Skip to content

Restore model-currency validation with sourced Fable and Astra landscape checks - #1101

Merged
seathatflowsinourveins merged 1 commit into
mainfrom
codex/model-landscape-check-20261014
Oct 14, 2026
Merged

seathatflowsinourveins merged 1 commit into
mainfrom
codex/model-landscape-check-20261014

Conversation

@seathatflowsinourveins

Copy link
Copy Markdown
Collaborator

Scope

The 2026-10-14 UTC rollover made Fable 5.1's September 1 release 43 days old and blocked every branch's publication validator. Add first-hand sourced landscape checks for Fable and GPT-6 Astra, the only other model within seven days of the 42-day release limit (41 days old).

  • Base commit: ef2fea184b86a06005ef079a88f6b023d028e361.
  • Lane: lane:foundation.
  • Owned paths: catalogs/foundation/model-currency.json, its native registration in manifests/evidence.json, and evidence/artifacts/model-landscape-check-20261014/receipt.json.
  • CC incident cap exception applies to this PR only; other currency PR work remains held.

SOTA sources

All pages were fetched first-hand on 2026-10-14 UTC:

  • Anthropic models comparison, Fable 5.1 model page, and September 1 release entry: claude-fable-5-1 remains the named Fable line; no newer Fable ID appears in these sources. The other listed Claude lines and same-generation restricted Mythos counterpart are recorded separately in the reason.
  • Official OpenAI documentation: model overview and GPT-6 Astra page: gpt-6-astra remains the Astra flagship, with no newer Astra ID shown; Sol and Luna remain separate named lines.
  • Repository policy at the base above: scripts/validate.py::model_currency_problems, 42-day release age and seven-day landscape check schema. Native evidence registration follows scripts/host_receipts.py::register_file.

Both newer_candidates arrays are empty for their named model lines; no successor is invented. Release dates, existing selections, consumer assignments, historical checks and validation policy remain unchanged. This source review does not measure quality, native model execution or account access.

Evidence-class table

Claim Evidence class Command / receipt
Current Fable/Astra IDs and absence of a newer same-line ID in the named official pages source_review First-hand source URLs and observations in incident receipt
Fable age43 and Astra age41 are the required expired/near-limit rows among21 models local_integration Effective-age census follows validator package-bound release rule; receipt
UTC-rollover validation failure before the check local_integration Exact idle validator rc1 and error retained in incident receipt
Model-currency schema/behavior contract synthetic python3 -m unittest tests.test_model_currency,22tests,rc0

Local commands run

Heavy commands run with systemd-run --user --scope -p CPUWeight=idle --quiet --.

python3 scripts/validate.py  # before, fresh origin/main
exit1
Publication validation failed:
model currency model claude-fable-5-1: release older than 42 days needs a sourced landscape check within 7 days

python3 -m unittest tests.test_model_currency
exit0;22tests

python3 scripts/evidence_manifest.py --check
exit0;11943files

gitleaks dir evidence/artifacts/model-landscape-check-20261014 --redact=100 --no-banner
exit0;no leaks

python3 scripts/validate.py  # after, registered PR tree
exit0;71components/11943hashedfiles/4profiles/240receipts

git diff --check
exit0

gitleaks git . --log-opts=ef2fea184b86a06005ef079a88f6b023d028e361..HEAD --redact=100 --no-banner
exit0;onecommit;no leaks

git merge-tree --write-tree origin/main HEAD
exit0

Decision record

evidence/artifacts/model-landscape-check-20261014/receipt.json records the source observations, exact fail-before output, age census, scope and limits. Only the two landscape_check objects are added; the native registry updates one catalog binding and adds one receipt binding.

Separate follow-up already reported to the CC: existing landscape checks dated October7 expire October15. Their earlier KEEP decisions are not silently renewed in this incident fix. MinerU ONNX is34days old today, below the requested35–42day near-limit window.

Host evidence

No host receipt or platform acceptance status changes. Local review and Astra incident-priority read are requested; CC lands after gates.

Checklist

  • No GitHub Actions changes.
  • No secrets printed or added; no new required secret.
  • No paid hosting/subscription/billing change.
  • Peer-owned files and worktrees preserved.
  • Independent local review, Astra read and actual-head CI before CC landing.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

lane:foundation Foundation lane: Claude/Codex setup, hosts, memory, RAG, research, workers

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant