chore(release): qualify 8.5.0 offline and retain a regraded baseline - #81
Merged
Merged
Conversation
Generalize the single hard-coded 8.4.0 picker exception into a fixed source-level registry of individually reviewed offline features, and add planning-models-v1 for 8.5.0 frozen to the reviewed PR #80 head c0064dc. The baseline stays 8.3.0, the last release measured by a live multi-provider campaign and canary. Offline releases deliberately never become baselines: chaining one onto another would let the newest measured evidence recede while each record still claimed a qualified parent. Release notes now state that no release after 8.3.0 has been measured live. The registry is source, not configuration. A record whose version and feature pair is absent is refused before any evidence is read, so it cannot nominate its own scope, borrow another feature's commit, or advance to a later version. No live model eval or paid canary was run for this candidate. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The release grader closure covers prompt surfaces and guides, so PR #80's changes to skills/flow-plan/SKILL.md and src/prompt-surfaces.ts invalidated the seal retained for 8.4.0. Regrade the same original 8.3.0 campaign and canary into a seal valid under current verifier sources. This retains existing evidence; it is not a rerun and produces no new measurement. Each authorized feature now owns its baseline directory in source, so neither a release record nor --bundles can redirect a release to another seal. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
|
You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard. |
…icates Independent review replaced the automated one, which was out of quota. It found no defect but two traps: verifyReleaseEvidence still passed a stale 8.4.0 baseline directory that the verifier now ignores, and no test covered a duplicated in-list guidance path, the only bypass attempt that clears the schema and is caught a layer down in assertPatchPaths. Also assert the registry holds one entry per version and per feature, so a lookup cannot be ambiguous. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Authorizes the reviewed 8.5.0 planning-model selector to release without a live
model campaign, after the operator was told the alternatives were a funded
campaign or holding the release.
What changed
registry of individually reviewed offline features. A record whose version and
feature pair is absent is refused before any evidence is read, so it cannot
nominate its own scope, borrow another feature's commit, or advance to a later
version.
planning-models-v1for 8.5.0, frozen to the reviewed PR feat(models): add planning specialist and explicit model handoff #80 headc0064dc, covering 13 picker, planning, prompt-surface, guide and packagingpaths. Everything else must satisfy the existing patch tooling allowance.
never become baselines: chaining one onto another would let the newest
measured evidence recede while each record still claimed a qualified parent.
release record nor
--bundlescan redirect a release to another seal.Why a new seal was needed
The release grader closure covers prompt surfaces and guides, so PR #80's edits
to
skills/flow-plan/SKILL.mdandsrc/prompt-surfaces.tsinvalidated the sealretained for 8.4.0. That failure was already present on
mainbefore thisbranch. The same original 8.3.0 campaign and canary were regraded into a seal
valid under current verifier sources. This retains existing evidence; it is not a
rerun and produces no new measurement.
Verification
artifact, the frozen source comparison and the regraded baseline, and generated
the disclosure notes. Initialization did not publish.
Disclosure
No live model eval or paid canary ran for this candidate. The generated notes
state that no release after 8.3.0 has been measured live.
🤖 Generated with Claude Code