Data refresh

What each refresh found, and what it did not publish.

ModelTree's dataset is refreshed by agents against primary sources, reviewed by an independent three-rubric panel, and gated deterministically. Every run is recorded here in full — including the runs that published nothing.

A run's working state is never committed. This page transcribes the durable record: the pull request body and the summary issue, both linked from each entry.

Runs recorded: 23
Pages fetched: 2,128
Claims proposed: 1,352
Edits published: 611
Items withheld: 439

Showing 1–3 of 3 runsfiltered by Stopped, published nothing

Clear filter

2026-09-05-df67a5

Targeted GPT-6 Astra refresh 2026-09-05

Scope requested: One creator, OpenAI, chosen because the scheduled refresh had missed GPT-6 Astra. The remaining 43 creators in organizations.json were deliberately not scouted and are named individually in found.unswept, so that "we did not look" is not reported as "we looked and nothing had changed".Stopped, published nothingSep 5, 20260 edits posted · 4 items withheld

Scouted OpenAI alone, re-fetching and re-hashing every byte rather than inheriting run 2026-09-05-ad1a1f evidence. Proposed 4 claims, all of which reached the 2-of-3 pilot threshold, and passed every deterministic data gate: evidence, source-approval, dataset (626 records) and scope. The run then stopped at publish and applied nothing. Adding the release and its family breaks two assertions that hard-code a dataset population count -- the datesCoarserThanADay inventory in web/src/data/validate.test.ts and the reordered-population count in web/src/lib/release-source.test.ts -- and both files are outside the ADR 0003 qualifying class. gate-scope adjudicated the repaired diff directly and refused it at exit 1, naming both paths as outOfClass. The page-weight blockers that stopped the previous run were measured and are no longer binding: the /compare payload now sits at 143,543 of 174,080 bytes with 30,537 spare, and the two asset-budget figures that drift are inside the class ADR 0015 admits, so they were re-recordable and were simply not reached.

PreflightRanClean tree, gh authenticated, node v24.14.0. Bare `npm` is refused by this machine's PowerShell execution policy while `npm.cmd` runs, so the toolchain is installed-and-blocked rather than absent; the policy was not changed. Trunk moved mid-run from 1f5d89e5 to 5e344962, a docs-and-ADR-only commit; web/ was confirmed byte-identical across that range with a difference control, so the local build is the tree CI measures.
ScoutRan14 URLs fetched and hashed against a 40-page budget. Three returned 403 and were recorded rather than worked around. platform.openai.com/docs/models redirects to developers.openai.com/api/docs/models and returned a byte-identical body (both artefacts exactly 364,123 bytes); both ends are approved origins, so the redirect crosses no trust boundary. The HTML model page is JS-hydrated and does not carry the specification lines as text, so the .md twin is cited instead, following the committed openai-gpt-daybreak-blue-docs convention.
ReviewRanThree blind reviewers on different model families, launched in parallel, none shown the scout reasoning, the running tally or another reviewer's verdict. 12 verdicts cast over 4 claims. All 4 reached the 2-of-3 pilot threshold. Two provenance rejections were recorded rather than overruled, and no reviewer was re-run.
GatesRancheck-bundle-pairing, gate-evidence, gate-source-approval, gate-dataset and gate-scope all passed on the dataset-only change. The full web/ validation then failed on two out-of-class population controls, which is what stopped the run.
PublishNot runNo data was applied, so no data pull request was opened. The accepted claims were reverted and the tree left clean. This ledger entry is published on its own, as run 2026-09-05-ad1a1f did.
DeployNot applicableNo data change reached main, so there is no data deployment to verify for this run.

What was found

Scouts: 1
Pages fetched and hashed: 14
Claims proposed: 4

Claims per creator bundle, with the review threshold its profile set
Creator	Policy	Threshold	Claims
openai	pilot	2-of-3	4

What those claims proposed to do
Kind	Count	Effect
Add	3	One source record for the GPT-6 Astra documentation page, one GPT-6 family, and one GPT-6 Astra release. The family and release travel together because gate-dataset fails a family holding no release.
Change	1	Advance openai-news-rss lastCheckedDate from 2026-08-31 to 2026-09-05 after a confirmed re-fetch at HTTP 200.

Not covered

No benchmark result was proposed. No sourced figure on an approved origin required reasoningMode, toolsEnabled or harness, so nothing was withheld on that ground; those fields were not populated because they pass gate-dataset but break the /compare key-compaction guard, whose repair touches web/src/lib/comparison.ts, a path outside the qualifying class.
No lineage claim was proposed. No consulted source states that GPT-6 Astra succeeds any GPT-5.6 release, and succession was not inferred from version numbers.
No safety or evaluation claim was attempted, because the safety overview page returned 403.

Not scouted this run

Creators this run did not look at, kept distinct from a creator that was scouted and found unchanged.

Creator	Last scouted	Why skipped
anthropic	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
google-deepmind	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
meta	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
xai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
mistral-ai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
deepseek	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
alibaba-cloud	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
microsoft	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
amazon	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
cohere	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
ai2	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
tii	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
nvidia	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
ai21-labs	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
zhipu-ai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
moonshot-ai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
eleutherai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
lg-ai-research	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
snowflake	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
upstage	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
ibm	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
baidu	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
tencent	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
bytedance-seed	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
stability-ai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
databricks	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
minimax	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
apple	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
hugging-face	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
01-ai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
sakana-ai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
sarvam-ai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
naver	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
aleph-alpha	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
reka-ai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
nous-research	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
liquid-ai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
xiaomi	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
ai-singapore	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
kyutai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
lelapa-ai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
maritaca-ai	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.
openbmb	Not established	Out of scope for this run, which was scoped to OpenAI alone to re-verify GPT-6 Astra. Not looked at this pass; no inference is available about whether it changed. lastScouted is omitted because it is not establishable from the committed ledger, whose previous entry records scouts: 44 against only 6 bundles.

Degraded discovery channels

openai — official-announcement: openai.com/news/ returned HTTP 403, reproducing the failure reported for 2026-09-04, so the profile's catalogued announcement channel is dark. Discovery survived only because openai.com/news/rss.xml still served HTTP 200. openai.com/index/gpt-6-astra and openai.com/index/safety-overview-gpt-6-astra also returned 403. No workaround was attempted: no scraping, no user-agent spoofing, no mirror and no press coverage as evidence.

What was evaluated

Reviewers: 3
Verdicts cast: 12
Accepted by panel: 4
Rejected by panel: 0

Deterministic gates and required checks — 7 of 9 checks passed. Exit 0 is a pass; exit 2 means the gate could not run and is never treated as one.
Check	Scope	Exit	Result
`check-bundle-pairing`	openai claim bundle	0	PassThe family add and the release add travel together in one bundle.
`gate-evidence`	openai claim bundle	0	Pass4 claims admissible under the pilot policy. Form only; ADR 0005 records that this gate does not re-fetch remote content.
`gate-source-approval`	openai claim bundle	0	Passpassed: true with 0 failures. anchor.commit 1f5d89e59f849e89c61a3cf3a4957bae36e2049b, selectedBy merge-base with refs/remotes/origin/main. Inherited source openai-news-rss; proposed source openai-gpt-6-astra-docs on the already-approved developers.openai.com origin.
`gate-dataset`	web/src/data with the 4 accepted claims applied	0	PassAll gates passed over 626 records. Counts would have been sources 286, families 86, releases 121.
`gate-scope`	the dataset-only diff	0	Passchanged 3, all in class (families.json, releases.json, sources.json), empty false. In class but insufficient: this diff cannot produce a green CI on its own.
`gate-scope`	the diff that green CI actually requires	1	Failchanged 5, outOfClass web/src/data/validate.test.ts and web/src/lib/release-source.test.ts, passed false. The repairs were verified to make those tests pass (73/73) and were then reverted. Note that validate.test.ts sits inside web/src/data/ yet is not in class: ALLOWED_PATHS is an exact list of 16 JSON files, not a directory prefix. This is the stop.
`npm run budget:compare`	the /compare payload with the release applied	0	Pass143,543 of 174,080 bytes with 30,537 spare, and 1,186 of 1,600 bytes per release. The ceiling raise and the payload trim between them cleared the blocker that stopped run 2026-09-05-ad1a1f; this limb was measured rather than assumed.
`npm run validate`	web/ with the 4 accepted claims applied	1	Fail5 failures over 2,860 tests. Two are the out-of-class population controls that stopped the run. Two are asset-budget measuredDrift failures inside the class ADR 0015 admits, so they were re-recordable and were not reached. One, web/scripts/run-check.test.ts, failed on a 180s timeout rather than an assertion and contains zero references to the dataset, so it is environmental and unrelated.
`asset-budgets.test.ts`not required	a clean trunk tree, as a control	0	Pass41/41 passed with no claims applied, establishing that the recorded figures are within tolerance on trunk itself and that the drift this run measured is caused by the tranche rather than inherited. Nothing in web/asset-budgets.json was touched, so the measuredWorstJsRaw exact-zero tripwire on routeGroups[0] was not moved, and no ceiling and no measuredDrift.maxFraction was changed.

Applied over a recorded dissent

These met their threshold and were applied. The objection stands on the record and was not overruled.

openai-gpt-6-family-addProvenanceHeld that the quotes describe GPT-6 Astra rather than a GPT-6 family: neither states that GPT-6 is a model generation nor that the categories apply family-wide. Outvoted 2-of-3 under the pilot policy and recorded rather than overruled.
openai-gpt-6-astra-release-addProvenanceAccepted the modalities, token limits, alias, uses, pricing and endpoints as directly quoted, but held that the feed item's pubDate states the item's publication time rather than the model's release date, and that API availability does not force proprietary-hosted over both because it does not state that weights are unavailable. Outvoted 2-of-3 under the pilot policy and recorded rather than overruled.

Posted 0 edits

Nothing reached the dataset. No branch, no commit, no pull request.

Not posted 4 items

Blocked by policy before it could run

openai-gpt-6-astra-docs-source-addAccepted 3-of-3 and not applied. It is cited by the release record, so it cannot land without it, and an uncited source is dead provenance that validate.test.ts refuses.Blocked by web/src/data/validate.test.ts datesCoarserThanADay exact inventoryweb/src/lib/release-source.test.ts reordered-population countgate-scope.mjs ALLOWED_PATHS (ADR 0003 qualifying class)
openai-news-rss-lastcheckedAccepted 3-of-3 and not applied. Advancing a verification date alone would assert that the feed was checked for a tranche that did not land, so it was reverted with the rest.Blocked by web/src/data/validate.test.ts datesCoarserThanADay exact inventoryweb/src/lib/release-source.test.ts reordered-population countgate-scope.mjs ALLOWED_PATHS (ADR 0003 qualifying class)
openai-gpt-6-family-addAccepted 2-of-3 over a recorded provenance objection and not applied. Its datePrecision unstated adds a ninth entry to the exact datesCoarserThanADay inventory in validate.test.ts. The unstated value is the honest one: no primary states a first release date for the family, and dating a family from one member's announcement is the step the provenance rubric names as a defect not to be extended.Blocked by web/src/data/validate.test.ts datesCoarserThanADay exact inventoryweb/src/lib/release-source.test.ts reordered-population countgate-scope.mjs ALLOWED_PATHS (ADR 0003 qualifying class)
openai-gpt-6-astra-release-addAccepted 2-of-3 over a recorded provenance objection and not applied. It cites two sources, taking the reordered population in release-source.test.ts from 110 to 111. Citing one source instead would leave the release date unsourced, so the count cannot be avoided honestly.Blocked by web/src/data/validate.test.ts datesCoarserThanADay exact inventoryweb/src/lib/release-source.test.ts reordered-population countgate-scope.mjs ALLOWED_PATHS (ADR 0003 qualifying class)

What this run does not prove

This run did not find nothing to do. It found a real, sourced, panel-accepted release it was not permitted to publish, and the obstruction is not the one the previous stopped run named.
The obstruction is structural rather than specific to GPT-6. Any release citing two or more sources moves the release-source.test.ts population, and any record whose date precision is not day moves the validate.test.ts inventory. Both are exact by design, and both are outside the class a refresh may touch, so an unattended refresh cannot add a release of either shape.
Two claims carried on 2-of-3 over a recorded provenance objection. The release date rests on a news-feed item pubDate rather than on prose, disclosed in the record summary exactly as the committed openai-gpt-5-6-cyber discloses its own announcement-derived date. A reader who holds that objection to be correct should read those two records as unsupported on that point.
The panel is three model families reading the same pages. 2-of-3 buys independence of reasoning, not independence of training, and a source that is itself wrong can carry all three.
OpenAI's catalogued official-announcement channel returned 403 throughout, so this run saw the announcement only through the RSS feed. Anything stated in the announcement prose and not in the feed or the documentation was invisible to it.
43 of 44 creators were not scouted at all. This run says nothing whatever about whether any of them changed.
gate-evidence verifies the form of evidence, not remote content. That a quote matches its page was established by this run fetching and hashing the page itself, and is not something the gate re-checks.

Follow-ups — proposed, not fixed

The two exact-population controls are the single highest-value fix for refresh throughput: either they should derive their figures from the dataset while keeping the anti-blindness property that made them exact, or they need admitting to the qualifying class. Filed as the follow-up in this run's summary issue rather than fixed here, because fixing them is itself the out-of-class change that stopped the run.
OpenAI's official-announcement source is dark at 403 while its RSS feed still serves 200. The profile currently depends on the feed alone for discovery, which no gate would report if the feed also went quiet.
The previous stopped run's conclusion that web/asset-budgets.json lies outside the scope a refresh may touch is wrong for that file: ADR 0015 admits it, field-scoped, and both figures it named are in ASSET_BUDGETS_REGENERABLE_FIELDS. The conclusion was right about comparison.test.ts. Recording the correction here so the next run does not inherit the whole-path reading.

2026-09-05-ad1a1f

Data refresh 2026-09-05

Scope requested: Every creator in organizations.json (44), plus the long-tail profile sweep. Six creators were taken to claim depth: openai, moonshot-ai, upstage, minimax, liquid-ai, xiaomi.Stopped, published nothingSep 5, 20260 edits posted · 26 items withheld

Scouted all 44 creators, extracted 26 claims across 6, ran the three-rubric panel (16 accepted, 10 rejected) and passed every deterministic data gate — then stopped at the publish stage and applied nothing. OpenAI GPT-6 Astra was accepted and gate-clean, but adding one release takes the shipped /compare payload 424 bytes past a ceiling held in web/src/lib/comparison.test.ts and pushes two recorded measurements in web/asset-budgets.json past their drift allowance. Both files are outside the scope ADR 0003 permits a refresh to touch, so the run filed #935 and published no dataset record. The working tree was returned to clean.

PreflightRanClean tree, gh authenticated as abdeslam-menacere, node v24.14.0, no open refresh pull request. Bare `npm` is refused by this machine's PowerShell execution policy while `npm.cmd` runs — recorded as installed-and-blocked, and the policy was not changed.
ScoutRan111 pages fetched and content-hashed across the 51 approved origins; 56 quotes verified verbatim against the exact hashed bytes with positive and negative controls. A Hugging Face gap sweep covered 36 organisations, matching 197 records and surfacing 395 leads.
ReviewRanThree rubrics run as independent parallel agents, each blind to the others and to the scout's reasoning. 78 verdicts cast over 26 claims. 16 met their bundle's own threshold, 10 did not.
GatesRangate-evidence passed on the only bundle contributing records and failed on five whose claims the panel had already rejected; gate-source-approval passed on all six; gate-dataset passed on the applied tree. The site's own npm run validate then failed on three page-weight assertions, which is what stopped the run.
PublishNot runNothing was published. The four accepted claims were reverted rather than committed, because landing them requires editing files outside the permitted scope. No branch content, no pull request carrying a dataset record.
DeployNot applicableNothing merged, so nothing deployed.

What was found

Scouts: 44
Pages fetched and hashed: 111
Claims proposed: 26

Claims per creator bundle, with the review threshold its profile set
Creator	Policy	Threshold	Claims
liquid-ai	long-tail	3-of-3	5
minimax	long-tail	3-of-3	5
moonshot-ai	long-tail	3-of-3	4
openai	pilot	2-of-3	4
upstage	long-tail	3-of-3	4
xiaomi	long-tail	3-of-3	4

What those claims proposed to do
Kind	Count	Effect
Add	26	Every claim this run proposed was an addition — 14 sources, 6 families, 6 releases. None reached the dataset.

Not covered

The 395 Hugging Face leads surfaced by the gap sweep were not converted into claims. The sweep ranks candidates; it does not establish that a repository is a creator's own release, and each lead still needs a primary source read before it can become a claim.
zhipu-ai GLM-5.3 was seen but deferred: the pages reachable this run did not carry a primary statement clean enough to quote for a dated release record.
No release was removed, superseded or marked legacy this run. The run looked for additions and did not audit existing records for lifecycle drift.
Four creators have never been scouted by any recorded run: kyutai, lelapa-ai, maritaca-ai, openbmb.

Not scouted this run

Creators this run did not look at, kept distinct from a creator that was scouted and found unchanged.

Creator	Last scouted	Why skipped
01-ai	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
ai-singapore	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
ai2	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
ai21-labs	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
aleph-alpha	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
alibaba-cloud	Sep 3, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
amazon	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
anthropic	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
apple	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
baidu	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
bytedance-seed	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
cohere	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
databricks	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
deepseek	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
eleutherai	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
google-deepmind	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
hugging-face	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
ibm	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
kyutai	Not established	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
lelapa-ai	Not established	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
lg-ai-research	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
maritaca-ai	Not established	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
meta	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
microsoft	Sep 3, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
mistral-ai	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
naver	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
nous-research	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
nvidia	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
openbmb	Not established	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
reka-ai	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
sakana-ai	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
sarvam-ai	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
snowflake	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
stability-ai	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
tencent	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
tii	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
xai	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.
zhipu-ai	Sep 1, 2026	Swept for discovery only — Hugging Face organisation listing and catalogued channels compared against the committed dataset — and not carried to claim depth this pass. Leads surfaced for this creator were left unconverted because the run stopped at the publish stage.

Degraded discovery channels

openai — official-announcement: openai.com/news/ returned 403 to this run, so OpenAI's announcement channel was dark. GPT-6 Astra was found instead through platform.openai.com primary documentation and the API changelog.
xai — official-announcement: x.ai/news returned 403 to this run, so xAI's announcement channel was dark and this run is not evidence that xAI shipped nothing.

What was evaluated

Reviewers: 3
Verdicts cast: 78
Accepted by panel: 16
Rejected by panel: 10

Deterministic gates and required checks — 3 of 5 checks passed. Exit 0 is a pass; exit 2 means the gate could not run and is never treated as one.
Check	Scope	Exit	Result
`gate-evidence`	openai.claims.json — the only bundle contributing records	0	Passpassed: true, 0 failures over 4 applicable claims. Every quote matched the hashed bytes and every claim carried exactly three verdicts, one per rubric.
`gate-evidence`not required	the five long-tail bundles (moonshot-ai, upstage, minimax, liquid-ai, xiaomi)	1	FailTen claims failed the review threshold, which is the gate agreeing with the panel rather than a defect: each is a family or release that reached 1 or 2 of the 3 accepts a long-tail creator requires. Zero evidence failures across all 56 quotes. None of these claims was applied, so no failure here reached the dataset.
`gate-source-approval`	all six claim bundles	0	PassEvery cited URL resolved to one of the 51 origins a reviewed profile catalogue or the committed dataset already stands behind. No run-approved source.
`gate-dataset`	the working tree with the four accepted claims applied	0	Passpassed: true, failures: []. Counts with the change applied — sources 287, publishers 52, organizations 44, families 86, releases 121, products 1, servingPlatforms 3, deployments 3, benchmarks 4, benchmarkResults 8, releaseEvents 7, usageObservations 2, usageSyntheses 0, modelFitStatements 7, modelFitEvidenceGaps 2.
`npm run validate`	the working tree with the four accepted claims applied	1	FailFour tests failed against a baseline of 2856/2856 passing on the same tree without the change, so the failures are caused by the dataset change and nothing else. /compare ships 143,784 bytes against a 143,360 ceiling (424 over) while the per-release figure improved 1,191 to 1,188 of 1,600; catalog measuredRaw drifted 13,527 against a 12,350 allowance; providers measuredWorstRaw drifted 19,666 against 12,884. This is the gate that stopped the run.

Applied over a recorded dissent

These met their threshold and were applied. The objection stands on the record and was not overruled.

openai-family-gpt-6ProvenanceHeld that status: 'current' rests on no quoted lifecycle statement — the documentation lists the model without describing its lifecycle — so the value is a reasonable reading rather than a recorded fact. Outvoted 2-of-3 under the pilot policy and recorded rather than overruled.
openai-release-gpt-6-astraProvenanceHeld that the 2026-09-03 release date rests on a changelog heading ('## September, 2026' / '### Sep 3') that never names GPT-6 Astra in the same breath, so attributing that date to this release adds a scope the source does not state. Outvoted 2-of-3 under the pilot policy and recorded rather than overruled.

Posted 0 edits

Nothing reached the dataset. No branch, no commit, no pull request.

Not posted 26 items

Rejected by the review panel

liquid-family-lfm2-5Reached 2 of 3 required accepts under the long-tail policy, so it was not applied. provenance: status: 'current' is unsupported — 'LFM2.5-8B-A1B is a general-purpose text-only model' and the createdAt quote state no lifecycle term. The required status field filled from no quote is a rejection.Blocked by rubric:provenance
liquid-release-lfm25-8b-a1bReached 1 of 3 required accepts under the long-tail policy, so it was not applied. provenance: status: 'current' has no lifecycle quote, and accessType: 'open-weight' with weightsDownloadable: true rest on the 'LFM Open License v1.0' name line, which states a licence name rather than downloadability. Either unsourced field is an independent rejection under the standing open-weight and status precedents. consistency: The release asserts license.osiApproved false but cites only Liquid AI and Hugging Face sources. No sourceId resolves to publisher open-source-initiative, violating the dataset's structural rule for every recorded OSI approval value.Blocked by rubric:provenancerubric:consistency
minimax-family-m3Reached 2 of 3 required accepts under the long-tail policy, so it was not applied. provenance: status: 'current' has no supporting quote — the card quote ('native multimodal model with 1M context...') and the createdAt quote state nothing about lifecycle. The unsourced required status field is a rejection regardless of the otherwise sound date basis.Blocked by rubric:provenance
minimax-release-m3Reached 1 of 3 required accepts under the long-tail policy, so it was not applied. provenance: Modalities are properly backed by the '"pipeline_tag":"image-text-to-text"' quote, but status: 'current' has no lifecycle quote, and accessType: 'open-weight' with weightsDownloadable: true are supported only by the 'MINIMAX COMMUNITY LICENSE' name line, not by any statement that weights are downloadable (mi4/open-weight precedent). Reject on the unsourced status and access fields. consistency: The release carries license.osiApproved false while citing only MiniMax and Hugging Face publishers. It omits an Open Source Initiative-published source such as osi-license-index, so validateDataset would reject the complete record.Blocked by rubric:provenancerubric:consistency
moonshot-family-kimi-k3Reached 2 of 3 required accepts under the long-tail policy, so it was not applied. provenance: status: 'current' has no supporting lifecycle quote — the card quote states modalities/context and the hub quote states createdAt, neither of which states a lifecycle state (the mi4 precedent rejects exactly this). The firstReleaseDate/dateBasis pairing is sound, but the unsourced required status field sinks the whole record.Blocked by rubric:provenance
moonshot-release-kimi-k3Reached 1 of 3 required accepts under the long-tail policy, so it was not applied. provenance: status: 'current' has no lifecycle quote, and accessType: 'open-weight' with weightsDownloadable: true rest only on the licence line 'Kimi K3 License Copyright (c) 2026 Moonshot AI', which states a licence name, not that weights are downloadable — the standing open-weight rejection. categories also includes 'coding' with no supporting quote. Any one of these is a rejection. consistency: The release records license.osiApproved as false but cites only moonshot-ai-kimi-k3-model-card and hugging-face-kimi-k3-hub-record. Neither source is published by open-source-initiative, which validateDataset requires for every osiApproved value, so this record would invalidate the dataset.Blocked by rubric:provenancerubric:consistency
upstage-family-solar-open-2Reached 2 of 3 required accepts under the long-tail policy, so it was not applied. provenance: status: 'current' is unsupported — the only quotes are the licence-distribution sentence and the createdAt, neither stating any lifecycle term. A required controlled-vocabulary field mapped from no quote is a rejection.Blocked by rubric:provenance
upstage-release-solar-open2-250bReached 1 of 3 required accepts under the long-tail policy, so it was not applied. provenance: Multiple fields lack quote support: inputModalities/outputModalities ['text'] have no attached quote (the pipeline tag is described in notes but never quoted), status: 'current' has no lifecycle quote, and accessType: 'open-weight' rests on 'distributed under the Upstage Solar License' — a licence name that does not state downloadable weights. Each is an independent rejection. consistency: The release asserts license.osiApproved false but its sourceIds contain only Upstage and Hugging Face sources. Because no Open Source Initiative-published source is cited, it violates validateDataset's mandatory OSI evidence-reference rule.Blocked by rubric:provenancerubric:consistency
xiaomi-family-mimo-v2-5Reached 2 of 3 required accepts under the long-tail policy, so it was not applied. provenance: status: 'current' has no supporting lifecycle quote — the modalities quote and the createdAt quote state none. The required status field mapped from no quote is a rejection, independent of the otherwise sound modality and date evidence.Blocked by rubric:provenance
xiaomi-release-mimo-v2-5Reached 2 of 3 required accepts under the long-tail policy, so it was not applied. provenance: osiApproved: true is properly evidenced by the opensource.org MIT page title, but status: 'current' has no lifecycle quote, and accessType: 'open-weight' with weightsDownloadable: true have no quote stating weights are downloadable (the 'license: mit' front matter states a name only). The unsourced status and access fields sink the whole record.Blocked by rubric:provenance

Accepted by the panel, then dropped

liquid-source-lfm25-cardnothing cites source liquid-ai-lfm2-5-8b-a1b-model-card once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion
liquid-source-lfm25-hubnothing cites source hugging-face-lfm2-5-8b-a1b-hub-record once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion
liquid-source-lfm-open-licensenothing cites source liquid-ai-lfm-open-license once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion
minimax-source-m3-cardnothing cites source minimax-m3-model-card once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion
minimax-source-m3-hubnothing cites source hugging-face-minimax-m3-hub-record once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion
minimax-source-m3-licensenothing cites source minimax-m3-license once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion
moonshot-source-kimi-k3-cardnothing cites source moonshot-ai-kimi-k3-model-card once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion
moonshot-source-kimi-k3-hubnothing cites source hugging-face-kimi-k3-hub-record once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion
upstage-source-solar-open2-cardnothing cites source upstage-solar-open2-250b-model-card once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion
upstage-source-solar-open2-hubnothing cites source hugging-face-solar-open2-250b-hub-record once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion
xiaomi-source-mimo-v25-cardnothing cites source xiaomi-mimo-v2-5-model-card once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion
xiaomi-source-mimo-v25-hubnothing cites source hugging-face-mimo-v2-5-hub-record once the panel's rejections are applied, and an uncited source is dead provenance that validate.test.ts refusesBlocked by web/src/data/validate.test.ts orphaned-source assertion

Blocked by policy before it could run

openai-source-gpt-6-astra-docsPanel-accepted and clean through gate-evidence, gate-source-approval and gate-dataset, then withheld at the publish stage. Applying it takes the /compare shipped payload to 143,784 bytes against a 143,360 ceiling in web/src/lib/comparison.test.ts, and moves two recorded measurements in web/asset-budgets.json past their 2% drift allowance. Both files sit outside the fifteen dataset documents plus the ledger that gate-scope.mjs admits, so ADR 0003 stops the run rather than letting it edit them.Blocked by ADR 0003 scope guardrailweb/src/lib/comparison.test.tsweb/asset-budgets.json#935
openai-source-api-changelogPanel-accepted and clean through gate-evidence, gate-source-approval and gate-dataset, then withheld at the publish stage. Applying it takes the /compare shipped payload to 143,784 bytes against a 143,360 ceiling in web/src/lib/comparison.test.ts, and moves two recorded measurements in web/asset-budgets.json past their 2% drift allowance. Both files sit outside the fifteen dataset documents plus the ledger that gate-scope.mjs admits, so ADR 0003 stops the run rather than letting it edit them.Blocked by ADR 0003 scope guardrailweb/src/lib/comparison.test.tsweb/asset-budgets.json#935
openai-family-gpt-6Panel-accepted and clean through gate-evidence, gate-source-approval and gate-dataset, then withheld at the publish stage. Applying it takes the /compare shipped payload to 143,784 bytes against a 143,360 ceiling in web/src/lib/comparison.test.ts, and moves two recorded measurements in web/asset-budgets.json past their 2% drift allowance. Both files sit outside the fifteen dataset documents plus the ledger that gate-scope.mjs admits, so ADR 0003 stops the run rather than letting it edit them.Blocked by ADR 0003 scope guardrailweb/src/lib/comparison.test.tsweb/asset-budgets.json#935
openai-release-gpt-6-astraPanel-accepted and clean through gate-evidence, gate-source-approval and gate-dataset, then withheld at the publish stage. Applying it takes the /compare shipped payload to 143,784 bytes against a 143,360 ceiling in web/src/lib/comparison.test.ts, and moves two recorded measurements in web/asset-budgets.json past their 2% drift allowance. Both files sit outside the fifteen dataset documents plus the ledger that gate-scope.mjs admits, so ADR 0003 stops the run rather than letting it edit them.Blocked by ADR 0003 scope guardrailweb/src/lib/comparison.test.tsweb/asset-budgets.json#935

What this run does not prove

This run did not find "nothing to do". It found a real, sourced, panel-accepted release it was not permitted to publish, which is why #935 was filed and left open.
Publishing GPT-6 Astra is blocked by page-weight artefacts, not by any doubt about the data. The /compare stopping rule was measured as met on both limbs: the catalogue grew and the per-release figure improved rather than merely holding.
The blocker is structural and not about this release. The baseline had 410 bytes of /compare headroom against roughly 834 bytes for a release and its two cited sources, so any new release from any creator now tips the same ceiling.
Thirty-eight creators were swept for discovery only and not carried to claim depth, so this run is not evidence that those creators shipped nothing. It is evidence that the discovery sweep surfaced no candidate strong enough to chase before the run stopped.
OpenAI's and xAI's announcement channels both returned 403, so neither creator's primary announcement signal was readable this run.
The 395 Hugging Face leads remain unconverted. A future run should expect them, not treat them as new.
New families were proposed with featured: false throughout. Promoting a family to the front page is an editorial decision this run did not make.
gate-dataset passing shows the applied dataset was internally coherent. It has no network and does not check that a source still says what it said when it was hashed.

Follow-ups — proposed, not fixed

#935 — the blocker this run filed: an auto-merging refresh cannot add a release while the /compare total and the asset-budget measurements live outside its permitted scope.
Xiaomi MiMo v2.5 cited an OSI licence source but its model card carried no LICENSE file this run, so the licence block could not be completed from a primary source.
Four creators (kyutai, lelapa-ai, maritaca-ai, openbmb) have never been scouted by any recorded run and should be prioritised by staleness on a future pass.
The four long-tail releases rejected on license.osiApproved would each have needed an opensource.org citation alongside the model card; a scout pass that fetches the OSI licence page up front would convert them.

2026-08-25-902306

Data refresh 2026-08-25

Scope requested: Every creator in organizations.json, plus the long-tail profile sweepStopped, published nothingAug 25, 20260 edits posted · 1 item withheld

Stopped at preflight and published nothing. ADR 0003 precondition 2 was still open — the approved-source binding enforced on the proposal-only path had no equivalent in the publishing skill set — and ADR 0003's guardrail is that the automation does not run until the skill set is corrected. No dataset change, no branch, no commit, no pull request, no deploy.

PreflightRanModelTree checkout, clean tree, gh authenticated, no open pull request from a previous refresh. Preconditions were then checked live and one was open.
ScoutNot runZero pages fetched, zero claims extracted. The publisher was not authorised to run.
ReviewNot runNo bundle existed to review.
GatesNot runNo bundle existed to gate. A read-only dataset health check was run separately and is recorded below.
PublishNot runNothing was applied, so there was nothing to publish.
DeployNot applicableNothing merged, so nothing deployed.

What was found

Scouts: 0
Pages fetched and hashed: 0
Claims proposed: 0

Not covered

Every creator in scope. Scouting never started, so no creator was checked for new releases.
Because scouting did not run, this run is not evidence that the dataset is current. It says only that the publisher was not authorised to run.

What was evaluated

Reviewers: 0
Verdicts cast: 0
Accepted by panel: 0
Rejected by panel: 0

Deterministic gates and required checks — 1 of 1 check passed. Exit 0 is a pass; exit 2 means the gate could not run and is never treated as one.
Check	Scope	Exit	Result
`gate-dataset`not required	committed dataset, read-only health check	0	PassRun standalone rather than as a run gate: passed: true, failures: []. Counts at the time — sources 47, publishers 8, organizations 4, families 11, releases 22, usageObservations 2, usageSyntheses 0, modelFitStatements 7, modelFitEvidenceGaps 3.

Posted 0 edits

Nothing reached the dataset. No branch, no commit, no pull request.

Not posted 1 item

Blocked by policy before it could run

entire-runADR 0003 precondition 2 was open. gates.py enforced an approved-source binding per claim; the publishing skill set's gate-evidence.mjs never inspected sourceId against any approved catalogue, and gate-dataset.mjs resolves sourceIds only referentially against a file the refresh may itself patch. On that rule the publishing path was strictly more permissive, which stops the automation.Blocked by ADR 0003 precondition 2#167

What this run does not prove

This run did not find "nothing to do". It found work for a human, which is why its summary issue was left open until the blocker was closed.
The read-only health check shows the published dataset was internally coherent. It does not show that it was current or true — gate-dataset has no network and does not check that a source still says what it said.

Follow-ups — proposed, not fixed

#167 — the blocker: the publishing gates must refuse a run that adds a source record to sources.json and cites it in the same change.
#205 — gate soundness is emergent across the set rather than provable gate by gate.
#210 — gate-scope reported green on a committed out-of-class change when run with no --base.
#209 — no gate input self-reported by the subject of the gate may have a default.
#168 — nothing detects drift between the Python and JavaScript gate implementations.