You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@Astro-Han Addressed all 14 review findings from merged #913 on top of #924. The report sanitizer, matrix validator, runtime policy, fixture lifecycle, runner/capability gates, budgets, owned-target proof, terminal qualification, and AX/semantic dispatch proof are now fail-closed. CI is fully green; hardened OpenAI gpt-5.4 L0 and L1 reruns pass. Ready for review.
The reason will be displayed to describe this comment to others. Learn more.
No product-level P0/P1 findings. The issues below affect the reliability of the real-provider evidence harness, so I am treating them as non-blocking P2s:
P2: The stale-window scenario builds its allowlist from the initial window set, so the replacement Current Window can be rejected.
P2: targetOwned can be decided by weaker conditions before fixture-PID trace binding runs.
P2: The L1 gate accepts fewer observations than the scenario prompt requires and evaluates only the qualifying subset of actions.
P2: Direct AppKit AX reports do not satisfy the new canonical report schema.
P2: wait and cursor_position are treated as observation-bound actions even though their schemas do not accept an observation ID.
P2: Driver traces are appended asynchronously without a flush before qualification reads them.
P2: Producer and policy provenance validation remains fail-open when fields are absent or unexpected.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Addresses every evidence-layer P2/P3 from the #913 review on top of merged #924.
real-runtime, status pass, matching provider/model/scenario,complete/end_turn, successful owned actions, budgets, fixture/forbidden evidence, and AX/semantic dispatch proofReal validation
OpenAI
gpt-5.4passed the hardened qualification gate:complete/end_turnclick_element, primary=1, danger=0, duplicate=0, AX dispatch evidenceVerification
npm run test:scripts(73/73 after merged test(cu): extend real AX provider evidence #924)git diff --check