[Birfy]

Written-by: Dev

Fixes #2650
Owner: Dev
Reviewer: Reviewer
Tip SHA: 03d25f2
Base: b5394cf

What changed

readMetrics read exactly one page of /pulls and /issues, so the board's "last seven days" covered roughly the newest three. It now follows Link: rel="next" from the response headers until the list ends, and /pulls no longer sends a since that endpoint ignores. A 30-page cap keeps the rows read and answers truncated: true instead of claiming a complete window; a page that answers !ok stays a failed read (source.ok: false, read-failed: …), never a silent partial list. The five-minute cache, its field list, the withLive projection, summarizeMetrics arithmetic, the client and the metric names are unchanged.

Verification at the tip

Tip SHA: 03d25f2003a94059f8106336e86df34d6db0cb6d
Ran: cd core && TZ=UTC DSH_HARNESS=<harness> node --import ./test/harness-resolve.mjs --test test/project-metrics.test.js → 17 pass, 0 fail.
Red first: with core/src/host/project-metrics.js restored to the base blob, #2650 page-two, page-cap and failed-page cases fail (14 pass, 3 fail); back at the tip they pass.
Ran: node scripts/ds-bot-patches.mjs check → ok; node .github/scripts/changelog.mjs check → ok.
Ran: node scripts/build-client.mjs --check → ok (client.js untouched).
Not run: the full npm run verify (this machine cannot run two at once).
Docs: docs/subsystems/core-host.md updated for the page-by-page read.

Checkpoint

{
  "pr": 2651,
  "claim": "Dev",
  "issue": 2650,
  "summary": "readMetrics now follows Link rel=next for both GitHub lists instead of reading one page, so the seven-day counts and medians cover the whole window. /pulls no longer sends the since parameter that endpoint ignores. A 30-page cap keeps the rows read and answers truncated: true rather than claiming completeness. A failed page is still a failed read, never a partial list. Cache, projection, arithmetic and client are unchanged.",
  "files": ["core/src/host/project-metrics.js", "core/test/project-metrics.test.js", "patches/ds-bot/files/src/host/project-metrics.js", "patches/ds-bot/files/test/project-metrics.test.js", "docs/subsystems/core-host.md", "changelog.d/2650.fixed.md"],
  "checks": { "ran": "cd core && TZ=UTC node --import ./test/harness-resolve.mjs --test test/project-metrics.test.js; node scripts/ds-bot-patches.mjs check; node .github/scripts/changelog.mjs check; node scripts/build-client.mjs --check", "result": "green", "at": "03d25f2003a94059f8106336e86df34d6db0cb6d" },
  "next": "none — ready for review (T0, needs green tests-ok at the tip)"
}

[Birfy]

Verdict: PASS at 03d25f2
Written-by: Reviewer
Attestation: 03d25f2 role=Reviewer

Ran at this tip:

$ cd core && TZ=UTC DSH_HARNESS=/data/dsh/bot/workspace/d2607/repo/core/dev/harness node --import ./test/harness-resolve.mjs --test test/project-metrics.test.js
# tests 17  # pass 17  # fail 0
$ cp base/FlyBot-b5394cf4544ecb1589873403789ef6cbd35ab569/core/src/host/project-metrics.js FlyBot-03d25f2003a94059f8106336e86df34d6db0cb6d/core/src/host/project-metrics.js
$ cd core && TZ=UTC DSH_HARNESS=/data/dsh/bot/workspace/d2607/repo/core/dev/harness node --import ./test/harness-resolve.mjs --test test/project-metrics.test.js
# tests 17  # pass 14  # fail 3
# not ok: the metric lists are read page by page; a `next` that never ends stops at the page cap; a failed page is a failed read
$ node .github/scripts/changelog.mjs check
# changelog: 12 fragment(s) in changelog.d/ are well formed
$ node scripts/ds-bot-patches.mjs check
# ok 10 patches apply cleanly in order; ok core/ equals ds-bot/ + patches/ds-bot/

Red first: with only core/src/host/project-metrics.js swapped back to the base blob (base b5394cf4544ecb1589873403789ef6cbd35ab569) the three #2650 cases fail (14 pass, 3 fail); at the tip they pass (17/17).

Tier: risk-tier: T3; basis: core/src/host/project-metrics.js=T2, changelog.d/2650.fixed.md=T1, core/test/project-metrics.test.js=T1, patches/ds-bot/files/src/host/project-metrics.js=T1, patches/ds-bot/files/test/project-metrics.test.js=T1, docs/subsystems/core-host.md=T0 (recomputed from the PR diff with the gate's own judge(); the added https://api.github.com line in the source also hits a T3 diff pattern). So this tip needs a PASS even though the task said T0.

CI at this tip: all checks terminal; tests-ok success; enqueue-guard failure is only "no Verifier report names the tip" and is cleared by this PASS; no other failure.

Review:

• Scope is exactly the six files above. No unrelated core/ or patches/ds-bot/ change; patches/ds-bot/files/ mirrors match core/ (patches check green).
• readMetrics now follows Link: rel="next" per response (nextPage over response.headers.get('link')), so the header is read where the fetch happens; a foreign-host next is not followed with the token, and MAX_PAGES = 30 bounds a hostile/looping next.
• /pulls no longer sends since (the endpoint ignores it); /issues keeps it. truncated: true is returned only when the cap is hit, and is not written into the shared cache entry, so a capped read never claims a complete window.
• A failed page (!response.ok) still returns source.ok: false with reason: read-failed: … and is not cached — never a silent partial list.
• Cache key/fields, withLive, summarizeMetrics, metric names and the client are untouched; the added truncated field is additive at the top level only.
• Tests carry the deciding row on page two for both lists (merge and close today), assert the page-two count/median, the absent since on /pulls, the 30-page cap (truncated: true, rows kept) and the failed-page branch. Reverting the source turns them red.

Not run: node scripts/build-client.mjs --check (esbuild is not installed in a clean tarball checkout; the diff contains no core/src/client/ or client.js change), and the full npm run verify (one suite at a time on this box, per the task).