review-runs-tracking: 2026-11 #1492

Activity

1. dormouse-bot

   added

   review-runs-trackingMonthly tracking issue for review-runs below-threshold findingsMonthly tracking issue for review-runs below-threshold findings

   on Nov 1, 2026
2. dormouse-bot commented on Nov 1, 2026

   

   dormouse-bot

   on Nov 1, 2026

   Last edited by dormouse-bot

   CollaboratorAuthor

   More actions

   Run 41316592241 — 2026-11-01T14:40:41Z

   Window 2026-10-31T13:54:05Z → run start (~24.5 h). 88 tend runs censused (not a page boundary): tend-mention 46 (39 success, 7 skipped), tend-review 14, tend-mention-relay 11, tend-ci-failures 10, tend-notifications 5, nightly 1, review-runs 1. Zero failures. The longest run was nightly 41311085957 at 22 min against the default 360-min cap, so nothing ran long or near the timeout. Token report: $20.04 over 22 priced sessions: tend-review 14 / $8.32, nightly $7.20, notifications 3 / $2.07, review-runs $1.31, mention 3 / $1.14. No outliers: the costliest review was $0.98 (#1475).

   Argos flagged 3 NetworkSettings screenshots on every PR because the stories read the real clock. Acted on: #1493

   • Evidence level: High (deterministic, reproduced)
   • Occurrences this run: 4 Argos builds (2071–2074, on bot PRs Restart a Copy button's feedback timer on a repeat click #1490 and Correct test comments that claim more than their tests check #1491)
   • Run ID: 41316592241
   • Workflow: https://github.com/diffplug/dormouse/actions/runs/41316592241
   • Detail: Nightly 41311085957 attributed the matching 3-screenshot diff on its comment- and test-only PRs to "drift" without finding a cause. Argos's public GraphQL (app.argos-ci.com/graphql, no auth) names the screenshots as modals-networksettings--{anywhere-enrolled,local-networks-enrolled,self-host}. Comparing the base and compared images of build 2073 shows one difference: "Paired Oct 31" against "Paired today". The stub's pairedDevices stamps the phone at Date.now() - 1h, and main's baseline (builds 2063/2064 for ef7a2be) rendered at 00:19 UTC, so every PR built after 01:00 differs. This is the same class of bug Pin the access-log story's timestamps to the story clock #1487 fixed for RemoteControlSection. Pin the NetworkSettings stories to a fixed clock so Argos stops flagging "Paired today" #1493 pins the NetworkSettings meta clock. Moving STORY_NOW to 00:10 temporarily reproduces "Paired Jan 14", which shows the pin now controls that line. All 20 stories pass in chromium. On Pin the NetworkSettings stories to a fixed clock so Argos stops flagging "Paired today" #1493 everything is green except Argos, which flags exactly the three predicted screenshots (builds 2075/2076) and waits for a maintainer to accept them. Next sweep: if Pin the NetworkSettings stories to a fixed clock so Argos stops flagging "Paired today" #1493 has merged, rerun Argos on Restart a Copy button's feedback timer on a repeat click #1490 and Correct test comments that claim more than their tests check #1491. Refuse a .. step behind a backslash in a browser binary path #1479 and Correct stale comments left by alert deferral's removal and dor remote routing #1480 need main merged in, because their diff, remotecontrolsection--paired-phones-and-access-log, predates Pin the access-log story's timestamps to the story clock #1487 on their branches. Deferred until Pin the NetworkSettings stories to a fixed clock so Argos stops flagging "Paired today" #1493 lands, so the merge does not just swap one stale diff for another.
   • Lesson for sessions: when Argos is red on a PR whose diff cannot reach a story, query the build's screenshotDiffs on the public GraphQL endpoint and compare the images instead of writing it off as drift.

   Sessions sampled: sound

   • Nightly 41311085957 ($7.20): it opened chore: update tend workflows (0.3.22 → 0.3.23) #1488 (tend 0.3.23), Test the pgstencil provenance gate before a Hosted production deploy #1489 (a verifyPackages test, mutation-checked), Restart a Copy button's feedback timer on a repeat click #1490 (CopyButton timer, with a test that fails first) and Correct test comments that claim more than their tests check #1491 (test-comment accuracy). It refreshed tend check: configuration drift on diffplug/dormouse #1069. The work was sound, apart from the Argos attribution above.
   • Notifications 41235405949: it saw that Pin the access-log story's timestamps to the story clock #1487 had re-baselined main, reran Argos on Log a one-time connection's end when it was replaced or disposed #1483 as a control, then on Abandon a Surface move whose late iframe consent would replace a newer question #1482 and fix: say an unsaved VS Code revocation lasts until the broker window closes #1485, and all three went green. That was good recovery without a push. Notifications 41262629717 correctly acknowledged 8 threads that were already handled, without posting.

   Maintainer corrections

   review-runs-corrections ran: 5 dispositions (#1478, #1481, #1482, #1483 and #1485, all bot PRs, merged), 16 comment rows and 0 review rows. The nedtwigg rows (#1473 ×3) apply bot findings. The rest are Cloudflare deploy comments and the audit status. No maintainer corrections.

   Live scan

   • Issues: none needs a worker. Security audit status #1137 turned PASS→FAIL at 11:30Z with one embargoed ledger finding (disclosure 11-08), which is maintainer-owned. VS Code: a revocation whose save failed lasts until the broker window closes, not until Dormouse quits #1484 stays open by design. [workflow-audit] 1 unexplained change(s) on 2026-10-31 #1486 was answered last sweep. Track Tauri GTK upgrade to remediate glib advisory #663 is maintainer-owned.
   • PRs: chore: update tend workflows (0.3.22 → 0.3.23) #1488 and Test the pgstencil provenance gate before a Hosted production deploy #1489 are green and await a maintainer (restricted). Refuse a .. step behind a backslash in a browser binary path #1479, Correct stale comments left by alert deferral's removal and dor remote routing #1480, Restart a Copy button's feedback timer on a repeat click #1490 and Correct test comments that claim more than their tests check #1491 are red only on Argos (see above). Drafts Design account-wide devices: the Relay keeps no device list #1473–Raise a prompt on a stray key in command mode instead of passing it #1476 have reviews at their live heads.
   • Default branch: red-default-branch-runs reached back to 2026-10-16T19:23:21Z with no candidates.
   • Waiting runs: only hosted-preview.yml, which is maintainer-gated. Dependabot: glib maybe: drop node-pty for a Rust backend #1 only, tracked by Track Tauri GTK upgrade to remediate glib advisory #663. Outage trackers: none.

   Pre-registered watches

   • tend-notifications under-delivery, occurrence 51: 5 runs in ~24 h against */15. No lever.
   • chore: update tend workflows (0.3.19 → 0.3.22) #1478 security-rule amendment: resolved. The maintainer merged it at 00:12Z, accepting the S41/S61 loosening for setup-runtime. Retire the watch.
   • Nightly/review double-post on nightly's red PRs: not repeated.

   Run 41491233370 — 2026-11-02T17:31:23Z

   Window 2026-11-01T14:31:10Z → run start. 175 tend runs censused (not a page boundary): tend-mention 80 (69 success, 11 skipped), tend-ci-failures 35, tend-review 29, tend-mention-relay 22, tend-notifications 5, nightly, weekly, triage, review-runs 1 each. Zero failures. The longest run was nightly 41475788338 at 28 min against the default 360-min cap. Token report: $39.54 over 38 priced sessions: tend-review 29 / $19.02, nightly $14.39, review-runs $3.05, notifications 3 / $1.49, mention 2 / $0.95, weekly $0.38, triage $0.26. The nightly cost twice the previous one ($7.20). Its log shows four parallel review subagents covering a large merged window (#1479–#1495), which produced 3 PRs. #1505 has merged, and #1503 and #1504 are green. The cost matches the work, so it is not an outlier finding.

   Sessions sampled: all sound

   • Nightly 41475788338 ($14.39): it found and fixed a VS Code terminalContext null crash (Wait out the extension host's bound on a VS Code terminal-context request #1503) and physical keys leaking past the mobile confirm dialog (Keep physical keys out of the mobile terminal while a gesture confirmation is open #1504, reproduced before fixing), and corrected stale comments (Correct stale comments and drop dead re-exports found in the nightly survey #1505, merged). It refreshed tend check: configuration drift on diffplug/dormouse #1069 without commenting.
   • Mention 41482070291 on Update dependency kysely to v0.29.8 #1502: nedtwigg asked it to "fix the merge conflict, merge once CI is green". It resolved the pnpm-lock.yaml conflict as 823bfbe, verified --frozen-lockfile and generate-deps.js, and declined to merge under restricted mode, explaining why. That handling is correct.
   • Review 41477437547 (19 min, Correct stale comments and drop dead re-exports found in the nightly survey #1505): a silent self-review, then a poll with --skip deploy --skip cleanup. This is designed behavior (tend#607).

   Maintainer corrections

   review-runs-corrections ran: 8 dispositions (#1479, #1480, #1488–#1491, #1493 and #1505, all merged), 21 comment rows and 0 review rows. The nedtwigg rows are a fix applying a bot finding on #1495 and the #1502 directive above. No maintainer corrections.

   Live scan

   • Issues: none needs a worker. Red team: November 2026 #1494 (red-team checklist) and Security audit status #1137 (audit back to PASS at 12:39Z) are maintainer- or automation-owned. VS Code: a revocation whose save failed lasts until the broker window closes, not until Dormouse quits #1484 is open by design. [workflow-audit] 1 unexplained change(s) on 2026-10-31 #1486 was answered on 10-31. tend check: configuration drift on diffplug/dormouse #1069 was refreshed by the nightly.
   • PRs: bot PRs Wait out the extension host's bound on a VS Code terminal-context request #1503 and Keep physical keys out of the mobile terminal while a gesture confirmation is open #1504 are green and await a maintainer (restricted). They have no review because self-review is silent by design. Update dependency kysely to v0.29.8 #1502 is green and awaits a maintainer. Draft Make iframe previews local-first, with an external-origin warning and allowlist #1474 is the maintainer's.
   • Default branch: red-default-branch-runs reached back to 2026-10-16T19:23:21Z with no candidates.
   • Waiting runs: only hosted-preview.yml (5 runs, maintainer-gated). Dependabot: glib maybe: drop node-pty for a Rust backend #1 only, tracked by Track Tauri GTK upgrade to remediate glib advisory #663. Outage trackers: none.

   Pre-registered watches

   • Argos clock drift (Pin the NetworkSettings stories to a fixed clock so Argos stops flagging "Paired today" #1493): resolved. Pin the NetworkSettings stories to a fixed clock so Argos stops flagging "Paired today" #1493 merged, and Refuse a .. step behind a backslash in a browser binary path #1479, Correct stale comments left by alert deferral's removal and dor remote routing #1480, Restart a Copy button's feedback timer on a repeat click #1490 and Correct test comments that claim more than their tests check #1491 merged after it. Retire the watch.
   • tend-notifications under-delivery, occurrence 52: 5 runs in ~26 h against */15. No lever.

   Run 41664275438 — 2026-11-03T15:39:58Z

   Window 2026-11-02T17:26:42Z → run start (~22 h). 528 tend runs censused (not a page boundary): tend-mention 274 (244 success, 24 skipped, 6 cancelled by concurrency), tend-mention-relay 92, tend-review 91, tend-ci-failures 64, tend-notifications 3, tend-triage 2, nightly 1, review-runs 1. Zero failures. The volume is a heavy maintainer push day (nedtwigg iterating #1515, #1528, #1533–#1540 with many "Merge main" pushes). The longest job was review 41495731021 at 22 min against the default 360-min cap. Token report: $88.63 over 119 priced sessions: tend-review 91 / $63.36, nightly $14.58, mention 21 / $6.73, triage 2 / $1.71, notifications 3 / $1.33, review-runs $0.92. 408 runs had no session artifact (relay, skipped mentions, pre-boot no-ops).

   Review fan-out on #1515 cost $7.33; findings all accepted. Low, record only

   • Evidence level: Low
   • Occurrences this run: 1
   • Run ID: 41664275438
   • Workflow: https://github.com/diffplug/dormouse/actions/runs/41664275438
   • Session: review 41539888040
   • Detail: On a 157-file, ~9k-line PR, the session started three general-purpose review subagents, all on claude-opus-5-5. They used ~3.4x the main session's tokens, roughly $5.7 of the $7.33; the main session alone was ~$1.6, in line with other large reviews. All five findings were fixed by nedtwigg (ee2b1b7, 2eebdc3), so the spend found real bugs. The model chose the fan-out; no skill asks for it. A side observation: token-usage.json token columns count the main session only while cost_usd includes subagents, which is why the row looked anomalous (17.6K output for $7.33). No upstream issue exists for that (searched max-sixty/tend); not filed, since the overlay requires a maintainer go-ahead per cross-repo issue and the cost figure itself is correct.

   Approve-then-withdraw on #1540 over Argos statuses not yet posted. Occurrence 1, Low, stochastic, record only

   • Evidence level: Low
   • Occurrences this run: 1 (plus a designed instance on Harden iframe previews: press reports, service workers, and metadata addresses #1537)
   • Run ID: 41664275438
   • Workflow: https://github.com/diffplug/dormouse/actions/runs/41664275438
   • Session: review 41592136243
   • Detail: Run 41591565510 declined to approve Remove the Terminal Context autorun #1540 because three Argos statuses were red awaiting a decision. Seven minutes later 41592136243 approved 4de3cd944 (a merge of main with no story changes) because Argos had not yet reported on that head, then polled, saw the same three red, and dismissed its approval at 06:58. Net state correct; three bot posts where one would do. This is the designed approve → poll → dismiss flow (cf. Settings → About: name official, development, and self-host builds #1106 in October); the only twist is that the bot's own prior review predicted the red. One occurrence; no instruction change warranted yet. Harden iframe previews: press reports, service workers, and metadata addresses #1537's withdrawal on a spec-lint failure the PR body already named is the same designed flow.

   Sessions sampled: all sound

   • Nightly 41651623976 ($14.58, 18 min): opened Keep a file link plain when its text is the file's name as printed #1553 (test-first fix in external-links.ts, 6 failures before, 97 passing after) and Correct stale comments and copy found in the nightly survey #1554 (comment/copy fixes); both merged unchanged by nedtwigg at 14:39Z. Refreshed tend check: configuration drift on diffplug/dormouse #1069's date only; no tend check --fix.
   • Reviews 41578088005 (Move the one-time phone page to its own origin #1536), 41578156825 (Harden iframe previews: press reports, service workers, and metadata addresses #1537), 41589548825 (Remove the Terminal Context autorun #1540 draft, silent), 41586591829 (Tighten the Relay bind warning, frame cookies, device codes, and the Hosted release check #1533): sound. No duplicate findings or inline replies across the ~40 re-reviews on Tighten the Relay bind warning, frame cookies, device codes, and the Hosted release check #1533–Remove the Terminal Context autorun #1540; the preflight blocked one same-SHA duplicate on Show pages and Tools in Pocket over the encrypted session #1515 (41539892556).
   • Mentions on Show pages and Tools in Pocket over the encrypted session #1515: three "Confirmed" replies to fixes; two landed after the approving run had resolved those threads. Redundant but harmless.

   Maintainer corrections

   review-runs-corrections ran: 5 dispositions (#1503, #1504, #1532, #1553, #1554, all bot PRs, merged), 80 comment rows and 0 review rows. The 28 nedtwigg rows apply bot findings ("Fixed in …", "Right: …") or close #1528/#331/#1484 by maintainer decision; the rest are Cloudflare, Renovate and audit automation. No maintainer corrections.

   Live scan

   • Issues: answered workflow-audit #1555: its one commit, c96f327, is the bot's Load ConvertTo-Json's module while warming Windows PowerShell in CI #1532 (PowerShell warm-up step only), merged by nedtwigg as 4cac744. Security audit status #1137 went PASS→FAIL at 12:10Z (security-local.md -> "Browser panes", private ledger), maintainer-owned; disclosures [audit-disclosure manual-iframe-service-worker-port-reuse/80] WARNING finding #1541–[audit-disclosure b1e6d8171ce7/66] FAIL failed check #1552 closed by automation. Other open issues are maintainer backlog or end with a bot response (Consider supporting OSC 7501 #1079 only gained a label).
   • PRs: no open bot PRs. Take the iframe leader chord only from the frame holding focus #1556 and Remove the Terminal Context autorun #1540 (nedtwigg) are approved at their live heads.
   • Default branch: red-default-branch-runs reached back to 2026-10-16T23:03:39Z; one candidate, security-audit 41631555350, which is the audit's FAIL outcome above, not a bot-fixable failure.
   • Waiting runs: only hosted-preview.yml (5 runs, maintainer-gated). Outage trackers: none.
   • Dependabot: 58 new alerts (Fix windows launching and shell detection #11–Better handling for incorrect characters typed in Command Mode #68) on pnpm-lock.yaml, all transitive, all created 14:37:20–14:37:39Z, about a minute after Move the one-time phone page to its own origin #1536 merged; Move the one-time phone page to its own origin #1536 did not change the lockfile. They are real against resolved versions (e.g. fast-uri@3.1.2 vs GHSA-qw65-cvwx-89v3 fixed in 3.1.7; dompurify@3.4.16 vs critical GHSA-cg2g-xh64-v4q7 fixed in 3.4.17). Under an hour old, so not yet live work. glib maybe: drop node-pty for a Rust backend #1 remains tracked by Track Tauri GTK upgrade to remediate glib advisory #663.

   Pre-registered watches

   • New: Dependabot pnpm-lock.yaml alert batch. Next sweep: if Dependabot or Renovate has opened no PRs for undici/dompurify/fast-uri/brace-expansion/sharp/zod/qs/@azure/msal-node, check for a red dynamic/dependabot/... run and whether Dependabot security updates are enabled; past a few days with no PR, it is live work (bump directly, respecting docs/specs/security-supply-chain.md cooldown).
   • tend-notifications under-delivery, occurrence 53: 3 runs in ~22 h against */15. No lever.