Adopt incremental eval-staleness workflow (agent-eval status/run + content-aware fingerprints)#92
Merged
Merged
Conversation
- Bump @vercel/agent-eval to ^1.3.0. - sync-evals: run `agent-eval refingerprint` after pulling evals instead of the local updateFingerprints (which re-stamped every result and hid real eval changes). Delete scripts/refingerprint.ts — the framework owns this now. - CI: the "which staleness is acceptable" policy lives in this repo, not the framework. scripts/check-stale.mjs reads `agent-eval status --json` and fails on new/changed evals except experiments listed in its ACCEPTED_STALE — so models we intentionally leave on an older eval stay green.
gaojude
force-pushed
the
jude/adopt-eval-staleness
branch
from
311b43c to
102c7d5
Compare
gaojude
changed the title
- pnpm-lock.yaml: resolve the now-published ^1.3.0. - One-time migration: `agent-eval refingerprint` (run via sync-evals at the pinned next.js SHA) stamps `contentFingerprint` into every committed summary.json, so future config-only changes carry forward instead of mass-staling legacy results. - .gitignore: ignore `.env*` (`.env.local` carries the sandbox/gateway token).
|
Review the following changes in direct dependencies. Learn more about Socket for GitHub.
|
jerilynzheng
approved these changes
Jul 3, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Adopt the incremental eval-staleness workflow from
@vercel/agent-eval@1.3.0(vercel-labs/agent-eval#164 + #165).What changes
@vercel/agent-evalto^1.3.0(lockfile resolved against the published release).scripts/sync-evals.ts: after pulling evals, runagent-eval refingerprint(carry config-only changes forward) instead of the localupdateFingerprints, which re-stamped every result and silently hid real eval changes. Deletescripts/refingerprint.ts— the framework owns this now.eval-cache-check): the "which staleness is acceptable" policy lives here, not in the framework. Newscripts/check-stale.mjsreadsagent-eval status --jsonand fails on new/changed evals except experiments listed in itsACCEPTED_STALE.contentFingerprintstamped into every committedsummary.json(1082 files), so future config-only changes carry forward instead of mass-staling legacy results.Workflow after this
Bump the pinned next.js SHA, then:
agent-eval(status + pick which experiments to run) → for any model you're intentionally leaving behind, add it toACCEPTED_STALEinscripts/check-stale.mjs→ CI passes.Verified locally against the published 1.3.0:
pnpm sync-evals 9e4d718…→node scripts/check-stale.mjs→ "Eval cache OK", exit 0.