Freshness: staleness budget + fail-loud --verify-live + monthly reality-check cron (#49) - #55
Merged
Merged
Conversation
…ive + monthly cron (#49) The guard chain locked README ⟷ sample but nothing checked sample ⟷ reality: --verify-live only PRINTED drift warnings (exit 0 — a log nobody reads is not a signal) and nothing bounded the snapshot's age, so `captured 2026-07-02` would silently rot into an asserted number (#27/#29) with everything green. - `--check-staleness [--max-age-days N=120] [--today ISO]`: pure date math on captured_at with an INJECTED today — the tests never read the wall clock (a wall-clock assertion would itself be the time bomb). Over budget / future captured_at / malformed dates → loud exit 1. - `--verify-live` now exits 1 when any recorded ground_truth has drifted (behavior change: warn → fail-loud, consistent with the whole guard chain). - `.github/workflows/freshness.yml`: monthly cron + manual dispatch runs both. Deliberately NOT PR CI — a time/network failure must never redden unrelated PRs; a red scheduled run is the maintainer's refresh signal. No ${{ }} expressions (injection-safe by construction). - evals/look-up/README.md: § Refreshing the snapshot — the human procedure both error messages point to (context7's column is a recorded judgment; re-capture is deliberately a human act — stated, not hidden). Tests (TDD, RED→GREEN): +4 staleness + 1 verify-live exit codes → 71 Python.
Via scripts/sync-test-counts.sh after the +5 staleness/verify-live tests. Milestone prose left frozen.
kiki830621
added a commit
that referenced
this pull request
Jul 5, 2026
- permissions: contents: read at workflow level — pip install + pytest execute PR-influenced code; don't hand them a write-scoped GITHUB_TOKEN (MEDIUM). - timeout-minutes: 10 on the python job (hang guard). - De-overclaim "matplotlib-version-independent by design" → COORDINATE-independent: the semantic guard compares <text> content, so a matplotlib serialization change (entities/tspan) would false-FAIL loudly, never silently pass; matplotlib gets a <4 upper bound via #49's requirements pin (MEDIUM, ci.yml comment + Testing ×2). - README.md CI line was left stale ("swift only") — now includes the Python suite (MEDIUM). - Guard-scope comment: the two R1 HIGHs (snapshot-SOURCE tampering passes the PR gate; wiki Home numbers zero-guarded) are closed by sibling #49/PR #55 — the freshness workflow gains a pull_request paths trigger on the sample (live registry check on any PR touching it) and a Home-pages guard test; landed there to avoid cross-PR count/file conflicts. Cross-referenced in both issues.
Closes the R1 findings from BOTH ensembles (the two #48 HIGHs land here by design — same guard family, avoids cross-PR count/file conflicts; cross-referenced): - HIGH (#48): snapshot-SOURCE tampering merged green — the PR-gate guards only lock rendered artifacts ⟷ sample. freshness.yml gains a `pull_request: paths:` trigger on the sample/corpus: any PR touching them runs the LIVE registry check as a hard gate; a fabricated ground_truth reddens that PR. Network cost only on the rare sample-touching PR. - HIGH (#48): wiki Home pages' hand-written competitive numbers (0 of 6, 6/6|0/6 table, react v18 vs 19.2.7) had ZERO guard yet publish to the public wiki — new test_home_pages_match_sample derives every token from the sample (exact table row binds both scores in order; prose score; react pair; capture date; both languages). - HIGH (#49): GitHub auto-disables scheduled workflows after ~60d of repo inactivity — the cron dies quietly exactly when the repo goes dormant. Acknowledged in freshness.yml + README with the recovery path (Actions tab / manual run); the paths trigger + manual dispatch are the activity-driven backstops. - MED: --verify-live treated fetch-failure (None) as "no drift" → false GREEN on outage + permanent silent-skip for unsupported ecosystems. Now: per-lib unverified warning, "verified X of N" summary, exit 1 on drift OR unverified (distinct messages). Test extended (None stub → 1). - MED (alarm fatigue): drift on a deliberately fast-moving corpus is EXPECTED, so a monthly hard-red would train the maintainer to ignore it. Scheduled runs now mark the drift step continue-on-error (informational); the staleness budget is the hard red (one deliberate nag per budget). PR runs keep both hard. - MED: staleness tests hardcoded dates against the committed sample's captured_at — the documented refresh procedure would break them. Now pinned via a temp fixture; refresh-proof (README notes it). - MED: root README still said --verify-live "warns" → exit-1 semantics + staleness documented; module docstring likewise (the 2nd stale spot the DA found). - LOW: --today now strict (same ISO round-trip as captured_at; lenient "20261030" rejected, tested); permissions: contents: read + timeout on freshness.yml; 120d rationale documented; "date is an attestation, not a proof" honesty note added; matplotlib capped <4 (the #48 serialization-drift MEDIUM). Tests: +1 (Home guard) → 72 Python / 216 total.
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
1 participant
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.
Refs #49
Summary
Closes the sample ⟷ reality gap the guard chain left open. Two mechanical signals, both run monthly by a new scheduled workflow (and runnable by hand):
--check-staleness(new): exit 1 whencaptured_atexceeds the age budget (default 120d,--max-age-days). Pure date math with an injected--today— tests never read the wall clock (a wall-clock assertion would itself be the time bomb this catches). Future/malformed dates fail loud.--verify-live(behavior change): now exits 1 on drift instead of only printing warnings — a log nobody reads is not a signal..github/workflows/freshness.yml: monthly cron +workflow_dispatch. Deliberately NOT PR CI — a time/network-based failure must never redden unrelated PRs; a red scheduled run is the maintainer's refresh signal.evals/look-up/README.md § Refreshing the snapshot: the human procedure both error messages point to. context7's column is a recorded judgment — re-capture is structurally a human act (stated, not hidden).Tests
TDD (RED→GREEN): +4 staleness (pure math / fresh / over-budget / boundary / future / malformed) + 1
--verify-liveexit codes (registry stubbed). 215 total (144 Swift + 71 Python), counts synced viasync-test-counts.sh.Security
freshness.ymlcontains no${{ }}expressions — triggers areschedule/workflow_dispatch(no event-payload text), static commands + stock actions only.Relation to #48 / PR #54
Complementary halves: #48 = deterministic guards at the PR gate (README⟷sample⟷SVG); #49 = time/network reality checks on a schedule. Separate workflow file → no ci.yml conflict; the only shared file (
Testing.md) is touched in different hunks.Checklist
Refs #49)/idd-verify #49🤖 /idd-all batch (#48 #49) · Do NOT add Closes/Fixes/Resolves.