Add Pixel Canary eval results - #128
Merged
Merged
Conversation
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Adds Pixel Canary (
stealth/pixel-canary) through Vercel AI Gateway and OpenCode, with baseline and AGENTS.md experiments, saved results for all 31 current evals, and the regenerated leaderboard export. Registers it as tier 1 without displacing an existing model; other models' exported data is unchanged.Both configurations use four attempts with early exit and a 2,400-second attempt limit. They reuse the existing pinned OpenCode binary with model metadata supplied through
extraProviders. Reasoning effort uses the Gateway default; pricing remains unavailable.Scoring exception: baseline
agent-048-prefetch-url-dataretains three genuine validation failures and one 900-second AI-judge timeout as an acknowledged failure. The 28/31 score therefore includes an ungraded timeout, rather than four genuine model failures for that eval. The classification and retry provenance record this explicitly; the cancelled regrading attempt is excluded.The AGENTS.md variant passes
refresh-settingsandprefetch-url-data, with no regressions. Its only failed eval isuses-nextjs.Validation:
pnpm typecheckpnpm test:cost— 9 tests passednode scripts/check-stale.mjs— Eval cache OKpnpm export-results --check