# Observed production-provider before/after runtime transcript Worktree: `/home/ubuntu/wt/pr-24406`. Base `6e792117436662208bf2b6eb71ee8d0d88817e43`; candidate `f78d4efb83406ee495171e1252d176384a5aa582`. 40 deterministic matrix cases plus one real public GitCafe payload. Every case calls the actual `GitCafePullRequestProvider.make` → `getChangeRequest` with production GitCafeApi/credential layers, replacing only HttpClient via the existing fakeGitCafe boundary. No copied check normalization or aggregate algorithm. No servers, browser, or live T3 data. All 12/12 matrix rows R1–R6, A1–A2, P1–P4 captured before and after. Candidate assertions verify returned rows, aggregate value AND property presence, HTTP check-request counts, and absence of writes. Equal-ID first-encounter behavior is observed implementation behavior, not an upstream ordering guarantee. Deterministic fixtures deliberately send contradictory `failing:1` counts for complete listings. P3/P4 instead use the four explicit counts tables. `matrix-fixtures.json` contains every input and hand-specified expectation; `proof-driver.test.ts` contains complete synthetic pull/status/auxiliary routes. ## Exact regression and suite results Base command (candidate tests with only the two production source files swapped to base): ```sh vp test run packages/source-control-gitcafe/src/server/gitCafePullRequestJson.test.ts packages/source-control-gitcafe/src/server/GitCafePullRequestProvider.test.ts -t "keeps only the newest run of a re-run check|reads the checks state from each check’s newest run" ``` Observed: exit 1, **2 failed / 47 skipped**. Converter actual `[failure,success]`, expected `[success]`; provider actual `[[Check,failure],[Check,success]]`, expected `[[Check,success]]`. Exact original ASCII command, full failures and stacks are in `base-regressions.log` and `commands.json`. Candidate command: ```sh vp test run packages/source-control-gitcafe/src/server/gitCafePullRequestJson.test.ts packages/source-control-gitcafe/src/server/GitCafePullRequestProvider.test.ts ``` Observed: exit 0, **49 passed**. Separate matrix driver: base exit 0 (observation mode); candidate exit 0 (all expected outputs asserted). The driver writes JSON from returned provider values because vp suppresses successful test stdout. ## Principal stale-count regression: actual JSON before/base ```json { "checks": [ { "name": "Check", "status": "cancelled" }, { "name": "Check", "status": "success" } ], "checksState": "failing" } ``` after/candidate ```json { "checks": [ { "name": "Check", "status": "success" } ], "checksState": "passing" } ``` Input: completed cancelled `99`, completed success `100`, same `Check` identity; status counts `{total:2,failing:1,pending:0}`. **Deterministic incident-shaped fixture**, not a claim of a captured live stale aggregate. ## Expanded matrix: actual rows and aggregate | Case | Before checks | Before state | After checks | After state | Checks GETs before/after | |---|---|---|---|---|---| | R1-failure-reverseFalse | Check:failure; Check:success | failing | Check:success | passing | 1/1 | | R1-failure-reverseTrue | Check:failure; Check:success | failing | Check:success | passing | 1/1 | | R1-cancelled-reverseFalse | Check:cancelled; Check:success | failing | Check:success | passing | 1/1 | | R1-cancelled-reverseTrue | Check:cancelled; Check:success | failing | Check:success | passing | 1/1 | | R2-reverseFalse | Check:failure; Check:success | failing | Check:failure | failing | 1/1 | | R2-reverseTrue | Check:failure; Check:success | failing | Check:failure | failing | 1/1 | | R3-queued | Check:failure; Check:pending | failing | Check:pending | pending | 1/1 | | R3-in_progress | Check:failure; Check:pending | failing | Check:pending | pending | 1/1 | | R3-action_required | Check:action-required; Check:failure | failing | Check:action-required | pending | 1/1 | | R4-githubApp | Check:failure; Check:success | failing | Check:failure; Check:success | failing | 1/1 | | R4-provider | Check:failure; Check:success | failing | Check:failure; Check:success | failing | 1/1 | | R4-name | A:failure; B:success | failing | A:failure; B:success | failing | 1/1 | | R5-status | Check:failure; Check:success | failing | Check:failure; Check:success | failing | 1/1 | | R5-missing | Check:failure; Check:success | failing | Check:failure; Check:success | failing | 1/1 | | R5-null | Check:failure; Check:success | failing | Check:failure; Check:success | failing | 1/1 | | R5-nondecimal | Check:failure; Check:success | failing | Check:failure; Check:success | failing | 1/1 | | R5-mixed | Check:failure; Check:success | failing | Check:failure; Check:success | failing | 1/1 | | R6-bigint | Check:failure; Check:success | failing | Check:success | passing | 1/1 | | R6-equal-reverseFalse | Check:failure; Check:success | failing | Check:failure | failing | 1/1 | | R6-equal-reverseTrue | Check:failure; Check:success | failing | Check:success | passing | 1/1 | | R6-null-absent | Check:failure; Check:success | failing | Check:success | passing | 1/1 | | A1-failure | A:failure; Check:success; Pending:pending | failing | A:failure; Check:success; Pending:pending | failing | 1/1 | | A1-cancelled | A:cancelled; Check:success; Pending:pending | failing | A:cancelled; Check:success; Pending:pending | failing | 1/1 | | A1-queued | A:pending; Check:success | failing | A:pending; Check:success | pending | 1/1 | | A1-action_required | A:action-required; Check:success | failing | A:action-required; Check:success | pending | 1/1 | | A2-success-neutral-skipped | neutral:neutral; skipped:skipped; success:success | failing | neutral:neutral; skipped:skipped; success:success | passing | 1/1 | | A2-empty | [] | failing | [] | omitted | 1/1 | | A2-neutral-skipped | neutral:neutral; skipped:skipped | failing | neutral:neutral; skipped:skipped | omitted | 1/1 | | P1-cross-page | Check:failure; Check:success | failing | Check:success | passing | 2/2 | | P2-limit | Check:failure ×10; Check:success ×10 | failing | Check:success | failing | 10/10 | | P2-repeat | Check:failure ×2; Check:success ×2 | failing | Check:success | failing | 2/2 | | P2-empty | [] | failing | [] | failing | 1/1 | | P3-zero | Check:success ×2 | omitted | Check:success | omitted | 2/2 | | P3-failing-pending | Check:success ×2 | failing | Check:success | failing | 2/2 | | P3-pending | Check:success ×2 | pending | Check:success | pending | 2/2 | | P3-neither | Check:success ×2 | passing | Check:success | passing | 2/2 | | P4-zero | [] | omitted | [] | omitted | 0/0 | | P4-failing-pending | [] | failing | [] | failing | 0/0 | | P4-pending | [] | pending | [] | pending | 0/0 | | P4-neither | [] | passing | [] | passing | 0/0 | ## Real GitCafe cancelled/rerun response Captured read-only using `cafe api` from public [#24508](https://git.cafe/pingdotgg/t3code/pulls/24508), head `ca3661c314bc03e2fede54d1c666fd0082694133`; checks page `next:null`, 32 items. The uploaded sanitized fixture preserves check records and status counts; author identity is replaced by an unavailable actor. Auxiliary reviewer/label/change routes are empty deterministic fixtures. Replay route namespace is remapped to owner/repo/7. No credentials or headers are captured. Actual `Native fingerprint diff`: older `120937060075` cancelled, newer `120937063370` success, provider github, app github-actions. Base returns both; candidate returns success only. Total returned rows: **32 → 30** (also another replaced run). Status counts already report `{total:30,pending:0,failing:0}`, so both aggregates are passing. This real response proves row cleanup; it does **not** reproduce live stale failing counts. ```json { "before": { "id": "LIVE-cancelled-rerun", "checks": [ [ "Build and Typecheck", "success" ], [ "Changes", "success" ], [ "Check", "success" ], [ "CodeRabbit", "success" ], [ "Collect PR targets", "success" ], [ "Deploy relay preview", "skipped" ], [ "Deploy web preview", "skipped" ], [ "EAS Preview", "skipped" ], [ "Label PR 24508", "success" ], [ "Label PR size", "success" ], [ "Lint", "success" ], [ "Macroscope - Approvability Check", "neutral" ], [ "Macroscope - Correctness Check", "success" ], [ "Macroscope - Effect Service Conventions", "skipped" ], [ "Macroscope - Effect Service Conventions", "skipped" ], [ "Mobile Native Static Analysis", "skipped" ], [ "Native fingerprint diff", "cancelled" ], [ "Native fingerprint diff", "success" ], [ "Prepare PR size config", "success" ], [ "Release Smoke", "skipped" ], [ "Rust", "skipped" ], [ "Sync PR size label definitions", "skipped" ], [ "Test Server 1", "success" ], [ "Test Server 2", "success" ], [ "Test Server 3", "success" ], [ "Test Server 4", "success" ], [ "Test Server 5", "success" ], [ "Test Server 6", "success" ], [ "Test Server 7", "success" ], [ "Test Server 8", "success" ], [ "Test Web", "skipped" ], [ "Test", "success" ] ], "checksState": "passing", "checksStatePresent": true, "checkRequests": 1, "writes": [] }, "after": { "id": "LIVE-cancelled-rerun", "checks": [ [ "Build and Typecheck", "success" ], [ "Changes", "success" ], [ "Check", "success" ], [ "CodeRabbit", "success" ], [ "Collect PR targets", "success" ], [ "Deploy relay preview", "skipped" ], [ "Deploy web preview", "skipped" ], [ "EAS Preview", "skipped" ], [ "Label PR 24508", "success" ], [ "Label PR size", "success" ], [ "Lint", "success" ], [ "Macroscope - Approvability Check", "neutral" ], [ "Macroscope - Correctness Check", "success" ], [ "Macroscope - Effect Service Conventions", "skipped" ], [ "Mobile Native Static Analysis", "skipped" ], [ "Native fingerprint diff", "success" ], [ "Prepare PR size config", "success" ], [ "Release Smoke", "skipped" ], [ "Rust", "skipped" ], [ "Sync PR size label definitions", "skipped" ], [ "Test Server 1", "success" ], [ "Test Server 2", "success" ], [ "Test Server 3", "success" ], [ "Test Server 4", "success" ], [ "Test Server 5", "success" ], [ "Test Server 6", "success" ], [ "Test Server 7", "success" ], [ "Test Server 8", "success" ], [ "Test Web", "skipped" ], [ "Test", "success" ] ], "checksState": "passing", "checksStatePresent": true, "checkRequests": 1, "writes": [] } } ``` ## Live search bounds and environment Requested candidate commit checks endpoint returned 0 items. Remote PR head had advanced to `8c1fe57e57479a8dc9eb053ab0d61ae7791a035a`; that head had passing counts. `cafe ci status --pr 24406` reported NOT_FOUND/no CI runs and `cafe ci list --limit 50` returned no items; checks themselves remained accessible. A bounded scan of 20 open PR heads through cafe found the cancelled/rerun above. No GitHub API/gh reads were used. Environment: machine marker fs-18, Linux x86_64, Node v26.12.1, vp v1.3.0, Vitest v5.0.5, Effect 4.0.5, installed worktree dependencies. Only the specified worktree was used for execution. Base production source files were restored byte-for-byte in finally; no branch checkout or HEAD change. Temporary untracked package driver removed. No tracked diff at completion; no commit/push/watch/comment/body edit. Unverified matrix rows: **none**. Bounded provenance gap: no accessible live payload with stale failing status counts was found; the principal aggregate regression is proven with a clearly labeled deterministic fixture and the existing production provider. UI/other clients are outside this server-only dispatcher scope. FRICTION category|happened|prevention Upstream drift|Remote PR advanced after task brief; requested local SHA remained unchanged|Pin local candidate and base, record remote observation separately. Payload provenance|Live cancelled/rerun existed but aggregate counts were already corrected|Keep real row-cleanup proof distinct from deterministic stale-count reproduction. CLI coverage|CI listing returned no runs although commit checks were readable|Use typed cafe first, then cafe api for raw commit-check response bodies. Test stdout|vp suppresses successful test stdout|Driver writes exact returned-value JSON artifacts in addition to logging.