nx` : `ag_stage`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. PIPELINE STAGE FOUR. Rivals coded 0 as a category statement: all four are indexes and none runs a research agent over its own corpus. The operator named the shape as gather, write, criticise. The estate ALREADY OWNS THE CRITIC and it is the strongest thing it can bring here: nxr_assess grades every numeric claim against its cited source and gates publish, and cvx_answer_verify corroborates across independent origins. The 2026 frontier adds the missing half, which is GOVERNANCE OF WHAT THE AGENT WRITES INTO MEMORY -- multi-agent knowledge-graph construction with domain-expert review [agentkg2026] and FAIR services with provenance and DOIs described as the foundation for agentic scientific workflows [foxden2026]. Done-rule: a stage that cannot write to the store until the critic has graded it, proven by a planted false claim that never lands dossierwatching ag_stage Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Content-addressed identity for every acquired document Source present; authored stronger claim: `cid_of` in `runtime/nx_canon_cid.nx`, verified at emit. EXCEED. The datum IS its sha256, so an acquired document has one identity independent of which of the 39 sources delivered it, and the same paper harvested three ways collapses to one object by construction rather than by a dedup pass that can be wrong. All four rivals key on mutable external identifiers and record where a copy lives; none makes the bytes their own name. Symbol verified present at nx_canon_cid.nx line 144 Adoption: `LIB-WIRED importers=116 nonval=86` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of OpenAlex, CORE, Semantic Scholar, OpenAIRE; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
Tamper-evident append-only acquisition store Source present; authored stronger claim: `ss_commit` in `runtime/nx_seg_store.nx`, verified at emit. EXCEED. History is sacred and acquisition is additive, so a source that later withdraws or rewrites a record cannot silently rewrite ours and the withdrawal is itself a fact we hold. This is the property that makes a coverage measurement reproducible months later, which no hosted aggregator offers because their corpus is mutable underneath the query. Symbol verified present at nx_seg_store.nx line 1516 Adoption: `LIB-WIRED importers=498 nonval=372` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of OpenAlex, CORE, Semantic Scholar, OpenAIRE; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
Fetch-time provenance binding a mirror to the url it claims Source present; authored stronger claim: `crg_prov_state` in `runtime/nx_compare_refs_gate.nx`, verified at emit. EXCEED, and it is the guard that made this board's own citations honest on the day it was written. Three states because two would lie -- PROVEN, MISMATCH and UNPROVEN -- and the binding is made at FETCH time, the only moment both facts are in hand. It earned itself here immediately: of the four rival documentation mirrors fetched for this board, TWO were non-content pages returning a clean status 200, an Apache directory index and a ReDoc JavaScript shell, and a register without fetch-time binding would have carried both as ordinary citations. Symbol verified present at nx_compare_refs_gate.nx line 402 Adoption: `GATE:LIVE trial=-` — fully adopted (top of its ladder). dossier Pros Source symbol found; stronger capability claim is authored; ahead of OpenAlex, CORE, Semantic Scholar, OpenAIRE; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
Citation still served: does the source still serve the bytes we pinned Measured: `rd_same` exists in `runtime/nx_refdrift.nx`, verified at emit. OPEN, AND THIS ROW IS A CORRECTION OF A PUBLISHED CLAIM ON A SIBLING BOARD RATHER THAN A NEW CONTRACT. On 2026-09-06 nx_domain_admit reported UNGROUNDED symbol rd_same not word-boundary-present in runtime/nx_refdrift.nx against the deepresearch row that carries exceed 1 and a note claiming LANDED. A grep over buildroot/runtime with corpus_complete=1 found func rd_same in exactly one file and it is nx_rebuild_drain.nx, an unrelated organ that shares the rd_ prefix, so a careless verification would have CONFIRMED the symbol and closed the finding. The complete outline of nx_refdrift.nx reports 64 declarations and its functions are rd_slurp, rd_put2, rd_ymd, rd_basename, rd_sha, rd_recite_row and main: THE FOUR-STATE COMPARATOR IS REAL AND RUNS INLINE IN MAIN, SO THE CAPABILITY EXISTS AND ITS CONTRACTED ENTRY POINT WAS NEVER WRITTEN. The remedy is to give the capability the named entry the board contracted, never to repoint the row at an unrelated helper, and it is carried here as an open rung so the correction is visible on a board instead of living in a transcript 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 OpenAlex, CORE, Semantic Scholar, OpenAIRECons 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)	●	○	○	○	○
Lexical retrieval over the acquired corpus Measured: `idf_q10` exists in `runtime/nx_intlog.nx`, verified at emit. Integer Okapi BM25, deterministic and reproducible. Every rival ranks; this is a par row and is here because a coverage board that cannot retrieve what it acquired has measured a warehouse and not a library rival marks uncited — the Yes, Best or Part codes on this row are an observation read with no reference mark behind them Adoption: `LIB-WIRED importers=9 nonval=7` — fully adopted (top of its ladder). Pros Implementing source symbol found; runtime acceptance separate; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	●	●	●	●
Frontier noticing driven by a board's own data Source present; authored stronger claim: `fs_scan_source` in `runtime/nx_frontier_scan.nx`, verified at emit. EXCEED. Seeds are derived from a domain's OWN matrix rows and keyword field, the outside is pulled through the sovereign fetcher over sources that are themselves data, and findings are diffed against our rows to emit gap proposals that name the row each candidate challenges. All four rivals are coded 0 as a category statement: they answer or index for whoever asks and none maintains a capability board to notice a gap in, so there is no comparable surface and nothing is fabricated here. Symbol verified present at nx_frontier_scan.nx line 622. ITS MEASURED LIMIT IS THIS BOARD'S REASON TO EXIST: it was wired to two sources (arXiv and HN Algolia) until 2026-09-06 and to seventeen by 2026-09-27 (GitHub, Hugging Face, Gitea, the SIGGRAPH 2026, Eurographics 2026 and SIGGRAPH Asia 2026 indices, seven arXiv venue-tag rows and a cs.GR row, each proven reachable before admission), and the registry raises that ceiling to 62 declared sources plus the institutional tier Adoption: `RUN-BY:clock` — fully adopted (top of its ladder). Pros Source symbol found; stronger capability claim is authored; ahead of OpenAlex, CORE, Semantic Scholar, OpenAIRE; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	◉	○	○	○	○
Venue tags as frontier source rows DARK — `runtime/nx_frontier_scan.nx` EXISTS but does not declare `fs_venue_rows`: 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. RR16 (2026-09-27). The arXiv comment field is a venue register: co:SIGGRAPH 2026 (85 entries), co:SIGGRAPH Asia 2026 (55), the Eurographics family (22), co:CVPR 2026 with a human filter (120 at the cap) were fetched, digested and landed as DATED rows in frontier_sources.conf beside co:ECCV 2026, co:NeurIPS 2026 and co:IEEE VR 2026; the first attest run over graphics printed the venue rows beside arXiv and HN [kesen_sig2026] [arxiv_co_siga2026]. Watch contract: fs_venue_rows names the per-venue NEWER, SAME, OLDER or UNDATED line in attest. Rival cells coded 0: no rival publishes a per-venue frontier verdict in a mirrored source. dossierwatching fs_venue_rows Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Static venue indices and the Eurographics digital library DARK — `runtime/nx_frontier_scan.nx` EXISTS but does not declare `fs_static_venue_rows`: 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. RR17. Ke-Sen Huang's Eurographics 2026 (108 entries) and SIGGRAPH Asia 2026 (7 conditional) indices joined the SIGGRAPH 2026 row as static month-tagged sources [kesen_eg2026] [kesen_siga2026]; the Eurographics Digital Library's OAI-PMH endpoint joined the registry UNPROBED. Watch contract: attest reads each index and the SIGGRAPH Asia count grows to the proceedings by December. Rival cells coded 0: not witnessed by a mirrored source. dossierwatching fs_static_venue_rows Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Venue completeness by DBLP DARK — `runtime/nx_srcreach.nx` EXISTS but does not declare `sr_dblp_population`: 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. RR18. A venue index read at a cap is a prefix (CVPR 2026 human read 120 at its cap [arxiv_co_cvpr2026_human]); DBLP lists every paper of a venue and year and is the denominator a mirrored index is reconciled against. Watch contract: mirrored over DBLP-listed per venue with the partition summing. Rival cells coded 0: not witnessed by a mirrored source. dossierwatching sr_dblp_population Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Labs, standards bodies and governments on the registry DARK — `runtime/nx_srcreach.nx` EXISTS but does not declare `sr_lab_rows`: 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. RR19. NVIDIA, Meta, Disney, MPI Perceiving Systems, Google, Microsoft, Adobe, Epic and Apple records, the Khronos and MPEG registries [mpeg_gsc_202609], NIST [nist_skin_reflectance_202609], DARPA [darpa_vital_202609], CORDIS and NIH RePORTER: 22 rows admitted 2026-09-27 under the six declared kinds (labs as publisher until a lab kind lands), srcreach load rows=62 ok=62. Watch contract: nx_srcreach probe reads every row REACHABLE or names its blocker, and a lab kind is admitted to the reader. Rival cells coded 0: not witnessed by a mirrored source. dossierwatching sr_lab_rows Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Login walls and script shells named as blockers with the sovereign browser as the door DARK — `runtime/nx_srcreach.nx` EXISTS but does not declare `sr_shell_blocker`: 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. RR20. Measured 2026-09-27: the MetaHuman news site redirects to a login while the developer documentation site serves the same notes as 259454 bytes of HTML; earlier vendor pages returned status-0 fragments and an Angular shell. Watch contract: the probe classifies AUTH-REQUIRED and SCRIPT-SHELL separately, a vendor row carries an alternate witness url, and the sovereign browser is the door for shells. Rival cells coded 0: not witnessed by a mirrored source. dossierwatching sr_shell_blocker Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Page to the month boundary DARK — `runtime/nx_arxiv_digest_lib.nx` EXISTS but does not declare `ad_page_month`: 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. RR21. A newest-first listing fetched at max_results is a prefix; the digester pages with start= until the first entry older than the bar month and nx_arxiv_census reconciles the month against totalResults. Watch contract: every barscan row prints entries read equal to the month's population or names the cap it stopped at. Rival cells coded 0: not witnessed by a mirrored source. dossierwatching ad_page_month Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Homonym subject classifier before newest Measured: `fs_cls_admit` exists in `runtime/nx_frontier_scan.nx`, verified at emit. RR22 (debt 1789839847). The newest listing item was the KV-cache InfiniGen homonym on procgen, and the 2026-09-27 wet-appearance scan's newest entry is a quantum-gravity paper on the phrase wet hair. Built 2026-09-27 in the attester: each entry classified ON or OFF by the listing's own taxonomy (arXiv category terms, any-of against the classes the board declares in knowledge/subject_classes.conf) before any max is taken, OFF items named on their own row and excluded, the partition printed on the summary line, a homonym-only listing attesting nothing. Subject tokens were refused as the axis because every entry on both listings carries the query token. Rival cells coded 0: not witnessed by a mirrored source. Adoption: `RUN-BY:clock` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of OpenAlex, CORE, Semantic Scholar, OpenAIRE; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Full text behind every bar claim DARK — `runtime/nx_arxiv_digest_lib.nx` EXISTS but does not declare `ad_fulltext_claim`: 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. RR23. A bar written from an abstract is a claim about an abstract; the open-access full text is read through nx_pdf_text and the sentence a bar rests on is quoted with its byte offset in the mirrored file. Watch contract: the sotabar rows landed 2026-09-27 on graphics, gameengine, skintwin and softbody each gain a quoted sentence with its offset or name the paywall that prevents it. Rival cells coded 0: not witnessed by a mirrored source. dossierwatching ad_fulltext_claim Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
PubMed as a dated frontier source for clinical bars DARK — `runtime/nx_frontier_scan.nx` EXISTS but does not declare `fs_pubmed_attest`: 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. RR24. The clinical record lives on PubMed, not arXiv: 16 arXiv entries back to 2006 against 28 PubMed records in twelve months for the same aesthetic-surgery 3D subject, both fetched and read by the researcher on 2026-09-28. The registry already carries the pubmed row; the attester has no E-utilities grammar, so clinical bars are hand-attested until this lands. Rival cells coded 0: not witnessed by a mirrored source. dossierwatching fs_pubmed_attest Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
FULLTEXT
open-format full text first -- arXiv HTML, then the LaTeX source, then JATS XML, PDF last Open — watching `runtime/nx_research_fulltext.nx` : `ft_open_fulltext`, re-measured on every compare beat. Ship that symbol and this mark flips itself; the `comparewatch-` plane row flips with it. RR25. MEASURED 2026-09-28 on one paper (2604.02320): the arXiv HTML read through nx_dr_extract gave 73 KB of clean text and recovered the capture facts the PDF route lost, where nx_pdf_text produced about 66 KB with split words and about 11.3 MB of undecoded stream bytes. The four columns were not read for full-text claims today, so their cells are 0 by the board's rule, never proven absent dossierwatching ft_open_fulltext Pros none measured yetCons open contract; expected source declaration not established	○	○	○	○	○
Dataset and media hunt from any repository, licence-classed Measured: `fs_hunt` exists in `runtime/nx_frontier_scan.nx`, verified at emit. RR26. Operator 2026-10-03: build the researcher to get from anywhere, licence-aware. nx_frontier_scan hunt <domain> <tag> <subject> asks every hunt row of knowledge/research_sources.conf (Europe PMC core [rr26_europepmc_core_202610], Zenodo, DataCite [rr26_datacite_dois_202610] and its figshare and OSF clients, Dryad, the PhysioNet catalogue [rr26_physionet_catalogue_202610], Wikimedia Commons, Crossref for PeerJ, data.gov), reads each hit's licence from its own record, classes it and writes a ranked worklist. First live run (skintwin, skin reflectance spectral): 10 of 10 sources reachable, 1102 records, 198 admitted, ACQUIRE 114 / CALIBRATE-ONLY 1 / PRIVATE-REFERENCE 17 / UNKNOWN 65 / REFUSED 1, partition sums. Rival cells coded 0: not witnessed by a mirrored source. Adoption: `RUN-BY:clock` — fully adopted (top of its ladder). dossier Pros Implementing source symbol found; runtime acceptance separate; ahead of OpenAlex, CORE, Semantic Scholar, OpenAIRE; fully adopted on the estate ladderCons none on the measured axes (rival marks are documented presence, not depth)	●	○	○	○	○
Licence class and acquisition rule for every hunt hit Measured: `lc_decide` exists in `runtime/nx_licclass_lib.nx`, verified at emit. RR26. The operator's rule as data (knowledge/license_classes.conf, 48 pattern rows strictest first, 9 verdict rows): ACQUIRE for public domain and attribution-only, CALIBRATE-ONLY for share-alike and no-derivatives, ACQUIRE-PRIVATE-REFERENCE for non-commercial and all-rights-reserved, REFUSED for research-only health-data licences, UNKNOWN never acquirable and every conf defect failing closed to it. nx_licclass_gate 35/35 GREEN, a last-match mutant killed by seven teeth. Rival cells coded 0: not witnessed by a mirrored source. Adoption: `LIB-UNCENSUSED` — PARTIAL: a library absent from the import-graph census (younger than the last eco-graph build, or unreachable from buildroot/runtime). not adopted: a library absent from the import-graph census (younger than the last eco-graph build, or unreachable from buildroot/runtime) Pros Implementing source symbol found; runtime acceptance separate; ahead of OpenAlex, CORE, Semantic Scholar, OpenAIRECons not adopted yet: a library absent from the import-graph census (younger than the last eco-graph build, or unreachable from buildroot/runtime)	●	○	○	○	○

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	8	rows carrying a Yes, Best or Part in a rival column
cited	6	a reference mark resolves to a pinned mirror
uncited	2	an observation read, badged until a reference lands
rows, summed	8	RECONCILED: the parts sum to the total 8

Delivery and evidence

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

Risk register

Risk	Likelihood x impact	Mitigation
A reachability probe becomes a politeness incident: probing 39 hosts plus a derived institutional tier is itself outbound load, and a probe that ignores the budget is the defect it exists to prevent	likely x high	RR2 is explicitly dependent on RR1 and gated by RR3's budget; the probe is a client of the pace budget and not an exception to it.
Coverage measured over undeduplicated records inflates by construction, so an early coverage number could look like progress that is only double counting	possible x high	RR7 precedes every coverage claim in M3, and RR2's partition counts SOURCES not records, so the M1 number cannot be inflated by duplicates.
A blocker recorded once becomes a fossil: codeberg TLS and semanticscholar throttling are true today and may be fixed upstream tomorrow, and a latched blocker reads identically whether it cleared a month ago or is firing now	likely x medium	RR2's states are re-measured every pass rather than latched, and the registry carries no status column at all, so a cleared blocker clears itself on the next probe instead of needing someone to remember.
The pipeline rungs RR9 to RR12 are the field's own frontier and could absorb the whole board while the reach rungs, which are the operator's actual ask, stay open	possible x high	M1 and M2 contain no pipeline rung. The ladder is ordered so reach and cadence complete before compile begins, and rungrole marks RR10 an arm so it cannot be mistaken for progress on the coverage number.

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	RR13	measure	A PUBLISHED EXCEED CLAIM WAS FOUND UNGROUNDED, AND THE COINCIDENCE THAT NEARLY HID IT IS THE FINDING. nx_domain_admit reported UNGROUNDED symbol rd_same not word-boundary-present in runtime/nx_refdrift.nx. A grep over buildroot/runtime with corpus_complete=1 found func rd_same in exactly one file: nx_rebuild_drain.nx, an unrelated organ sharing the rd_ prefix. A verification that greps for the symbol name alone would have CONFIRMED it and closed the finding. Only the complete outline of nx_refdrift.nx settled it: 64 declarations, functions rd_slurp rd_put2 rd_ymd rd_basename rd_sha rd_recite_row and main, and no comparator among them. THE CAPABILITY IS REAL AND RUNS INLINE IN MAIN; ITS CONTRACTED ENTRY POINT WAS NEVER WRITTEN. Verify a symbol IN THE ORGAN THAT NAMES IT, never in the tree.
2026-09-06	RR2	measure	TWO OF FOUR RIVAL DOCUMENTATION MIRRORS WERE NON-CONTENT PAGES RETURNING A CLEAN STATUS 200. graph.openaire.eu/docs/apis is an Apache directory index of 2813 bytes and api.semanticscholar.org/api-docs is a ReDoc JavaScript shell of 1752 bytes whose real specification loads at view time from graph/v1/swagger.json. Both saved as successful bodies. This is the stub class the estate has measured before, and it happened on the very board being built to measure reachability, which is the argument for RR2 stated better than any specification could: A HOST THAT ANSWERS IS NOT A HOST THAT DELIVERED WHAT YOU ASKED FOR, and a reachability probe that only checks status is a probe that would have called both of these fine.
2026-09-06	RR6	queue	THE CADENCE RUNG IS NOT NEW WORK, IT IS A BLOCKED SIBLING GETTING ITS BLOCKERS. deepresearch DN1 fs_sweep_all has been open since 2026-08-20 with a done-rule naming exactly two preconditions: a second key-free scholarly index admitted as a source row, and a per-source pace budget the run prints and obeys. openalex is now a registry row and RR3 is the budget. When RR6 lands, DN1's done-rule is satisfiable and the two rungs should be reconciled rather than both left open, because one capability carried as two open contracts on two boards is how a backlog grows without work being missed.
2026-09-06	RR14	retract	A CLAIM IN THIS PLAN'S OWN FIRST DRAFT WAS FALSE ABOUT OUR OWN ESTATE, AND ONE nx_memfind CALL REFUTED IT. The pos line asserted there is no declared source registry and no harvest protocol. nishi-library reached v0.0.102 with a fifty-plus source registry across nine tiers, a generic OAI-PMH adapter with resumptionToken cursors, DV2.0 hub-satellite-link with FTS5, 2.2 million works and 1.87 million citation edges, LIVE since 2026-04-29. THE DISCRIMINATING QUESTION WAS NOT WHETHER THE CAPABILITY EXISTS BUT WHICH STACK IT EXISTS IN, and asking the memory plane BEFORE writing the claim costs one call while publishing it costs the next reader the whole investigation. The capability search this estate mandates covers ORGANS; it does not cover retired-stack capability, and nx_memfind is the instrument that does.
2026-09-06	RR2	lesson	THIS BOARD'S FIRST CITATION SET FAILED THIS BOARD'S OWN THESIS, AND THE ESTATE'S GATE NAMED IT. nx_compare_refs_gate reported OPEN-REDIRECT-RATCHET-ROSE-NEW OPEN-REDIRECT researchreach.refs openalex_api MOVED: the mirror was fetched with hops=1 and its canonical read CANON-HEAD-TRUNCATED, so it is NOT acquitted as a capture of the url the row cites. A register that only checks that a mirror EXISTS and that its pin matches would have passed it, because both were true. THE FETCH SUCCEEDED, THE PIN WAS CORRECT, AND THE CITATION WAS STILL NOT PROVEN TO BE OF THE PAGE CITED. This is the exact reason RR2's done-rule refuses to treat an HTTP 200 as reach: a host that answers is not a host that delivered what you asked for.
2026-09-06	RR1	lesson	AN APPEND TO regen.list DOUBLE-APPLIED AFTER A 503, DESPITE FOLLOWING THE SPACED-READS DISCIPLINE EXACTLY. The call returned 503 Outcome Unknown at stage read-after-accept. Two spaced size reads both returned 1052, so the append was re-issued, and the re-issue reported total=1080 -- which is 1052 plus 14 plus 14, proving the first call had landed all along and both reads were stale. THE READ PATH LAGS THE WRITE PATH FOR EVERY VERB, SO NO NUMBER OF READS IS EVIDENCE THAT A WRITE DID NOT LAND. Only the writer's own reported total settled it, and it settled it instantly. For an append, predict the resulting byte count and read it back from the WRITE, never from a separate size call.
2026-09-06	RR1	land	RR1 AND RR3 SHIPPED AS SOURCE AND BITE-PROVEN ON THE LAPTOP BUILD FARM, BECAUSE THE NAS COMPILER WAS REFUSING ALL DAY. nx_srcreg_lib.nx (10788 B, sr_load plus the closed kind and protocol vocabularies and sr_classify) and nx_srcreach.nx (9552 B, sr_pace plus the load and plan verbs) both landed in buildroot/runtime. THE LOCAL LANE IS NOT A CONCESSION, IT IS THE DOCUMENTED REMEDY: nx_sov_build_run --build-only on the laptop is governor-free, it compiled first try to 29605 bytes through nx_cc into nxasm_x86 with no gcc anywhere, and the laptop measured load1_centi=19 against the NAS's 1918. PROVEN IN BOTH DIRECTIONS ON A FIXTURE ASSEMBLED AT RUNTIME: four planted defects each FIRED and each was NAMED by its key -- BAD-KIND, BAD-PROTO, BAD-PACE, BAD-FIELDS -- with the partition summing exactly, rows=6 ok=2 bad_fields=1 bad_kind=1 bad_proto=1 bad_pace=1 sum=6 verdict=REFUSED exit=1. THE POSITIVE CONTROL IS THE HALF THAT MATTERS MOST and it is why this is not a guard that refuses everything: a clean three-row registry read rows=3 ok=3 sum=3 verdict=OK exit=0. The budget was proven both ways too: 3 sources at 10 requests each is 30 planned, under a ceiling of 100 it passes exit 0, over a ceiling of 20 it prints REFUSED-OVER-CEILING and exits 1 saying Nothing was fetched. The arithmetic is independently checkable from the printed values: longest_source_wall_ms 45000 is 9 waits times the 5000 ms source, sequential_wall_ms 81000 is 9 times 3000 plus 1000 plus 5000.
2026-09-06	RR2	lesson	sr_probe WAS DELIBERATELY NOT DEFINED, AND THE ORGAN SAYS SO IN ITS OWN HEADER. Defining the symbol would flip the RR2 watch cell on the published board with no capability behind it, and a completion signal that keys on a name rewards writing the name. RR2 needs the network leg and stays an OPEN contract until it can actually probe. The two rungs that could be honestly finished were finished; the one that could not was left visibly unfinished.
2026-09-06	RR1	queue	THE NAS BUILD IS QUEUED, NOT SKIPPED. nx_build_admit read QUEUE on three different conjuncts across the session -- procs_blocked 12 of 8 as an I/O storm, then procs_running 14 of 8 as CPU saturation -- so the build was handed to /api/build, which absorbs a QUEUE verdict by launching nx_buildq rather than failing. The organ is source-landed and locally proven; when headroom appears the NAS artifact follows, and the RR1 and RR3 watch cells flip themselves on the next emit because the contracts name the real symbols sr_load and sr_pace. NOTHING HERE NEEDS A SEAT TO REMEMBER IT.
2026-09-06	RR1	measure	THE READER CORRECTED ITS OWN BOARD ON ITS FIRST LIVE RUN. Every prose claim on this domain said 39 declared sources, hand-counted while the registry was written. nx_srcreach load knowledge/research_sources.conf reports rows=40 ok=40 bad_fields=0 bad_kind=0 bad_proto=0 bad_pace=0 sum=40 verdict=OK on 10177 bytes. THE ORGAN IS RIGHT AND THE PROSE WAS NOT, and that is precisely the argument for RR1 existing at all: a population figure that nobody re-measures is a claim with an expiry date nobody recorded. Corrected to 40 wherever it was published.
2026-09-06	RR3	measure	THE CADENCE QUESTION NOW HAS A NUMBER, COMPUTED BEFORE ANYTHING IS FETCHED. Live on the real registry: 40 sources at 3 requests each is 120 planned_requests, longest_source_wall_ms 20000 set by the one host measured to throttle us, sequential_wall_ms 280000. Against a ceiling of 100 it printed REFUSED-OVER-CEILING and exited 1 with Nothing was fetched. The arithmetic is checkable from the printed values alone: the declared paces sum to 140000 ms and two waits per source is 280000, and 2 x the 10000 ms semanticscholar pace is the 20000 floor. THAT IS THE ANSWER TO not smashing our nas resources -- the cost is stated and refused in advance rather than discovered from the array afterwards.
2026-09-06	RR2	land	RR2 SHIPPED AND THE FIRST REAL REACHABILITY CENSUS RAN. nx_srcreach PROMOTED 795d0912 at 50392 B, REGISTERED, CAPPED, LIVE. Live over all 40 declared sources, paced per host, every probe into ONE fixed scratch name: probed=40 reachable=19 rate_limited=0 auth_required=3 not_found=6 other_status=12 transport_fail=0 unprobed=0 sum=40 rows=40. THE PARTITION SUMS TO THE POPULATION and every non-reachable row printed its key and the remedy that would close it.
2026-09-06	RR2	retract	DO NOT READ 19 OF 40 AS 21 BROKEN SOURCES -- TWO OF THE THREE THINGS THAT CENSUS MEASURED ARE DEFECTS IN OUR DATA AND OUR PROBE, NOT IN THE HOSTS. ONE: the registry's base-url field is a BASE and the probe treats it as a TARGET. api.github.com answers 200 bare while huggingface.co/api returns 404 because the real call needs a sub-path, so six NOT-FOUND rows and much of the OTHER-STATUS class are the probe asking a url nobody serves rather than a source we cannot reach. The registry owes a per-row PROBE PATH column and until it has one this census measures url shape as much as reach. TWO: codeberg came back OTHER-STATUS and transport_fail is ZERO across all forty, so the reported codeberg-fails-our-TLS class DID NOT REPRODUCE in this run; it is recorded as unconfirmed here rather than repeated. ★★★★★★A CENSUS THAT SUMS IS NOT THE SAME AS A CENSUS THAT MEANS WHAT ITS LABEL SAYS, and the first duty of a new instrument's first run is to say which of its findings are about the subject and which are about itself.
2026-09-06	RR2	lesson	THREE DEFECTS IN THE ORGAN, FOUND BY RUNNING IT AND NAMED RATHER THAN LEFT FOR THE NEXT READER. ONE, the request ceiling doubles as the probe limit, so there is no way to say probe five of forty -- they must become separate arguments. TWO, the budget and the probe disagree on the wire: at one request per source the budget correctly prints wall 0 because a single request needs no inter-request wait, while sr_probe sleeps pace_ms after EVERY request, so the model and the behaviour describe different things. THREE, the per-row lines print only after the whole sweep returns, so an interrupted census yields nothing -- the all-or-nothing shape this estate has already paid for elsewhere. Each is a rung, not a rewrite.
2026-09-06	RR2	land	ALL THREE SELF-REPORTED DEFECTS FIXED AND VERIFIED LIVE, PLUS THE PROBE-PATH FIELD THE FIRST CENSUS ASKED FOR. nx_srcreach PROMOTED 0f3b4759 at 50872 B. ONE, the probe limit is now its own argument: probe conf 1 100 5 reports probed=5 max_probe=5 against ceiling=100, which was impossible when one number served both questions. TWO, the budget models the behaviour that exists rather than the one a formula suggests: reqs x pace, not (reqs-1) x pace, because sr_probe sleeps after EVERY request including the last to a host -- at one request per source the old model printed wall 0 while the run slept 140000 ms, and it now prints longest_source_wall_ms=10000 sequential_wall_ms=140000, which is the cost that will actually be paid. THREE, every row prints AS IT IS DECIDED rather than after the sweep returns, so an interrupted census keeps what it found. FOUR, the optional tenth field probe_path is honoured, and it is PROVEN by outcome and not by inspection: huggingface read NOT-FOUND against its bare base, and with /models?limit=1 the same probe reads REACHABLE -- probed=2 reachable=2 not_found=0. THE DIAGNOSIS THAT THE BASE-URL WAS THE DEFECT RATHER THAN THE SOURCE IS THEREFORE CONFIRMED BY EXPERIMENT, not asserted. The partition still sums to the population in every run.
2026-09-06	RR13	land	RR13 CLOSED: rd_same EXISTS AND IS LIVE (nx_refdrift promoted 4a87c99c, 65360 B). The discovery that made it correct is that the byte comparator was never missing -- rp_same in nx_rowparse_lib already was it, so writing a second would have been the duplicate-ruler defect. What had no name was the FOUR-STATE DECISION wrapping it. Extracted without widening: the caller still performs both fetches and both slurps under exactly its prior guards. NEUTRALITY PROVEN RATHER THAN ASSERTED -- nx_contentdiff live vs candidate reports runs_scanned=260 lost_from_live=0 lost_permil=0 symmetry=ONE-WAY-GAIN verdict=GREEN, a strict superset. nx_domain_admit now reports ungrounded=0 on deepresearch, and that board's row names the real symbol again with its exceed left at 0 for the owning lane to re-assert. ★★★★★THE MISSING ENTRY POINT WAS NOT THE MISSING CAPABILITY: when a contracted symbol is absent, ask whether the primitive already exists under another name before writing a second one.
2026-09-06	RR1	land	THE LIAR GATE, DONE PROPERLY, AND A PRECISION ERROR IN MY OWN CLAIM CORRECTED FIRST. I had described nx_filehash_gate as bite-proven when what it carries is an INTERNAL gv_bite tooth (fires-on-bad, silent-on-good). That is not the same proof as nx_gate_bite, the estate's MUTATION tester, and conflating them is exactly the kind of overclaim this board exists to refuse. nx_srcreach_gate was therefore built and put through the real thing: nx_gate_bite over runtime/nx_srcreg_lib.nx reports baseline green=1, a COMPILING mutant KILLED at cmp site 0, restore verified BYTE-IDENTICAL to pristine at 11347 bytes, twin RESTORED with mutant_live=0, valid_mutants=1 killed=1 not_reached=0, VERDICT GREEN -- non-vacuity PROVEN rather than asserted. The gate itself is 31/31 GREEN with every planted class caught by name, a positive control proving it is not a refuse-everything guard, both closed vocabularies proven to refuse an unknown token rather than default to the permissive value, and the budget asserted against hand-computed constants through gv_check_eq so the number tested and the number printed are one expression.
2026-09-06	RR1	lesson	MUTATING THE ESTATE'S IDENTITY RULER WAS DELIBERATELY NOT DONE, AND THE REASON IS THE POINT. nx_filehash sits under every content pin in every refs register on this estate, and nx_gate_bite works by replacing the subject and rebuilding. The custody report from the run above shows the harness restores correctly -- twins_checked=1 restored=1 mutant_live=0 -- but one successful restore on a low-stakes subject is not evidence for a load-bearing one during an I/O storm, and a failed restore there breaks every pin check at once. So nx_filehash_gate stands on its internal bite tooth plus two published KATs it carries itself, and that limit is STATED rather than papered over. ★★★★★A PROOF YOU DECLINED TO RUN MUST BE NAMED AS DECLINED, NEVER QUIETLY OMITTED -- the difference between an honest gap and an overclaim is whether the reader is told.
2026-09-06	RR1	land	THE BOARD IS NOW SOTA-CLASS, WHICH IN THIS ESTATE IS A MECHANISM AND NOT AN ADJECTIVE. researchreach.sota landed at 12326 bytes, making it the twelfth sota-class domain against eleven before. EVERY GRADE ROW CARRIES A MEASURED BINDING that the one symbol ruler re-checks on EVERY publish, and the rule that makes it worth having is that a Yes, Best or Part over an ABSENT symbol is the UNBACKED class and REFUSES TO PUBLISH. A claim on that page therefore cannot outlive the capability behind it, and no seat can grade this board green by writing prose. Eight bindings, every one verified present on disk before its row was written and every one verified IN THE ORGAN THAT NAMES IT rather than by a grep across the tree: cid_of, ss_commit, crg_prov_state, rd_same, fs_scan_source, sr_load, sr_pace, sr_probe. THE ABSENT CAPABILITIES ARE GRADED No AND BOUND TO NOTHING ON PURPOSE -- OAI-PMH, registry expansion, DOI dedup, full-text acquisition, layout parsing, hybrid retrieval, agentic compile -- because a No over an absent symbol is the only honest pairing and binding one to an aspirational name is precisely how a watch cell flips on a word instead of a capability.
2026-09-06	RR1	measure	THE SOTA PAGE STATES THE BEHIND CELL IN NUMBERS RATHER THAN IN A FOOTNOTE. Coverage of the scholarly record reads 40 declared sources and ZERO harvested against roughly 250M, 300M, 200M and 180M records for OpenAlex, CORE, Semantic Scholar and OpenAIRE -- rival figures quoted as ORDER-OF-MAGNITUDE READS of their own published descriptions and explicitly NOT re-measured by us, so they are scale and not audited counts. The honest comparison is not 40 against 250 million: it is that we have declared a denominator and they have a corpus. Five rows are graded Best and each is a governance property no aggregator offers -- content-addressed identity, a tamper-evident store, fetch-time provenance, drift over time, and board-driven frontier noticing -- so the page claims a real lead on PROVENANCE while conceding COVERAGE outright.
2026-09-06	RR1	lesson	THE APPEND ABOVE DOUBLE-APPLIED AFTER TWO 503s AND THE DEDUPLICATION IS THIS ROW. Both Outcome-Unknown calls had landed. I checked before re-issuing, using a PAYLOAD TOKEN rather than a size -- the method this estate prescribes precisely because size lags -- and the token count read 0 while the text was already there twice by the time the next read ran. SECOND OCCURRENCE TODAY, and it tightens the rule rather than restating it: FOR AN APPEND, ONLY THE WRITE'S OWN total= FIELD IS AUTHORITATIVE. Not a size, not a token count, not two spaced reads, and not any number of verbs on the reader side -- every one of them shares the lagging read path. The one instrument that does not is the writer's own receipt, and when a 503 destroys that receipt there is no sound pre-check available at all: the only safe shapes are an idempotent verb, a caller-supplied nonce the writer echoes, or an anchor-destroying edit whose re-issue self-adjudicates.
2026-09-15	RR1	land	SOTA BOARD EXPANDED 2026-09-15 FROM FOUR RIVALS TO SEVEN (Crossref, Europe PMC, OpenCitations) so the field pass's own liar-kill bars hold: 8 columns, 20 rows, 3 quantitative rows, every Best evidenced. The two Best rows the pass had named UNEVIDENCED (content-addressed identity, append-only store) carried a symbol verification but no date or gate token; both now name the 2026-09-06 verification and the store's promoted gates. Seven refs added, every mirror pinned by sha256, every live count read from the reply and never from prose.
2026-09-15	RR2	measure	LIVE COUNTS READ FROM THE SOURCES' OWN APIs THROUGH THE SOVEREIGN FETCHER WITH NO KEY: OpenAIRE numFound 389982852 (2233 B), Crossref total-results 186608495 (2472 B), Europe PMC hitCount 5596567 for one term (1156 B), OpenCitations citation-count 200 (37 B), all 2026-09-15; OpenAlex meta.count 325977999 from the 2026-09-06 mirror, which corrects the ~250M cell this board had quoted from prose. CORE and Semantic Scholar stay order-of-magnitude reads and the page says so.
2026-09-15	RR2	lesson	TWO CANDIDATE COLUMNS WERE STUBS WITH A CLEAN 200 AND THE FETCH-TIME READ IS WHAT REFUSED THEM: BASE served an Anubis proof-of-work challenge (7843 B) and Unpaywall a 1421 B Vue shell. A column graded from either would have been eight cells of fiction under a perfectly valid pin. Neither is cited; the mirrors stay on disk as the record of the refusal.
2026-09-23	RR8	queue	INTAKE 1790173548 routed by nx_intake_route (nishi-review-via-codex, epoch 1790173579) from operator-via-codex: Build a sovereign legal-open-access research resolver, callable by MCP and API. Given DOI, title or URL, discover Unpaywall, author, institutional and publisher OA versions without bypassing controls; -- sponsor intake-1790173548 plus2
2026-09-23	RR8	queue	INTAKE 1790173820 routed by nx_intake_route (nishi-review-via-codex, epoch 1790173910) from operator-via-codex: Create a Nishi sovereign research-publication and corpus pipeline. It must preserve original bytes and provenance, validate DOI or arXiv identity/title/authors/version/date/license, retain reading ord -- sponsor intake-1790173820 plus2
2026-09-27	RR16	land	Venue-tag rows and static venue rows landed in knowledge/frontier_sources.conf 2026-09-27 after each url was fetched through nx_research_fetch with status 200 and a provenance row; the attest run that consumes them is the next leg.
2026-09-27	RR19	measure	nx_srcreach load before the landing: rows=40 ok=40 verdict=OK; after the landing the same verb is the gate that the new rows parse.
2026-09-27	RR20	lesson	A SEARCH-ENGINE SUMMARY IS NOT A WITNESS AND THE ESTATE CAUGHT ONE: a search result asserted an NVIDIA hybrid real-time subsurface talk at SIGGRAPH 2026; the mirrored Advances course page and NVIDIA's events page (1010241 bytes) carry the token subsurface zero times, so no bar was raised on it. The researcher's mirror-then-count discipline is what makes the oracle triage safe.
2026-09-27	RR16	measure	FIRST ATTEST OVER GRAPHICS WITH THE VENUE ROWS, 2026-09-27 (run artifact /tmp/fs_attest_graphics.txt, 109 lines at the time of this row): the venue-tag rows answered for the first bars (co:SIGGRAPH 2026 returned an item for procedural generation, the Eurographics 2026 static index 4 items dated 2026-05, the SIGGRAPH Asia 2026 static index 1 item) and then 30 of the first 40 venue-tag queries returned UNREACHABLE, printed as UNOBSERVABLE and never as zero. Seven venue rows times seven bars is 49 arXiv requests in one run with no per-host pace between them, which is arXiv rate limiting honestly reported, not a dead source.
2026-09-27	RR16	queue	The per-source pace budget (RR3) must govern nx_frontier_scan's fan-out per HOST across all rows (one arXiv request per pace interval, whatever row it serves), or the venue rows coalesce into ONE OR-query per seed; until one of those lands, attest reads the venue rows as UNOBSERVABLE after the first bars and no bar is misdated by it.
2026-09-27	RR17	lesson	A STATIC INDEX ROW DATED TO ITS PROCEEDINGS MONTH READS NEWER THAN EVERY CURRENT BAR: the siga2026 row was admitted with month=2026-12 (the December proceedings) while its items are September's conditional acceptances, so attest printed NEWER 2026-12 against September bars. Re-dated to 2026-09, the month the list is served; a second field for the proceedings month is a grammar the conf does not yet carry.
2026-09-27	RR16	land	THE FAN-OUT IS PACED (2026-09-27, same session as the venue rows): nx_frontier_scan gained fs_pace_wait at its ONE fetch site, fs_pace_for (the pace for a url's host read from knowledge/research_sources.conf field pace_ms, the registry RR3 already governs; the largest declared pace for a host the registry does not name; PACE-UNKNOWN announced and no wait when the registry is unreadable), fs_url_host and fs_num_at, with six static module slots (90504 -> 96160 bytes). Sovereign build 126144 bytes sha256 be8a1fb3ba6ce17a050ac07708f3806777a34a57bd94d3ea9182da2aec51ace9; nx_contentdiff lost ONE run, ct_pick, an unreferenced symbol name the organ never calls (0 occurrences in its source), gained the registry path and the newer bar-ruler refusal strings the live binary predated; nx_behaveprobe seeds graphics IDENTICAL (8852 bytes both); PROMOTED with prev banked and _offc refreshed. First measured wait in the paced attest: PACE host=export.arxiv.org pace_ms=3000 elapsed_ms=303 waited_ms=2697, and the venue row that read UNREACHABLE unpaced now answers (co:SIGGRAPH Asia 2026 for the GI bar: Gaussian Light Transport 2609.11430, NEWER than the 2026-07 bar). Scope stated: consecutive requests to one host are paced, which is the measured burst shape; a per-host table for interleaved bursts is the next rung only if one is measured.
2026-09-27	RR16	measure	LIKE FOR LIKE, SAME BOARD, SAME BARS, SAME ROWS, SAME AFTERNOON: the unpaced attest over graphics read 22 of its first 33 venue-tag queries UNREACHABLE (30 of the first 40); the paced attest read 0 of its first 33 UNREACHABLE, every wait announced (PACE host=export.arxiv.org pace_ms=3000, waits of 2486 to 2704 ms after 296 to 514 ms gaps), and the rows that had read UNOBSERVABLE now carry items: co:SIGGRAPH Asia 2026 for the GI bar (Gaussian Light Transport 2609.11430, 2026-09, NEWER than the 2026-07 bar), co:SIGGRAPH 2026 for real-time rendering (4 items, newest 2026-08), co:CVPR 2026 (8 items, newest 2026-07), co:ECCV 2026 (8 items, newest 2026-08). The number the fix is judged by is the unreachable count at equal query count, 22 to 0; wall time rose by the sum of the announced waits, which is the price the registry's own pace declares.
2026-09-27	RR16	measure	paced attest run complete 2026-09-27 (nx_frontier_scan attest graphics, promoted be8a1fb3): bars=7 fetches=112 venue-tag fan-out rows=49 verdict=UNREACHABLE rows=0 newer_sources=8 stale_bars=1; every consecutive export.arxiv.org request printed PACE pace_ms=3000 waited_ms 2575 to 2718; the matched unpaced run read 30 of its first 40 venue-tag rows UNREACHABLE. The paced reach surfaced graphics SB-RTGI STALE (Gaussian Light Transport 2609.11430 under the SIGGRAPH Asia 2026 tag) that the throttled run could not raise
2026-09-27	RR16	lesson	an UNREACHABLE fan-out row is an abstention that acquits: nx_barfresh cannot read STALE from a listing the scanner never received, so a throttled researcher keeps every bar looking fresh. Pacing is therefore a freshness instrument, not a courtesy to the host; count UNREACHABLE rows beside stale_bars on every attest
2026-09-27	RR16	land	the fleet refs referee nx_compare_refs_gate names its offenders now (promoted 2026-09-27, 165376 B sha256 3e93bce3, prev banked 140080 B sha256 242e51a8): dupkey= and missing= were counters with no worklist row in crg_scan, so a duplicate key (softbody arxiv_softbody_202609, this lane's own inherited row) was found only because its UNCITED sibling named it; the scan prints DUP-KEY file= key= and MIRROR-MISSING file= key= mirror= beside the counts. Content ruler: lost 1 (sys_openat_append, the base-class symbol the matched-vintage control nx_vasc_gate lost through the same lane today), gained the fio and framed-append closures the served binary predated; promoted under allow_capability_loss with that run named. Acceptance is the next fleet run printing the one missing mirror by register and key
2026-09-27	RR16	lesson	a referee that counts a defect class without naming a row makes every fix a search; the two counters in the estate's own citation referee had been printed for weeks with nobody able to act on them. When a counter is added, add its row in the same edit
2026-09-27	RR16	measure	acceptance of the promoted refs referee, fleet run 2026-09-27 (38/44, verdict unchanged, offenders now named): MIRROR-MISSING file=buildroot/knowledge/compare/motion-rigmodels-20260919.refs key=rigmodels_viewer_20260919 mirror=knowledge/compare/motion-rigmodels-acceptance-20260919.json (a 09-19 motion row citing a mirror outside knowledge/fetched -- for that row's owner); DUP-KEY fires on no live register (dupkey=0) and both rows fire on the gate's own scan-defects fixture bad.refs key=gamma, so the bite came free from the fixture the gate already planted
2026-09-27	RR22	measure	the two homonyms named by debt 1789839847, read from the mirrored bytes through nx_arxiv_digest 2026-09-27: 2609.16215 (the KV-cache placement paper in the all:infinigen listing) carries category cs.AI, its siblings 2604.05012 cs.AR,cs.AI and 2406.19707 cs.LG,cs.DC; 2512.11025 (the phrase wet hair) carries hep-th,gr-qc,hep-ph. The controls: SceneForge 2605.14399 cs.CV,cs.GR and Infinigen-Articulated 2505.10755 cs.RO,cs.GR on the same listing; 2401.15628 (wet porous materials rendering) cs.GR on the wet listing. Subject tokens cannot separate a homonym that shares the query token (every entry above contains infinigen or wet hair), the listing's own taxonomy separates all of them, so the classifier's axis is category terms against a per-board set declared as data in knowledge/subject_classes.conf (9 boards seeded; any-of, not primary-only, because Infinigen-Articulated is primary cs.RO)
2026-09-27	RR22	land	classifier BUILT AND PROMOTED 2026-09-27: nx_frontier_scan 133344 B sha256 ff32fccd (prev 126144 B sha256 be8a1fb3 banked in the prev slot), six anchored CAS edits (fs_cls_load, fs_cls_has, fs_cls_admit, fs_w_class; fs_newest_month admits only ON or UNCLASSIFIED items into its max and keeps the any-item max as off_subject_newer; the ATTEST branch refuses a homonym-only listing by name; fs_attest_domain prints CLASSES domain= allowed= or UNDECLARED), knowledge/subject_classes.conf seeded for 9 boards; contentdiff GREEN one-way gain (lost 0, gained 26 runs), behaveprobe seeds procgen IDENTICAL 4872 B, nx_frontier_scan_gate 27/27 GREEN, _offc refreshed. Rung symbol and matrix row repointed from the digester's ad_subject_classify to the attester's fs_cls_admit (the DARK-CONTRACT law: name the real entry point)
2026-09-27	RR22	measure	DONE-RULE MET 2026-09-27 on live bytes with the promoted scanner: (1) procgen SB-PROCGEN's own listing read on=6 off=2, the KV-cache homonyms 2609.16215 (cs.AI) and 2604.05012 (cs.AR,cs.AI) named OFF-SUBJECT, the September month carried by AgriGen 2609.25725 (cs.RO,cs.GR) and GAE 2609.24981 (cs.CV); (2) the mirrored wet listing knowledge/fetched/cmp_skintwin_arxiv_wet_202609.xml through monthprobe cls=cs.GR,cs.CV,physics.optics,physics.med-ph,eess.IV read items=11 dated=11 newest=2025-04 on=2 off=9 unclassified=0 off_subject_newer=2025-12, the quantum-gravity Wet Hair paper 2512.11025 (hep-th,gr-qc,hep-ph) named OFF and the wet porous rendering paper 2401.15628 (cs.GR) ON; the skintwin attest fan-out named 8 OFF rows and the procgen attest 4. The bite is permanent in the referee: nx_frontier_scan_gate gained five classprobe teeth (a planted cs.AI homonym as the newest fixture item, an on-subject May item, an undated on-subject item, a dated item with no category; newest=2026-06 not 2026-07, OFF row named, off_subject_newer visible, and the neg-control: undeclared classes take the homonym month and say undeclared), 27/27 -> 32/32 GREEN; scanner re-promoted 133824 B sha256 0c4996d4 (monthprobe takes cls= on its own argv so the referee shares no production conf), gate promoted 73880 B sha256 63fceb5e with the same three base-class symbols the day's matched-vintage control lost. Debt 1789839847 is answered by mechanism and left open for its owner's adjudication
2026-09-27	RR22	lesson	classify by the listing's own taxonomy before taking a max, and declare the allowed classes as data per board: a token axis cannot separate a homonym that shares the query token, a threshold axis is a magic number, a taxonomy axis is derivable and the calibration (which arXiv terms a board lives in) is a conf row with its witnesses. The any-of rule is itself calibrated: a primary-only rule would have excluded a real procedural generator (Infinigen-Articulated, primary cs.RO)
2026-09-28	RR16	land	HOST-BORROWED GRAMMAR 2026-09-28: a bar whose ref url is a query the source templates do not spell (abs: instead of all:, a parenthesised OR, a venue tag) read ATTEST-UNATTESTABLE no-dated-source-row-knows-this-listing although its host is a dated source whose item and date grammar are exactly what that listing answers in; nx_frontier_scan now lends the first host-matching row's grammar when no template prefix matches and announces ATTEST-GRAMMAR borrowed-by-host (four anchored CAS edits; build 134968 B sha256 a691fd96; contentdiff one-way gain of the two announcement strings; behaveprobe seeds IDENTICAL; promoted; gate 32/32; _offc refreshed). Live: skintwin 7 bars, 5 attested by the organ (the NIST and MetaHuman html refs correctly not), barscans appended with on-subject months; SB-FABRIC re-dated to 2026-03 and SB-WET to 2025-04 because the classifier now excludes the June cond-mat yarn paper and the December quantum-gravity homonym the 09-27 hand rows had carried
2026-09-28	RR16	lesson	an attester that matches a citation by template PREFIX can only attest the queries it was taught to spell; the grammar a listing answers in is a property of its HOST, so match the host and lend the grammar, and say so on the row. Until this landed every custom-query arXiv bar in the estate would have read UNATTESTED on the first of October
2026-09-28	RR16	land	the citation referee no longer convicts a binary mirror on an HTML axis (2026-09-28): a redirected PDF or ZIP-based office mirror read CANON-HEAD-TRUNCATED (no head closes in a PDF) and joined the open-redirect ratchet forever; the Jensen 1999 mirror raised that ratchet 20 -> 21 the hour it landed. nx_compare_refs_gate now recognises the PDF and ZIP magics in crg_is_binary_doc, crg_canon_of_mirror returns a fourth negative class, the row is censused as CANON-NOT-APPLICABLE-BINARY in its own partition bucket (pv[21], summed into the redirected partition, never open), the summary prints not_applicable_binary=, and a neg-control fixture assembled byte by byte proves the class fires (promoted 167336 B sha256 c14ed0da, contentdiff one-way gain, prev 3e93bce3 banked). Its pin is the identity proof a canonical would have been
2026-09-28	RR16	measure	publisher blocked 2026-09-28: the scoped regen refused every domain with FORK-STALE (buildroot/_offc/nx_swcompare_matrix.elf af966543 1283352 B against the promoted 603de9cd 1341712 B shipped by another seat at 1790603620 per organ_ship.jrnl); the fork was refreshed by copying the promoted serving-root binary (nx_filecopy, 1341712 B verified byte-for-byte), the documented remedy for a buildroot/_offc generator left behind by a promote. The ship loop's OFFC-TWIN stage refreshes nishihost/_offc, not buildroot/_offc: the third root the publisher forks, still owed a stage of its own
2026-09-28	RR16	measure	fleet refs referee on the promoted binary 2026-09-28 (167376 B sha256 fddf4815): 39/45, up from 38/44 before the binary class; the Jensen 1999 mirror reads CANON-NOT-APPLICABLE-BINARY, not_applicable_binary=3 (the swarm verma2015.pdf and the automation osworld results .xlsx were the same misreading), canon_unknown 19 -> 16, the open-redirect set 21 -> 18 with both identity teeth (open equals moved plus canon-unknown; worklist rows equal the open count) PASSING again after the first ship had broken them by excluding the class from the worklist but not from the count (the OPEN count lived in a second expression, redirected minus benign). Residue, all other owners: 1 missing mirror (motion-rigmodels 09-19), 6 pins on beat-rewritten library files, 45 uncited keys in nine domains, 2 provenance mismatches (datavis, loadgov), the open-redirect ratchet at 18 against a baseline of 14, the moved ratchet without an attribution snapshot
2026-09-28	RR16	land	THE THIRD ROOT HAS ITS STAGE (2026-09-28): nx_offc_install <target> promoted now also refreshes buildroot/_offc/<target>.elf from the PROMOTED root binary whenever that mirror already exists (oi_refresh_buildroot, the one atomic install oi_install_from_to parameterised by destination; a mirror is maintained, never created), and announces buildroot_offc=REFRESHED or ABSENT or FAILED on the same line; the ship loop's OFFC-TWIN stage forks this installer, so every future generator promote refreshes the publisher's fork without a hand copy. Promoted 37616 B sha256 c53a1539 under allow_capability_loss for the one rewritten success sentence (the served binary carried no source paths; the candidate gains them); witnessed live: buildroot_offc=ABSENT on nx_kubelka_gate, REFRESHED on nx_swcompare_matrix; the FAILED state is named, not witnessed
2026-09-28	RR16	land	THE CITATION REFEREE'S WORKLIST HAS REPEATABLE REMEDIES (2026-09-28, operator: repeatable capabilities, never hand edits): nx_compare_cite grew mark (UNCITED, label prefix or rung:ID or auto from the grounds field), markmany (a whole board's uncited keys in one call, all-or-none, one publish per touched file), labels, repin (PIN-MISMATCH: the row's own url re-fetched through the researcher into an immutable knowledge/fetched mirror, a redirect refused), reprov (PROV-MISMATCH: the url the fetch journal holds), canon and recanon (OPEN-REDIRECT: re-fetched at the declared canonical with zero hops); one row matcher shared with the legacy path; gate nx_compare_cite_gate 11 to 24 teeth with planted refusals, promoted 3e870fc3 / 9f5878ab; the gate caught the first markmany parser on its live run (a loop-exit sentinel in the cursor, the estate's own law) and the fix shipped inside the hour. Live on the fleet referee's rows: 2 repins, 2 reprovs, 1 recanon, 45 UNCITED marks across 9 boards, the motion-competitor side register merged into motion.refs and retired
2026-09-28	RR9	measure	MEASURED 2026-09-28 ON A MODERN ARXIV PDF: nx_pdf_text over Large-scale Codec Avatars (2604.02320v2, 32064633 B, knowledge/fetched/cmp_twin_lca_2604_02320v2.pdf) wrote about 66 KB of body text with kerning-split words (multi-vie wstudio, achie vehigh) and ligatures lost as replacement bytes, then about 11.3 MB of binary stream content: nx_txtscan ctl read ctl_runs 961750 ctl_bytes 2018613 verdict CORRUPT. The organ's own LIMITS line names the causes (no object streams, no xref streams, no glyph-position layout). RR23 cannot land on a reader that cannot read the paper, and the twiningest board's studio camera counts stay UNREAD for the same reason.
2026-09-28	RR21	measure	THE DIGESTER CUTS INSIDE A CODEPOINT: nx_arxiv_digest truncates each abstract head at abstract_bytes without backing off to a UTF-8 boundary, so a 600-byte digest of the body-shape listing is refused by nx_fs as malformed at byte 12385 (inside a math abstract carrying multi-byte symbols) and the estate's own reader cannot read the estate's own digest. Remedy: back the cut off over continuation bytes inside ad_digest, with a gate tooth on a fixture whose cap lands mid-codepoint.
2026-09-29	RR21	land	SHIPPED 2026-09-28: nx_arxiv_digest's abstract cap no longer cuts inside a UTF-8 codepoint (ad_utf8_floor backs off over continuation bytes to the last complete codepoint); nx_arxiv_digest_gate 44/44 GREEN with the floor's known answers, a cut inside and after a two-byte codepoint and a neg-control that cuts by plain bytes; gate bc66b1e1 and CLI 7d4598f2 promoted; verified on the real body-shape listing, whose 43,680-byte digest now reads end to end.
2026-09-29	RR9	measure	THE OPEN FORM BEAT THE PDF ON THE SAME PAPER, 2026-09-28: Large-scale Codec Avatars (2604.02320) read through nx_pdf_text gave about 66 KB of text with split words and lost ligatures followed by about 11.3 MB of undecoded stream bytes (verdict CORRUPT); the same paper's arXiv HTML read through nx_dr_extract gave 73 KB of clean text and recovered the capture facts the PDF route lost (200 calibrated synchronised 4K cameras, about 5,000 frames per subject, 2,737 studio participants). The HTML carries each equation twice (the rendering and its TeX annotation), which RR25 must collapse to one copy. RR9 stays the fallback for papers with no open form.
2026-09-29	RR16	measure	REACH GAP 2026-09-28: https://civitai.red/ answered the researcher 403 and nothing was saved (RF-DEFINITIVE, not retried); the sibling hubs answered 200 and are pinned on twiningest (the civitai.com public API listing and the civitaiarchive.com index). A 403 from a hub the operator named is a wall to classify (login, region or bot challenge), never evidence that the site is absent.
2026-09-29	RR24	queue	The Europe PMC dated source row landed in knowledge/frontier_sources.conf (src europepmc, ISO firstPublicationDate), appended the same session. The attest proof is owed on a board whose clinical bar cites a Europe PMC query, and the mirrored Europe PMC listing's first record (an organoid hair-regeneration paper) is off-subject for aesthetic 3D imaging, so a biomedical subject classifier must stand in front of its newest month before that month can be trusted (RR22's arXiv categories do not apply to PubMed records).
2026-10-01	RR2	queue	ATTEST VERDICT ACQUITS ON FULL ABSTENTION, measured 2026-10-01 and confirmed in source: the ATTEST-DONE summary in buildroot/runtime/nx_frontier_scan.nx sets the verdict from stale_bars alone (verdict=ATTESTED, exit 0, whenever stale_bars is 0), so the twiningest attest that morning read ATTESTED with appended 0 and unattestable 11 of 11 bars, every arXiv own-ref leg ATTEST-UNATTESTABLE reason=source-unreachable. nx_barfresh was not fooled (it reads barscan rows and the board stayed UNATTESTED until a seat read the listings), but any caller that branches on the organ's exit or verdict line is. Remedy, not built in this lane: a third state, verdict=UNOBSERVABLE with exit 3 when appended plus already is 0 and unattestable is above 0, PARTIAL when only some bars are unattestable, and a planted all-unreachable fixture as the gate tooth. Same morning, the reach fact behind it: export.arxiv.org answered HTTP 429 to the NAS on every attempt for hours and arxiv.org search dropped the connection intermittently, while the OpenAlex works API (filter primary_location.source.id S4306400194, the arXiv source, with title_and_abstract.search and sort by publication_date descending) answered every request; it indexes arXiv about a week behind, enough for a month-granular bar, and it read every arXiv bar this seat attested on aesthetictwin and twiningest (mirrors rf_0fd48a64, rf_1c19978b, rf_0423ce9b and nine more under knowledge/fetched/). Candidate: OpenAlex as a second key-free dated source row for arXiv-hosted bars, announced as borrowed, with its indexing lag printed beside the newest month.
2026-10-01	RR24	queue	A CLINICAL BAR NOW WAITS ON THIS RUNG: aesthetictwin SB-SURGOUTCOME-2609 (predicting the surgical outcome from a 3D face or body twin, landed 2026-10-01) cites the Europe PMC listing surg_europepmc_listing_202610 (knowledge/fetched/rf_570050b4.raw), because PubMed HTML answered nx_research_fetch with a bot wall while NIH E-utilities efetch text and the Europe PMC REST search both answered; twiningest SB-TI-CLINIC-2608 was read the same way. Until fs_pubmed_attest exists every reading of those two bars is a seat reading. First live subjects for its done-rule: attest aesthetictwin prints SB-SURGOUTCOME-2609 ATTEST newest 2026-09 from the Europe PMC row, and attest twiningest prints SB-TI-CLINIC-2608 newest 2026-08.
2026-10-01	RR22	queue	A CATEGORY CLASSIFIER CANNOT SEPARATE THIS HOMONYM CLASS: the aesthetictwin face-anatomy listing arxiv_faceanat_202609, read 2026-10-01 through OpenAlex, had September and August rows that are homonyms on the word face, some inside cs.CV itself (SignDino sign language, ophthalmic pretraining, sheep facial pain) beside off-category ones (DITTO teleoperation, dental CBCT, a medical-imaging privacy audit); the newest on-subject first submissions are TongueReenact 2607.28039 (2026-07-30) and GNM Head 2607.23687 (first submitted 2026-07-26, v3 2026-09-11). The seat re-dated SB-FACEANAT-2609 and MB8 to 2026-07. aesthetictwin has no cls row in knowledge/subject_classes.conf, and a category row would still admit the cs.CV homonyms, so the remedy is the ref query (require a head-model, skull or musculoskeletal term) plus a check of which Atom date the arXiv row reads (published versus updated, since a v3 must not read as a new entry). Until then the next attest of these refs can write newest 2026-09 and convict both re-dated bars STALE by homonym.
2026-10-01	RR16	queue	BARFRESH 2026-10-01 (seat residue-fleet, T0 1790856940): STALE before and after, 4 of 4 bars UNATTESTED; this seat wrote NO barscan here and names two ABSTENTIONS. RB1 (openalex_api): help.openalex.org re-fetched over the sovereign fetcher (mirror knowledge/fetched/cmp_researchreach_openalex_help_202610.html, 20201 B, sha256 b542cdca68571d1a, changed since the 6 Sep pin) carries no dated mark at all (zero occurrences of the substring 202), so a month cannot be attested from the bar's own ref; the bar needs a dated ref (a release or changelog listing) before any barscan is honest. SB-VENUEREACH-2609 (kesen_sig2026): the SIGGRAPH 2026 index re-fetched byte-identical (288392 B, sha256 7e6c166c5ae0fca3), a static July 2026 listing whose newest month is 2026-07 while the bar is dated 2026-09 by its measurement day; attesting newest=2026-07 would not match the bar, and re-measuring its eight sources needs arXiv fetches this seat does not make. RB2 (agentkg2026) and RB3 (foxden2026) are arxiv.org abstract pages, left for the arXiv attest runner.
2026-10-03	RR26	land	SHIPPED 2026-10-03: nx_licclass_lib (new, buildroot/runtime/nx_licclass_lib.nx) with its rule as data in knowledge/license_classes.conf (48 pattern rows strictest first, 9 verdict rows, the OpenOximetry licence text quoted as the RESTRICTED-RESEARCH-ONLY grounds); nx_licclass_gate 35/35 GREEN, promoted 4780550c and registered, known answers on the live conf for Zenodo, Europe PMC, DataCite, Crossref (JSON-escaped url folded), PhysioNet licence name AND licence text REFUSED, an all-rights-reserved page PRIVATE-REFERENCE, empty and custom-terms evidence UNKNOWN, and three fail-closed neg-controls each beside a positive control; the hunt verb in nx_frontier_scan promoted 91c96e37 through the ship loop (gate GREEN, behaviour probe IDENTICAL on seeds, two lost runs adjudicated as the shared syscall shim's reworded mmap refusal and approved by evidence aa32be91); 10 hunt rows and 3 src rows (physionet, wikimedia-commons, datagov) admitted to knowledge/research_sources.conf, nx_srcreach load 65 of 65 OK with hunt rows ignored by the src reader; hunt_cover_permil 500 added to knowledge/frontier_scan.conf with its derivation
2026-10-03	RR26	measure	FIRST LIVE HUNTS 2026-10-03 (skintwin): reflectance (skin reflectance spectral) 10 of 10 sources reachable, 1102 records, 198 admitted, 897 irrelevant, 7 unpaired, ACQUIRE 114, CALIBRATE-ONLY 1, PRIVATE-REFERENCE 17, UNKNOWN 65, REFUSED 1, degraded 0, partition_sums 1; colour (skin color colorimetry) 949 records, 79 admitted, ACQUIRE 20, PRIVATE-REFERENCE 9, UNKNOWN 45, REFUSED 5 -- the four openox-repo versions (pattern restricted health data license, from PhysioNet's license.name) and ENCoDE (credentialed). Three method classes agree on the precedent: the organ's verdict from the API field, the gate's verdict from the licence TEXT mirror, and the seat's hand adjudication of the same day (skintwin.plan log 1791040900). The hunt named the International Skin Spectra Archive (figshare 28228571, CC BY 4.0 on the figshare record and the DataCite rightsList), mirrored for skintwin
2026-10-03	RR26	lesson	A LICENCE TEXT CAN NAME A STRICTER AND A LOOSER CLASS AT ONCE, AND THE ORDER OF THE RULE ROWS DECIDES WHICH DUTY SURVIVES: the CC BY 4.0 legal code carries public domain and CC0 once, in its closing notice that dedicates the licence text itself to the public domain, so with the public-domain rows first a CC BY text classed PD-CC0 and its attribution duty vanished. Found by reading the mirrored text before trusting the first gate GREEN; attribution rows now run first and a gate tooth holds the real text. Second lesson, same hour: Zenodo answered HTTP 400 to size=100 from an anonymous caller (UNREACHABLE on the first run, never an empty result); the row now asks size 25
2026-10-03	RR26	queue	OPEN: (1) a licence-class store does not exist yet (knowledge/licstore/ absent): acquired bytes sit in knowledge/fetched with the store class on the worklist row and the refs row; owed, a store index per class and the move. (2) The hunt shares one fetch receipt path (/tmp/nx_frontier_scan/hunt.out), so two concurrent hunts race on it; derive the path from domain and tag. (3) Relevance is token coverage and cannot tell human skin from lizard or frog skin (both ranked ACQUIRE on the reflectance run): a subject classifier for the biomedical record is owed, as RR22 is for arXiv. (4) The data.gov CKAN API answered 404 (api/3 and api/action) and the GSA mirror 503; its HTML page carries no licence, so every data.gov hit reads UNKNOWN by construction. (5) A figshare COLLECTION record carries no licence field (University of Minho hyperspectral faces, c.7163569): the hunt must follow to the member articles. (6) UNKNOWN rows (65 on the reflectance run, the NIST human skin reflectance set among them) need a licence-page read lane: fetch the rightsUri, class the text, append the verdict
2026-10-03	RR26	land	IDEA LEVEL SHIPPED 2026-10-03 (operator, verbatim: "we also have idea level that is just used for idea but doesnt require attribution"): a USE LEVEL orthogonal to the licence class, as data in knowledge/license_classes.conf (sha256 e4ad5d8e, 9377 B): U rows level
2026-10-03	RR26	queue	OPEN for the idea level: (1) nx_frontier_scan hunt does not yet print the IDEA-ONLY verdict beside the acquisition verdict on its worklist rows (another lane owns that file today); (2) CONTRACT-BOUND detection is phrase-based and cannot see a click-through gate a page enforces without naming it; a page whose access flow requires acceptance must be read by hand until a flow probe exists; (3) CONTRACTED is allowed at IDEA-ONLY on the reading that our own two-party contract grants use -- a contract that restricts idea use would need its own B row
2026-10-03	RR26	retract	THE 1102 RECORDS OF THE FIRST LIVE HUNT (log 1791043531) WERE A SAMPLE, NOT A POPULATION: three verifiers measured on 2026-10-03 that every hunt row fetched ONE page of a fixed size and never read the source total (Europe PMC rf_4208d023.raw hitCount 308 with a nextCursorMark, 100 read; Zenodo rf_522c046e.raw total 85569, 25 read; DataCite 42 read whole), HUNT-DONE printed no per-source total or capped flag, data.gov answered an off-query popular-datasets page and was counted reachable, and OSF admitted 0 with no item count. The ranked worklist of that run stands as proposals; its record count is withdrawn. The discriminating experiment is the paged re-run below.
2026-10-03	RR26	land	HUNT PAGING SHIPPED 2026-10-03: nx_frontier_scan a4dbd86d (189960 B, /api/promote with expect_sha256 after nx_contentdiff named its one lost run, the renamed receipt path; _offc twin installed) reads a huntpage row per source from knowledge/research_sources.conf (total marker, cursor/page/offset/none, page arg, next-token span, echo span; 10 rows landed), pages up to hunt_page_budget (THRESHOLD 10 in knowledge/frontier_scan.conf with its calibration), percent-encodes cursors, prints SOURCE-TOTAL fetched/total/pages/budget/capped/stop/paging/echo_permil/answer per source, classes a source OFF-QUERY when its records all miss the subject and it does not echo the query (counted unanswered, records excluded from totals), EMPTY-TOTAL when it declares zero and serves none, and HUNT-DONE now carries answered/off_query/no_items/unreachable with source_partition_sums, records_fetched, records_total_declared, capped_sources and population_complete; receipts are named /tmp/nx_frontier_scan/hunt_<domain>_<tag>.out. Offline twin huntfx reads <dir>/<key>.<page>.raw. nx_frontier_scan_gate 31eb04a1 (ship loop, loss approval 90dba7f9 = two superseded nx_syscalls arena messages, behavior review 9a824c6b) gained 12 hunt teeth on a four-source fixture (cursor to total, page budget binding with a present page 3 that must not be read, off-query neg-control, empty total): 45 of 45 GREEN against the promoted subject, 33 of 45 RED against the pre-change subject (the 12 hunt teeth are the ones that failed). The staged gate .new at 74320 B was a rebuild of the UNCHANGED gate source against the newer nx_syscalls (contentdiff: only the two arena messages lost), not mine; superseded by 31eb04a1. Zenodo join changed to AND (OR answered 85569, AND 26, rf_05694ba2.raw). nx_licclass_gate authorised (cap minted, nonce 1791045120) and exercised 64 of 64 GREEN.
2026-10-03	RR26	measure	PAGED RE-RUN 2026-10-03 skintwin reflectance (skin reflectance spectral), receipt /tmp/rr26_hunt_skintwin_reflectance.out: sources 10 = answered 9 + off_query 1 (datagov, 20 records served, none relevant, no echo) + no_items 0 + unreachable 0, partition sums. Per source fetched/total: europepmc 314/308 over 4 cursor pages (echo 1000), zenodo 26/26 (2 pages), datacite 42/42, figshare 26/26, osf 0/0 EMPTY-TOTAL, dryad 2/2, crossref-peerj 123/123 (2 offset pages), physionet 731 static whole catalogue, wikimedia-commons 500 over 10 pages CAPPED (publishes no total). records_fetched 1764, records_total_declared 527, admitted 378, acquire 126, private_reference 24, unknown 227, refused 1, population_complete 0 (wikimedia alone). Three method classes: the gate fixture of known answers (45 of 45), the totals checked against independent mirrors fetched before the change (Europe PMC 308, DataCite 42, Zenodo AND 26 from a separate probe), and the sums reconciled by hand (527 = 308+26+42+26+0+2+123; 1764 = 314+26+42+26+0+2+731+500+123).
2026-10-03	RR26	lesson	ONE PAGE READ WHOLE IS STILL A SAMPLE OF ITS SOURCE: reading every item of the response bounded the count by the response, never by the source, and the header comment said nothing was capped in silence while every source was capped. The source total is DATA beside the row, and a source that serves records none of which match and does not echo the query answered a different question, so it is unanswered, never reachable. And a gate that forks the live subject cannot prove a new verb before promote: the subject was promoted on the content ruler first, then the gate shipped through the loop against it.
2026-10-03	RR26	queue	OPEN: (1) europepmc counts 314 items over hitCount 308: the item delimiter matches six nested objects, so fetched can exceed total; name it as delimiter_overcount on SOURCE-TOTAL and tighten the delimiter; (2) wikimedia-commons is capped at 10 pages with no total (CirrusSearch ORs the words); find an AND form or a totalhits field; (3) the idea-level verdict is still not printed on hunt worklist rows (queue 1791045001); (4) the ship loop cannot prove an end-to-end gate against a staged subject (it forks the served binary), so a new verb needs promote-then-gate; a staged-subject mode for the loop is owed.
2026-10-03	RR26	retract	the paged reflectance hunt published records_fetched=1764 and Europe PMC fetched=314; an independent count of one isOpenAccess per result across the four mirrored pages (rf_4208d023 100, rf_46fb4c34 100, rf_1ca6152d 100, rf_f527f9f5 8) gives exactly 308 Europe PMC records, so the true totals are 308 and 1758. The 6 extra are phantom delimiter matches that landed in unpaired. Owed: tighten the Europe PMC item delimiter, a gate tooth that fetched never exceeds a declared total, per-source lines written to the durable worklist (not only /tmp), mirror= naming every page file, and a human-subject relevance classifier (off-subject records still rank ACQUIRE).

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.

SOTA field — 7 peers, 20 axes, measured bindings

Nishi vs OpenAlex, CORE, Semantic Scholar, OpenAIRE, Crossref, Europe PMC and OpenCitations. Every Nishi grade is BOUND to a symbol and re-measured on publish; an unbacked grade refuses to publish. Rival grades are presence reads of pinned mirrors and live replies, September 2026.

How the field is scored. A state-of-the-art comparison across the FULL competitor field: quantitative axes carry measured or published numbers (Nishi’s column is measured on this estate, competitor columns are researcher-sourced and cited in the note); grade axes carry Best / Yes / Part / No for every peer. A Nishi grade may be bound to an organ and a symbol and is then re-measured on every publish by the one symbol ruler the capability matrix above uses; an over-claim refuses to publish.

field, strip order:OpenAlexCORESemantic ScholarOpenAIRECrossrefEurope PMCOpenCitations·BestYesPartNo

Measured bindings. 10 of the Nishi grades on this page are bound to an organ and a symbol and re-measured on every publish by the one symbol ruler; the drift classes are counted below and their sum is re-derived beside the bound total.

Measured bindings, counted
Class	Count	Note
consistent	5	the grade and the symbol agree
stale-under	2	the grade says No while the organ carries the symbol; published with the badge, named by the gate until raised
unbacked	0	a claim over an absent symbol refuses to publish
unresolved	0
unreadable	0
unserved	3	the organ carries the symbol in source while its binary is not on the serving surface: built nowhere, invisible to symbol presence; published with the badge, never refused
partition sum	10	RECONCILED: the parts sum to the total 10

Reach and governance (where we are ahead)

Content-addressed identity for an acquired document

BEST. The datum IS its sha256, so the same paper harvested from a preprint server, an institutional repository and a publisher collapses to ONE object by construction rather than by a dedup pass that can be wrong. All seven rivals key on mutable external identifiers and record where a copy lives -- DOI at Crossref [crossref_live], PMID beside DOI at Europe PMC [epmc_live], DOI, PMID and OMID at OpenCitations [opencitations_docs] -- and none makes the bytes their own name. Symbol verified at nx_canon_cid.nx line 144 on 2026-09-06, the witness being the organ that names it.

bound to runtime/nx_canon_cid.nx:cid_of · measured PRESENT (decl) · UNSERVED — the hand grade claims the capability and the organ carries the symbol in SOURCE, but its binary is not on the serving surface: built nowhere -- this cell over-reports what RUNS and stays flagged until the organ is promoted

BEST

Tamper-evident append-only acquisition store

BEST. History is sacred and acquisition is additive, so a source that later withdraws or rewrites a record cannot silently rewrite ours -- and the withdrawal is itself a fact we hold. This is what makes a coverage measurement reproducible months later, which no hosted aggregator offers because their corpus is mutable underneath the query. Symbol verified at nx_seg_store.nx line 1516 on 2026-09-06; the store's own gates, nx_seg_store_gate and nx_sts_cas_gate, are promoted on the NAS. None of the three columns added on 2026-09-15 documents an append-only history in its mirror.

bound to runtime/nx_seg_store.nx:ss_commit · measured PRESENT (decl) · UNSERVED — the hand grade claims the capability and the organ carries the symbol in SOURCE, but its binary is not on the serving surface: built nowhere -- this cell over-reports what RUNS and stays flagged until the organ is promoted

BEST

Fetch-time provenance binding a mirror to the url it claims

BEST, and it earned itself on this very board within the hour it was written. Three states because two would lie: PROVEN, MISMATCH and UNPROVEN. Of the four rival documentation mirrors fetched for this domain, TWO were non-content pages returning a clean 200 -- an Apache directory index and a ReDoc JavaScript shell -- and a register without fetch-time binding would have carried both as ordinary citations. It then caught THIS lane's own OpenAlex citation as an open redirect whose canonical was the help-centre home page rather than the API overview it was cited as. Symbol verified at nx_compare_refs_gate.nx line 402. It earned itself again on 2026-09-15: two candidate columns, BASE and Unpaywall, answered with a proof-of-work challenge and a JavaScript shell, both status 200, and both were refused before a cell was written.

bound to runtime/nx_compare_refs_gate.nx:crg_prov_state · measured PRESENT (decl) · CONSISTENT

BEST

Does the source still serve the bytes we pinned

BEST. A pin proves the bytes on OUR disk did not change and says nothing about whether the SOURCE changed, so a citation can rot in silence with every pin check green. Four states, because folding a page that varies against ITSELF into DRIFTED would ship a false-positive generator: MATCH, DRIFTED, DYNAMIC, UNREACHABLE. GROUNDED 2026-09-06: this symbol was UNGROUNDED when this board was admitted -- the capability was real but ran inline in main and its contracted entry point had never been written -- and it was extracted the same day as the four-state classifier, NOT as a byte comparator, because rp_same already was the comparator and a second one would have been a duplicate ruler. Neutrality proven, not asserted: nx_contentdiff lost_from_live=0, ONE-WAY-GAIN.

bound to runtime/nx_refdrift.nx:rd_same · measured PRESENT (decl) · CONSISTENT

BEST

Frontier noticing driven by a board's own data

BEST. Seeds derived from a domain's OWN matrix rows and keyword field, the outside pulled through the sovereign fetcher over sources that are themselves data, findings diffed against our rows to emit gap proposals naming the row each candidate challenges. All seven rivals answer or index for whoever asks and none maintains a capability board to notice a gap in, so there is no comparable surface. ITS MEASURED LIMIT IS THIS BOARD'S REASON TO EXIST: it is wired to exactly two sources today.

bound to runtime/nx_frontier_scan.nx:fs_scan_source · measured PRESENT (decl) · CONSISTENT

BEST

The reach layer (young, live, and measured)

Source registry as data with closed vocabularies

LANDED 2026-09-06. 40 sources, 7 kinds, 6 protocols, a per-host pace and a licence, declaring INTENT ONLY and never status -- a status column in a declaration file is a constant wearing the shape of a measurement. The vocabularies are CLOSED: an unknown protocol REFUSES rather than defaulting to the permissive value, which is the silent-downgrade class this estate has already measured in a routing table. OpenAIRE is graded Part because its own published API index names a dspace-eprints-api for the institutional tier; the other six publish no ingest configuration as a readable contract (the Crossref, Europe PMC and OpenCitations mirrors describe query surfaces, not ingest). The reader itself is mutation-proven (nx_srcreach_gate 31/31, killed=1).

bound to runtime/nx_srcreg_lib.nx:sr_load · measured PRESENT (decl) · UNSERVED — the hand grade claims the capability and the organ carries the symbol in SOURCE, but its binary is not on the serving surface: built nowhere -- this cell over-reports what RUNS and stays flagged until the organ is promoted

YES

Outbound budget printed and obeyed before the first request

BEST, and coded so because none of the seven publishes a client-side request budget at all -- Crossref documents a server-side polite pool [crossref_docs], which is the server's throttle and not a client stating its own count before the first request -- so there is no comparable surface, not a feature we lost. A sweep that cannot state its own request count is a load incident waiting to happen, so the run PRINTS the planned count and REFUSES over the ceiling with nothing fetched. CORRECTED THE DAY IT SHIPPED: the budget first modelled inter-request waits and printed wall 0 while the run actually slept 140 seconds, because the probe sleeps after EVERY request. It now models the behaviour that exists rather than the one a formula suggests. Asserted against hand-computed constants in its gate so the number tested and the number printed are ONE expression.

bound to runtime/nx_srcreach.nx:sr_pace · measured PRESENT (decl) · CONSISTENT

BEST

Reachability measured per source with the blocker NAMED

BEST as a measured absence of a comparable surface: NONE of the seven publishes a per-source reachability verdict, so this is a lead nobody currently holds rather than a race we won. Seven named states over the WHOLE registry with UNPROBED counted as its own, because a source nobody probed must never read as one that passed, and TRANSPORT-FAIL kept separate from every status class, because a host that never completed a handshake said something different from one that answered 404. Each non-reachable row carries the remedy that would close it. FIRST LIVE CENSUS 2026-09-06: probed 40, reachable 19, auth-required 3, not-found 6, other-status 12, transport-fail 0, unprobed 0, partition summing to 40.

bound to runtime/nx_srcreach.nx:sr_probe · measured PRESENT (decl) · CONSISTENT

BEST

Coverage of the scholarly record

THE CELL WHERE WE ARE PLAINLY BEHIND AND THE NUMBER SAYS IT. Ours is a DECLARED registry, not a harvested corpus: 40 sources reachable-tested, zero works ingested on the sovereign side. The OpenAlex, OpenAIRE and Crossref figures are the counts their own APIs returned to the sovereign fetcher on the dates shown [openalex_live] [openaire_live] [crossref_live], re-measurable by anyone with the url; the CORE and Semantic Scholar figures remain order-of-magnitude reads of their published descriptions and are NOT re-measured by us; the Europe PMC figure is the one printed on its own service page [epmc_docs]. The honest comparison is not 40 against 300 million; it is that we have declared a denominator and they have a corpus.

bound to runtime/nx_srcreg_lib.nx:sr_load · measured PRESENT (decl) · STALE-UNDER — the hand grade says No while the organ carries the symbol: this cell under-reports and stays flagged until the grade is raised

OpenAlex 325977999 works (meta.count, live 2026-09-06)CORE ~300M+ metadata records with full textSemantic Scholar ~200M+ papersOpenAIRE 389982852 research products (numFound, live 2026-09-15)Crossref 186608495 works (total-results, live 2026-09-15)Europe PMC over 33 million publications (its own service page, 2026-09-15)OpenCitations n/a (a citation index; the mirror states no work total)

40 declared sources, 0 harvested

Nishi, measured

The harvest and compile pipeline (absent, and bound to nothing)

OAI-PMH incremental harvest with resumption

CORE and OpenAIRE are built on harvesting repositories. Crossref and Europe PMC each list an OAI-PMH service in their own navigation [crossref_docs] [epmc_docs] -- a surface they SERVE, which is not evidence that they harvest by it, so both are No on this row; the OpenCitations mirror names none. Ours is ABSENT on the sovereign side and deliberately bound to no symbol, because a No over an absent symbol is the only honest pairing and binding it to an aspirational name is how a watch cell flips on a word. NOTE THE ESTATE ALREADY HAS THIS IN A FORBIDDEN STACK: nishi-library ships a generic OAI-PMH adapter with resumptionToken cursors covering DOAB, OAPEN, AAAI and NSF-PAR, in Python. The rung is a PORT with parity measured before the legacy path is retired.

Registry-of-registries expansion to the institutional tier

OpenDOAR and ROAR each enumerate institutional repositories publishing an OAI-PMH endpoint, so reach grows by ADMITTING A REGISTRY rather than by pasting a thousand urls. Both registry rows sit in our conf today and nothing walks them. The three columns added on 2026-09-15 document no registry walk in their mirrors. Absent, bound to nothing.

Cross-source deduplication on a stable identifier

All four original rivals dedupe on DOI and it is the core of what an aggregator IS, so all four are Best and this is a genuine behind. Crossref is the registration agency that mints the DOI in the first place [crossref_docs], so identity there is by construction rather than by a dedup pass, graded Best; Europe PMC records carry pmid and doi side by side in the live reply [epmc_live] and its mirror documents no dedup, graded Part; OpenCitations resolves DOI, PMID and its own OMID onto one citation record [opencitations_docs], graded Yes. Without a resolvable identifier the same paper harvested three ways is three records, and any coverage number measured over undeduplicated records is inflated by construction. Absent, bound to nothing.

Open-access full text acquired, not merely indexed

CORE is Best because full text is the axis it is built on, and Europe PMC stands beside it: 10.2 million full-text articles and a fullTextXML operation on its own service page [epmc_docs]. Crossref carries text-mining LINKS in the live record [crossref_live] and acquires no text; OpenCitations holds citations, not documents. A record you can count but not read contributes nothing to a reader or a reranker, which is why this is graded separately from coverage. Absent, bound to nothing.

Layout-aware parsing: a PDF keeps its maths and tables

The estate has nx_pdf_text, which recovers a text STREAM and not a STRUCTURE, so equations and tables arrive as flattened prose and a reader cannot tell a caption from a cell. That is not this capability and is deliberately not graded as it. Absent, bound to nothing.

Hybrid retrieval: a knowledge graph and a vector index on one query

OpenAIRE is Best on its own published graph API; OpenAlex and Semantic Scholar publish entity graphs; OpenCitations is a citation graph queried by identifier with no vector leg documented [opencitations_docs], graded Part; the Crossref and Europe PMC mirrors make no graph claim. The 2026 frontier states the architecture directly -- models interpret, embeddings organize, graphs emerge -- and our substrate for BOTH halves already exists separately, so the contract is the JOIN and not either leg. Absent, bound to nothing.

Agentic compile whose writes a critic can refuse

All seven are indexes and none runs a research agent over its own corpus, so all seven are No and this is a lead nobody holds. OUR HEAD START IS UNUSUAL AND IS THE STRONGEST THING WE BRING: nxr_assess already grades every numeric claim against its cited source and GATES PUBLISH, auto-invoked in-run. What the 2026 frontier adds is governance of what the agent writes into MEMORY. Absent on this board, bound to nothing, and the nearest thing we own lives on /compare/deepresearch.

The door: how each source answers the sovereign fetcher (measured 2026-09-15)

Mints the persistent identifier as its registration agency

Nishi mints no resolvable public identifier: our identity is the content address in the first row, which names bytes and resolves nowhere, so No is the honest pairing and this row is bound to nothing. Crossref is the DOI registration agency -- its documentation tree is organised around content registration and its API answers /works/{doi}/agency with the registration agency for any DOI [crossref_docs]. The others index or link identifiers minted elsewhere; OpenAlex and OpenCitations assign internal keys of their own and register no DOI.

Answers an unauthenticated query with status 200

Nishi publishes no research API today, so No. MEASURED through the sovereign fetcher with no key and no account: OpenAlex 200 on 2026-09-06 [openalex_live]; OpenAIRE 200, Crossref 200, Europe PMC 200 and OpenCitations 200 on 2026-09-15 [openaire_live] [crossref_live] [epmc_live] [opencitations_live], the last with an access token encouraged in its documentation and not required [opencitations_docs]. CORE documents a Bearer apikey header on every example call [core_api], graded No. Semantic Scholar is graded No as UNMEASURED: its mirror is a shell that cannot show it, and the host was rate-limited against us on 2026-09-06, so no probe was spent on it.

Bytes that unauthenticated query returned

The transport-level receipt behind the row above, one number per column, each re-derivable from the pinned mirror of that reply. The Europe PMC reply also carried a nextCursorMark and the Crossref reply a total-results field, which is how cursor paging and the corpus count on the coverage row were read rather than assumed.

OpenAlex 17563 (2026-09-06)CORE key requiredSemantic Scholar UNMEASUREDOpenAIRE 2233 (2026-09-15)Crossref 2472 (2026-09-15)Europe PMC 1156 (2026-09-15)OpenCitations 37 (2026-09-15)

no endpoint

Nishi, measured

Sources reachable in the sovereign probe census

MEASURED by the first live census on 2026-09-06 and printed in full on the reachability row above. None of the seven publishes a per-source reachability verdict, so their cells hold n/a rather than a number invented for symmetry, and the partition summing to 40 is the check that our number is a population and not a sample.

bound to runtime/nx_srcreach.nx:sr_probe · measured PRESENT (decl) · STALE-UNDER — the hand grade says No while the organ carries the symbol: this cell under-reports and stays flagged until the grade is raised

OpenAlex n/aCORE n/aSemantic Scholar n/aOpenAIRE n/aCrossref n/aEurope PMC n/aOpenCitations n/a

19 of 40 (auth-required 3, not-found 6, other-status 12, transport-fail 0, unprobed 0)

Nishi, measured

7 competitors20 axes3 quantitativeNishi best 7 · absent 9bound 10 · stale-under 2 · unbacked 0 · unserved 3

Honest verdict. Nishi is not a peer of these seven on coverage and this page says so in the cells rather than in a footnote: they hold hundreds of millions of works and we hold a 40-source registry that was declared on 2026-09-06. Three of their counts were read live from their own APIs by the sovereign fetcher rather than from prose: OpenAIRE 389982852 research products and Crossref 186608495 works on 2026-09-15, OpenAlex 325977999 works on 2026-09-06. What Nishi has that none of them has is provenance the reader can check -- content-addressed identity so a document's name IS its bytes, a tamper-evident append-only store so a source that rewrites history cannot rewrite ours, fetch-time binding of every mirror to the url it claims, and a drift check that asks whether that url still serves those bytes. Those four are graded Best and each is bound to a symbol that was verified in the organ naming it. The reach layer is real but young: the registry reader, the outbound budget and the reachability probe all landed 2026-09-06, all three are LIVE and the budget and classifier are mutation-proven (nx_srcreach_gate 31/31, nx_gate_bite valid_mutants=1 killed=1, restore byte-identical). THE FIRST LIVE CENSUS IS THE HONEST HEADLINE AND IT INDICTS OUR OWN DATA, NOT THE SOURCES: 19 of 40 reachable, and the six NOT-FOUND rows were the probe treating a BASE url as a TARGET -- proven by adding one probe path and watching huggingface go NOT-FOUND to REACHABLE. The harvest half is entirely absent: no OAI-PMH, no registry expansion, no DOI dedup, no full-text acquisition, no parsing, no graph. Those are graded No and bound to nothing, because a No over an absent symbol is the only honest pairing. AND THE ESTATE HAS SOLVED MOST OF THIS ONCE ALREADY, in a stack the sovereignty order forbids in the build path: nishi-library v0.0.102 carries 2.2M works, a 50-plus source registry across nine tiers and a generic OAI-PMH adapter with resumption cursors, in Python over SQLite. The gap is therefore a PORT and a MEASUREMENT, not a greenfield build, and pretending otherwise would be the more flattering story. EXPANDED 2026-09-15 from four rivals to seven so the field pass's own bars hold (8 columns, 20 rows, 3 quantitative rows), and the expansion was measured rather than pasted: two candidate columns, BASE and Unpaywall, answered the fetcher with a proof-of-work challenge page and a JavaScript shell, both status 200, and were refused before a cell was written.

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

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

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

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 30 capabilities: 24 exclusive to it and 6 shared with other boards. Shared rows on this page: 6 = home here 2 + home elsewhere 2 + ambiguous 0 + undecided 2 + refused 0 (partition 6); undecided = unmapped 0 + unproven 2 + unevidenced 0; homonyms 0; since the previous census: new 0, resolved 0. Census FRESH age_s=7207.

Capability	Organ	Boards grading it	Home	Rule	Evidence	Standing on this board
idf_q10	`nx_intlog.nx`	search+, webscraping, deepresearch*+, researchreach+	deepresearch	rung	plan	home elsewhere adopt: `@xrow deepresearch\|idf_q10\|home by rung (fleet duplicate census)`
crg_prov_state	`nx_compare_refs_gate.nx`	deepresearch, researchreach*	researchreach	rung	plan	home here
rd_same	`nx_refdrift.nx`	deepresearch, researchreach*	researchreach	rung	plan	home here
ss_commit	`nx_seg_store.nx`	authz, surveys, librarian, communitypulse, researchreach	—	none	unproven	undecided
cid_of	`nx_canon_cid.nx`	librarian*, researchreach	librarian	rung	plan	home elsewhere adopt: `@xrow librarian\|cid_of\|home by rung (fleet duplicate census)`
fs_scan_source	`nx_frontier_scan.nx`	intelmine, researchreach	—	none	unproven	undecided

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

The record census (nx_goalmap record) reads the invoked-tool population and every plan queue row and files each organ or directive that NO matrix, plan or gates row names. A row here is a callout the boards missed: adjudicate it onto a board or declare it infrastructure. Census state STALE (age 1396312 s), sources read 4 of 7 declared — a BLIND census is a FLOOR: unread sources can only add rows.

kind	name	board	source	evidence

rows shown 0|this board's directives 0|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.

[arxiv_csdl_202609] arXiv API query over cat:cs.DL (Digital Libraries), the 60 most recent by submission date, retrieved 2026-09-06. Enumerated in full: 61 title elements of 61 total, of which 7 carry a published date in 2026-09, the newest being 2026-09-03. Entries relied on include 2609.03105 FOXDEN, 2609.00969 The zbMATH Open Knowledge Graph, 2608.29612 LLMs Interpret Embeddings Organize Graphs Emerge, 2608.28642 From Extraction to Governed Memory, 2608.23760 Towards a Definition of the Computational Architecture of Open Scholarly Infrastructures, 2608.19026 Institutional Books Enriched Text, and 2608.24976 A Comparative Evaluation of Digitization Pipelines for Historiographical Sources. publisher · read in our library knowledge/fetched/cmp_scholarsource_arxiv_csdl_202609.atom · pin h35c2b83b138cd8cc0628e5bae18693a193609129e33fa2cbdf1eee85af3f4292 · accessed 2026-09-06 · mirrored-indexGrounds: THE DATED BAR ITSELF. This listing is the evidence that the bar month for this board is 2026-09 and not an inherited target, and it grounds the Layout-aware parsing row where the 2026 digitization and Institutional Books pipelines are named. It is cited rather than paraphrased because a bar without a month is a bar from whenever a seat first read it.
[oaipmh_spec] Open Archives Initiative. The Open Archives Initiative Protocol for Metadata Harvesting, version 2.0. Protocol specification, openarchives.org. publisher · read in our library knowledge/fetched/cmp_scholarsource_oaipmh_spec.html · pin hd81e32299f773b460c9dc8494bc3fe69d5e919fe4d9ef74c39d1a0e3ed7c2b98 · accessed 2026-09-06 · published-standardGrounds: The OAI-PMH incremental harvest row. The specification is the authority for the six verbs, for resumption tokens over long result sets, and for the from and until selective-harvest window that makes incremental harvesting possible rather than repeated full re-query. It is also the standard the two registry rows in knowledge/research_sources.conf expand into.
[openalex_api] OpenAlex Help Center -- the documentation surface for OpenAlex, the open catalog of the global research system. Retrieved 2026-09-06 at its canonical url. publisher · read in our library knowledge/fetched/cmp_researchreach_openalex_help.html · pin ha4405ec3a16a534736c39084c43a6c945b3850c27c080a0c4eed4798cd7ae0bf · accessed 2026-09-06 · vendor-docGrounds: CORRECTED THE DAY IT WAS WRITTEN, AND THE CORRECTION IS THE POINT. This row first cited https://docs.openalex.org/how-to-use-the-api/api-overview and nx_compare_refs_gate flagged it OPEN-REDIRECT ... MOVED with canonical CANON-HEAD-TRUNCATED. The old docs url REDIRECTS to help.openalex.org and the mirror's own head declares rel=canonical https://help.openalex.org/ with the title OpenAlex Help Center -- so the bytes were the HELP CENTRE HOME PAGE and not the API overview the row claimed. The fetch had succeeded, the pin was correct, and the citation was still of the wrong document. Re-fetched at the canonical url it now returns hops=0 with a byte-identical sha256, which is what proves the redirect target and this row are the same page. It grounds the documentation-surface claim ONLY; the key-free REST claim moved to openalex_live because a landing page does not demonstrate an API.
[openalex_live] OpenAlex Works API, live unauthenticated query https://api.openalex.org/works?per-page=1 returning HTTP 200 with 17563 bytes of JSON on 2026-09-06 with no API key and no account. publisher · read in our library knowledge/fetched/cmp_researchreach_openalex_root.json · pin he65661d6d16b7f45eb4920e828dc049eae2dae65d21bcf46c477ec302cb456d1 · accessed 2026-09-06 · source-readGrounds: THE KEY-FREE REST CLAIM, GROUNDED BY EXERCISING IT RATHER THAN BY READING A PAGE THAT DESCRIBES IT. This is the second key-free scholarly index deepresearch rung DN1 has required since 2026-08-20 before its fleet sweep may run wide, and a successful unauthenticated 200 is the strongest available evidence that the requirement is met. It also grounds the openalex registry row. NOTE ON THE PACE VALUE, STATED RATHER THAN QUIETLY DROPPED: the 1000 ms pace in knowledge/research_sources.conf was first justified as adopted from published polite-pool guidance, and no mirror on this board contains that guidance, so that derivation is WITHDRAWN and the value is now declared as a conservative default for an index we have not rate-measured.
[core_api] CORE. API v3 documentation -- search and retrieval over the aggregated open-access corpus including full text. Vendor documentation. publisher · read in our library knowledge/fetched/cmp_researchreach_core_api.html · pin hbcf2e3256d9926bcaedfc37dd33621d41754e5442ae507220fe09f97e6d7f62f · accessed 2026-09-06 · vendor-docGrounds: The CORE column: coded 2 on Open-access full text actually acquired and on OAI-PMH incremental harvest, and coded 0 on Hybrid retrieval because the mirrored documentation describes a corpus and search surface and makes no graph claim. A real 148034-byte documentation mirror, which is why depth claims about CORE are permitted where they are not for the two weak mirrors.
[openaire_api] OpenAIRE. Graph API v4 documentation -- the query surface over the OpenAIRE research graph. Vendor documentation. publisher · read in our library knowledge/fetched/cmp_researchreach_openaire_graphapi.html · pin h8155c1ae773ef75354df51060ffc4f95d156b3eeb8b142e88cf1087162b25e70 · accessed 2026-09-06 · vendor-docGrounds: The OpenAIRE column on the Hybrid retrieval row, where it is coded 2. This is the real 22900-byte documentation page and it is the only OpenAIRE mirror any graph claim on this board rests on.
[openaire_docs] OpenAIRE. Index of the published API documentation tree, listing authentication, broker-api, dspace-eprints-api and graph-api-v4. Retrieved 2026-09-06. publisher · read in our library knowledge/fetched/cmp_researchreach_openaire_api.html · pin h64891614c95e875781d4ca7c95b40f6bda6382a092b02be0f87f408b921f0491 · accessed 2026-09-06 · mirrored-indexGrounds: DECLARED WEAK: this mirror is an Apache directory index of 2813 bytes and NOT documentation, and it returned a clean status 200, which is the stub class this estate has measured before. It proves exactly one narrow fact and is cited only for that fact -- that OpenAIRE publishes a dspace-eprints-api alongside its graph API, which is the institutional-repository tier the Registry-of-registries expansion row and the OAI-PMH row rely on. No depth claim rests on it.
[s2_api] Semantic Scholar. Academic Graph API documentation entry page, declaring an Academic Graph API, a Recommendations API and a Datasets API, and loading its OpenAPI specification at runtime from graph/v1/swagger.json. publisher · read in our library knowledge/fetched/cmp_researchreach_s2_api.html · pin hb7533929074f34552a47fa750dab762541ef3d7240232b1fe852d463323eb3c6 · accessed 2026-09-06 · vendor-docGrounds: DECLARED WEAK: this mirror is a ReDoc JavaScript shell of 1752 bytes whose real specification is fetched by script at view time, so the document contains the API NAMES and none of their content. It is cited only for presence of those three surfaces, which is what the Semantic Scholar column codes rest on. IT WAS DELIBERATELY NOT RE-FETCHED to obtain a stronger mirror, because the sibling probe lane measured this host RATE-LIMITED against us on 2026-09-06 and spending requests on a throttled host to improve our own citation is the exact behaviour the Per-source pace budget row exists to govern.
[agentkg2026] Agent-driven compilation of scientific knowledge. LLMs Interpret, Embeddings Organize, Graphs Emerge. arXiv:2608.29612, August 2026. publisher · read in our library knowledge/fetched/cmp_researchreach_agentkg_2608_29612.html · pin h46d95ce7921cf03491b7cec80c7bd0c7087677dad361b1215fc4426faa6ba403 · accessed 2026-09-06 · published-paperGrounds: The 2026 frontier target for the Hybrid retrieval and Agentic multi-stage compile rows. Its title states the division of labour this board adopts as its architecture: interpretation by a model, organisation by embeddings, structure emerging as a graph -- which is the operator's stage three and stage four described as one system rather than two bolted together.
[foxden2026] FOXDEN: FAIR Services for AI-Ready Scientific Datasets. FAIR Open-Science Extensible Data Exchange Network, cyberinfrastructure for annotating datasets with structured metadata and provenance and publishing them with DOIs, deployed at CHESS and the National High Magnetic Field Laboratory. arXiv:2609.03105, September 2026. publisher · read in our library knowledge/fetched/cmp_researchreach_foxden_2609_03105.html · pin h9c100661e30fc6cb4d1cd160d946cc5b8ed01bc62b297c90b2635eb3987bef9d · accessed 2026-09-06 · published-paperGrounds: THE CURRENT-MONTH BAR, published 2026-09-02. It grounds the Agentic multi-stage compile row: its abstract describes rich machine-readable metadata and provenance records published with DOIs as the foundation for future agentic scientific workflows, which is the governance half our critic already has and our acquisition store does not yet expose.
[arxiv_apimanual] arXiv API User's Manual, arXiv info. Mirrored 2026-09-06 at 148526 bytes, hops=0. Searched for its rate guidance: the token `seconds` occurs ZERO times and `rate limit` ZERO times in the whole body. publisher · read in our library knowledge/fetched/cmp_researchreach_arxiv_apimanual.html · pin hf6c3ce2936f629e03fd42faf911604a7b0677ae3475c9c8608a543e9a7579671 · accessed 2026-09-06 · vendor-docGrounds: CITED FOR AN ABSENCE, WHICH IS WHY IT IS HERE AT ALL. The Per-source pace budget row originally derived 3000 ms from arXiv's published courtesy rate of one request per three seconds. This mirror is the check on that claim and it does not contain it, so the derivation is withdrawn and the value is declared a conservative default. An absence claim needs its mirror exactly as much as a presence claim does, and the scope is stated rather than implied: this proves the guidance is not on THIS page, not that arXiv has never published one.
[crossref_docs] Crossref. REST API -- retrieve metadata. Vendor documentation page, mirrored 2026-09-15 at 71499 bytes, hops=0. Read for: the polite pool, the /licenses and /works/{doi}/agency routes, and an OAI-PMH entry in the site navigation. publisher · read in our library knowledge/fetched/cmp_researchreach_crossref_docs.html · pin hcd4c13052aa19731acce614bb52c2b1ebde73e001a350aa8e51603d274c8a890 · accessed 2026-09-15 · vendor-docGrounds: The Crossref column: the registration-agency row, the polite-pool sentence on the pace row, the OAI-PMH row (a served surface, not harvesting) and the dedup row. A real documentation mirror, so depth claims about Crossref are permitted where they are not for the two weak mirrors.
[crossref_live] Crossref Works API, live unauthenticated query returning HTTP 200 with 2472 bytes of JSON on 2026-09-15 with no API key: total-results 186608495, one item carrying DOI, license and text-mining link fields. publisher · read in our library knowledge/fetched/cmp_researchreach_crossref_live.json · pin h7b5cc343f1d785d3e6dc6ce3b1314385a195fa8b00c29b5e48bcac08a4b2b7a8 · accessed 2026-09-15 · source-readGrounds: The Crossref coverage count, the key-free door row and its byte row, the DOI identity cell on the first row and the text-mining-links-not-text cell on the full-text row. A count read from the reply, never from prose.
[epmc_docs] Europe PMC. RESTful Web Service documentation page, mirrored 2026-09-15 at 65400 bytes: over 33 million publications, 10.2 million full text articles, 6.5 million open access, a fullTextXML operation, and an OAI service in its navigation. publisher · read in our library knowledge/fetched/cmp_researchreach_epmc_docs.html · pin h96b08d09fbb166ffc242767bcc65a517f91e2ef063446decf63adbd6741da85c · accessed 2026-09-15 · vendor-docGrounds: The Europe PMC column: the coverage figure, the open-access full-text row (Best beside CORE) and the OAI-PMH row (a served surface, not harvesting). A real documentation mirror.
[epmc_live] Europe PMC search API, live unauthenticated query returning HTTP 200 with 1156 bytes of JSON on 2026-09-15 with no key: hitCount 5596567 for the term cancer, a nextCursorMark for cursor paging, and a record carrying pmid and doi side by side. publisher · read in our library knowledge/fetched/cmp_researchreach_epmc_live.json · pin h26173f2ac02b65c6e8b333fce432c41d09653b46fa4057d8eb2b6a8e99849fb0 · accessed 2026-09-15 · source-readGrounds: The key-free door row and its byte row, and the PMID-beside-DOI cells on the identity and dedup rows. The hit count is for one term and is deliberately NOT quoted as a corpus size.
[openaire_live] OpenAIRE Graph API v1, researchProducts, live unauthenticated query returning HTTP 200 with 2233 bytes of JSON on 2026-09-15 with no key: numFound 389982852. publisher · read in our library knowledge/fetched/cmp_researchreach_openaire_live.json · pin h487bcde11fe5db87e7e152539c6f47dba9a4b8889b82f623d28ef897fd66a2fd · accessed 2026-09-15 · source-readGrounds: The OpenAIRE coverage count, replacing an order-of-magnitude read with a number the API returned, and the key-free door row and its byte row.
[opencitations_docs] OpenCitations. The REST API for OpenCitations Index, version 2: citation data, references and citation counts by DOI, PMID and OMID, free access, with an access token encouraged for application use and bulk dumps offered for large-scale retrieval. Mirrored 2026-09-15 at 35936 bytes. publisher · read in our library knowledge/fetched/cmp_researchreach_opencitations_docs.html · pin h18537556bcf0f6fc1ba6e1c89f24381f948e7f6e9c6f78361ab8e6dcb03d6a7c · accessed 2026-09-15 · vendor-docGrounds: The OpenCitations column: identifier resolution on the identity and dedup rows, the citation-graph-without-a-vector-leg cell on the hybrid row, and the token-encouraged-not-required sentence on the door row. A real documentation mirror.
[opencitations_live] OpenCitations Index API, live unauthenticated citation-count call for doi:10.1108/jd-12-2013-0166 returning HTTP 200 with 37 bytes on 2026-09-15 with no token: count 57. publisher · read in our library knowledge/fetched/cmp_researchreach_opencitations_live.json · pin h6153c2617323fe0df845ccd41c66b83f4c673befc66e810fc90ff1da0c9cff98 · accessed 2026-09-15 · source-readGrounds: The key-free door row and its byte row for OpenCitations. The DOI is the one the documentation page uses in its own example, so the call is the documented shape and not one invented here.
[kesen_sig2026] Ke-Sen Huang's SIGGRAPH 2026 papers index (Los Angeles, 19-23 July 2026): 288 dt-B entries counted by the estate on the mirrored bytes, read 2026-09-27 publisher · read in our library knowledge/fetched/cmp_graphics_kesen_sig2026.html · pin h7e6c166c5ae0fca3a235a74073fb3d85e055ca46fe59d7b20e3abd2519bb206b · accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUEREACH-2609
[kesen_eg2026] Ke-Sen Huang's Eurographics 2026 papers index: 108 dt-B entries counted on the mirrored bytes, read 2026-09-27 publisher · read in our library knowledge/fetched/cmp_graphics_kesen_eg2026.html · pin h92ca15926547daecb0f36c2fa3af9948295dc6a7fbebe807a4f74aaadf6e029b · accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUEREACH-2609
[kesen_siga2026] Ke-Sen Huang's SIGGRAPH Asia 2026 papers index: 7 conditionally accepted entries on the mirrored bytes, read 2026-09-27 publisher · read in our library knowledge/fetched/cmp_graphics_kesen_siga2026.html · pin hb745c68e9eda411a35406d31894a14300abb2ef2c563745d38868dbe88c50119 · accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUEREACH-2609
[arxiv_co_siga2026] arXiv API venue-tag scan on the comment field, phrase SIGGRAPH Asia 2026, 55 entries, newest 2026-09-24, read 2026-09-27; the proof that the arXiv comment field is a venue register the estate can query publisher · read in our library knowledge/fetched/cmp_graphics_arxiv_co_siga2026.xml · pin h51d49183aafe7aa1b0dbf264bc0c029b4716c07ea617c03a35fa2b58286599df · accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUEREACH-2609
[arxiv_co_cvpr2026_human] arXiv API venue-tag scan, CVPR 2026 comment phrase AND a human-subject filter, 120 entries at the fetch cap, newest 2026-09-03, read 2026-09-27; a PREFIX of the tagged population, which is RR21's motivation publisher · read in our library knowledge/fetched/cmp_graphics_arxiv_co_cvpr2026_human.xml · pin hb4b8b9bbbb66992d6344463bdc7fe37983b97af5724709621f31c110b9e96e5c · accessed 2026-09-27 · mirrored-indexGrounds: SB-VENUEREACH-2609
[mpeg_gsc_202609] MPEG-I Gaussian Splat Coding working page, mirrored 2026-09-27 over the TLS 1.2 leg (the host has no TLS 1.3); a standards body reached as a source publisher · read in our library knowledge/fetched/cmp_gameengine_mpeg_gsc_202609.html · pin h615fe9720333bdd8cedcbe2e285fab55b8c985179d208d7cb1893a77b3e73c2e · accessed 2026-09-27 · published-standardGrounds: SB-VENUEREACH-2609
[nist_skin_reflectance_202609] NIST publication page for the Reference Data Set of Human Skin Reflectance, mirrored 2026-09-27; a government metrology record reached as a source publisher · read in our library knowledge/fetched/cmp_skintwin_nist_skin_reflectance_202609.html · pin h5630d69eaa5bd2ba7c801de22f6d27e3dbaa27aa5ffff60d191d8de120a7af6a · accessed 2026-09-27 · datasetGrounds: SB-VENUEREACH-2609
[darpa_vital_202609] DARPA VITAL programme page, mirrored 2026-09-27; a government research programme reached as a source publisher · read in our library knowledge/fetched/cmp_anatomy_darpa_vital_202609.html · pin haaa9a666338a493fa36bd1549f255bf8077ae4aa246ae827418f7bf3b38efb1a · accessed 2026-09-27 · vendor-docGrounds: SB-VENUEREACH-2609
[rr26_datacite_dois_202610] DataCite REST API, GET /dois with query and resource-type-id=dataset, live response read 2026-10-03: every DOI-minting data repository in one listing, each record carrying attributes.rightsList (SPDX rightsIdentifier and rightsUri when declared, an empty list or a custom-terms pointer when not), which is the licence field the hunt reads per hit; client-id narrows it to one repository (figshare.ars, cos.osf) publisher · read in our library knowledge/fetched/rf_8757b417.raw · pin h8ac24b24dbc8d9df81b6755ed200e801602fc58663c86e2bcea090e9a9f14da6 · accessed 2026-10-03 · mirrored-indexGrounds: Dataset and media hunt from any repository, licence-classed
[rr26_europepmc_core_202610] Europe PMC RESTful Web Service, search with resultType=core, live response read 2026-10-03: each result carries a license field (cc by, cc by-nc, cc by-nc-nd) that resultType=lite omits, and field-scoped ABSTRACT queries keep full-text noise out publisher · read in our library knowledge/fetched/rf_637eea9c.raw · pin hc62bb41d8fdd0edfaff40a65999998c2226c95d42d94fe4423071ece87ec2ac7 · accessed 2026-10-03 · mirrored-indexGrounds: Dataset and media hunt from any repository, licence-classed
[rr26_physionet_catalogue_202610] PhysioNet published-project API (api/v1/project/published), live response read 2026-10-03: one static listing of 731 project versions, each with license.name and access_policy (Open, Credentialed, Restricted); carries openox-repo 1.0.0 to 1.1.1 under the PhysioNet Restricted Health Data License 1.5.0, the precedent the hunt must refuse publisher · read in our library knowledge/fetched/rf_7bb3b2cc.raw · pin hc1aa9b56e152d44f513cc0fa15abf6c7369b6ad516648d5ec3f45b885400dc5a · accessed 2026-10-03 · mirrored-indexGrounds: Dataset and media hunt from any repository, licence-classed

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