Skip to content

fix: add conditional GitHub REST reads and quota-aware sweeps - #72

Closed
MrGTV-love wants to merge 9 commits into
mainfrom
fm/fm-github-conditional-reads
Closed

fix: add conditional GitHub REST reads and quota-aware sweeps#72
MrGTV-love wants to merge 9 commits into
mainfrom
fm/fm-github-conditional-reads

Conversation

@MrGTV-love

@MrGTV-love MrGTV-love commented Oct 9, 2026 •
edited
Loading

Copy link
Copy Markdown
Owner

Intent

I would like all bugs to be fixed so tomorrow can be focused entirely to Vernant and not problems preventning Vernant from getting built.

Context: GitHub rate limiting has repeatedly stopped fleet work (no-mistakes push and CI steps failing on gh pr view, a Vernant lane PR report published incomplete with rows missing). The measured scout report data/fm-fleet-github-graphql-budget/report.md (in the main firstmate home) found: the core (REST) bucket runs at about 80 percent of 5000/hour with no unusual load; the open-work ledger in 7 lane homes spent 88 calls per run until those homes were updated to PR #56 (done 2026-10-08); the contributions poll spends about 1150 core calls/hour re-reading 12 unchanged PRs; conditional GETs with If-None-Match that return 304 do not count against the limit; nothing in tracked code reads the remaining quota or degrades when it is low; gh api rate_limit reads wrong on this account - use X-RateLimit-* headers. GraphQL was exhausted on 2026-10-06 ~21:26Z, 2026-10-08 02:53Z and ~04:45Z by a burst of about 8x baseline whose consumer is unknown (the ledger and no-mistakes are ruled out).

What Changed

  • Route contributions, open-work ledger, PR review-list, and reviewer-suggestion REST reads through a shared ETag cache with conditional GETs, cached 304 responses, and pagination support.
  • Record quota from response headers and stop background sweeps below a configurable floor (default 15%), preserving prior observations as stale with reset-time diagnostics instead of publishing partial results.
  • Add HTTP-faithful fixtures and regressions for caching, concurrent responses, quota refusal, and recovery; update test-runner routing and configuration documentation.

Risk Assessment

✅ Low: The change is bounded to conditional REST reads and explicit quota-stale handling, with no substantiated material defects found in the complete source review.

Testing

Inspected the change and exercised ten live CLI/API scenarios using real firstmate commands and installed gh against an isolated local REST/GraphQL service. Final scenarios passed after correcting disposable setup issues; supplemental shim and deadline regressions also passed, with longer isolated timeouts needed to reach the named deadline readers. Saved command transcripts, HTTP responses, and persisted state, then removed temporary setup. No tracked files, production credentials, or fleet sessions were changed; no full suite, lint, or static analysis was run.

  • Live validation: ✅ go - 10 of 10 scenarios driven live against the product
Scenario Result Live Evidence
Read unchanged and changed paginated resources with safe cache fallback ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-01.json
Concurrent 304 serves its validated generation without rewriting another generation ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-02.json
HTTP failure and fresh or cached pagination refuse safely below the quota floor ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-03.json
Concurrent quota responses retain the minimum and reject older windows ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-04.json
Poll unchanged PRs conditionally and discover a new maintainer signal on an appended page ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-05.json
Quota-stale polling keeps observations, suppresses late-owner repeats, and recovers after reset ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-06.json
A completed low-quota issue remains measured while budget-skipped tail owners become stale ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-07.json
A low-quota PR wave blocks GraphQL and marks every unmeasured tail owner ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-08.json
Ledger keeps all published rows stale under quota refusal and refreshes after reset ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-09.json
PR advisory readers report all paginated blockers and reviewer evidence even below the sweep floor ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-10.json
Evidence: Live product command transcripts and HTTP response sequences

Source: Live product command transcripts and HTTP response sequences

LIVE PRODUCT VALIDATION
Real tracked firstmate commands and installed gh; disposable loopback HTTPS API and isolated fleet homes.
No public GitHub requests, production data, fleet sessions, credentials, linters, or broad repository suite were used.
Earlier setup attempts are preserved separately; final scenarios below reflect corrected disposable setups.

SCENARIO 1: Read unchanged and changed paginated resources with safe cache fallback
Observed result: pass
API responses: GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2=304 GET /repos/o/r/items?per_page=2=304 GET /repos/o/r/items?per_page=2&page=2=200 GET /repos/o/r/items?per_page=2=304 GET /repos/o/r/items?per_page=2&page=2=304 GET /repos/o/r/items?per_page=2=304 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2=304 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/commits?sha=bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb&path=dir%2Fa%20b.ts&per_page=100=200
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":1},{"id":2}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":1},{"id":2}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":1},{"id":2}],[{"id":3}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":1},{"id":2}],[{"id":3}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":1},{"id":2}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2
[{"id":9}]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2
[{"id":9}]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2
[{"id":9}]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2
[{"id":9}]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2
[{"id":9}]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/commits -f sha=bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb -f path=dir/a b.ts -f per_page=100
[{"sha":"aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa","author":{"login":"carol"}}]
exit=0

SCENARIO 2: Concurrent 304 serves its validated generation without rewriting another generation
Observed result: pass
API responses: GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2=304 GET /repos/o/r/items?per_page=2&page=2=200 GET /repos/o/r/items?per_page=2=304 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2&page=2=200 GET /repos/o/r/items?per_page=2=304 GET /repos/o/r/items?per_page=2&page=2=304 GET /repos/o/r/items?per_page=2=304
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2
[{"id":1}]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":3}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2
[{"id":2}]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":3}],[{"id":4}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":2}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2
[{"id":1}]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":3}],[{"id":4}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2
[{"id":2}]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":3}],[{"id":4}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":2}]]
exit=0

SCENARIO 3: HTTP failure and fresh or cached pagination refuse safely below the quota floor
Observed result: pass
API responses: GET /repos/o/r/error=502 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2&page=2=200 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2=200 GET /repos/o/r/items?per_page=2&page=2=200 GET /repos/o/r/items?per_page=2=304 GET /repos/o/r/items?per_page=2=304
$ bin/fm-gh-rest.sh get /repos/o/r/error
stderr: gh: isolated service failure (HTTP 502)
exit=1
$ bin/fm-gh-rest.sh guard
GitHub core quota low (600 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T13:47:46Z
exit=75
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --floor
stderr: GitHub core quota low (600 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T13:47:46Z
exit=75
$ bin/fm-gh-rest.sh guard
exit=0
$ bin/fm-gh-rest.sh guard
GitHub core quota low (600 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T13:47:46Z
exit=75
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":1}],[{"id":2}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --floor --paginate --slurp
stderr: GitHub core quota low (500 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T14:47:46Z
exit=75
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --paginate --slurp
[[{"id":1}],[{"id":2}]]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --floor --paginate --slurp
stderr: GitHub core quota low (500 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T14:47:46Z
exit=75
$ bin/fm-gh-rest.sh get /repos/o/r/items?per_page=2 --floor
[{"id":1}]
exit=0

SCENARIO 4: Concurrent quota responses retain the minimum and reject older windows
Observed result: pass
API responses: GET /repos/o/r/quota-seed=200 GET /repos/o/r/quota-low=200 GET /repos/o/r/quota-high=200 GET /repos/o/r/old-window=200 GET /repos/o/r/new-window=200
$ bin/fm-gh-rest.sh get /repos/o/r/quota-seed
[]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/quota-low
[]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/quota-high
[]
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r/old-window
[]
exit=0
$ bin/fm-gh-rest.sh guard
GitHub core quota low (749 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T13:47:46Z
exit=75
$ bin/fm-gh-rest.sh get /repos/o/r/new-window
[]
exit=0
$ bin/fm-gh-rest.sh guard
exit=0

SCENARIO 5: Poll unchanged PRs conditionally and discover a new maintainer signal on an appended page
Observed result: pass
API responses: GET /repos/o/r/pulls/8=200 GET /repos/o/r/commits/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/statuses?per_page=100=200 GET /repos/o/r/pulls/8/comments?per_page=100=200 GET /repos/o/r/issues/8/comments?per_page=100=200 GET /repos/o/r/commits/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/check-runs?filter=all&per_page=100=200 GET /repos/o/r=200 GET /repos/o/r/pulls/8/reviews?per_page=100=200 POST /graphql=200 GET /repos/o/r/pulls/8=304 GET /repos/o/r/pulls/8/reviews?per_page=100=304 GET /repos/o/r/commits/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/statuses?per_page=100=304 GET /repos/o/r=304 GET /repos/o/r/pulls/8/comments?per_page=100=304 GET /repos/o/r/issues/8/comments?per_page=100=304 GET /repos/o/r/commits/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/check-runs?filter=all&per_page=100=304 POST /graphql=200 GET /repos/o/r/pulls/8=304 GET /repos/o/r/commits/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/check-runs?filter=all&per_page=100=304 GET /repos/o/r/pulls/8/reviews?per_page=100=304 GET /repos/o/r=304 GET /repos/o/r/pulls/8/comments?per_page=100=304 GET /repos/o/r/commits/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/statuses?per_page=100=304 GET /repos/o/r/issues/8/comments?per_page=100=304 GET /repos/o/r/issues/8/comments?per_page=100&page=2=200 POST /graphql=200
$ bin/fm-contributions.sh poll
exit=0
$ bin/fm-contributions.sh poll
exit=0
$ bin/fm-contributions.sh poll
contribution-wake: check: contributions delivery 81d47dbd8d0ea5c0f95bf28be43c302f585568a08cccefb238a4ea8d7153353a
exit=0
$ bin/fm-fleet-snapshot.sh --contribution-input
{
  "backlog": {
    "present": true,
    "records": [
      {
        "captain_drop": false,
        "completion": {
          "date": null,
          "verb": null
        },
        "hold_set": null,
        "body_excerpt": null,
        "merged": null,
        "title": "Lab contribution",
        "local_note": null,
        "

... [17620 bytes truncated] ...

ation/ledger-live", "complete": true, "rows": [{"id": "0bb7cb632cd5d10acc5c68be", "category": "ready_not_started", "subject": "delivery", "owner": "firstmate", "next_action": "informational: queued and not selected for dispatch", "age_seconds": null, "limit_seconds": 1800, "overdue": false, "evidence": "", "informational": true}, {"id": "4b42e7d0e8233623e839ddeb", "category": "red_check", "subject": "https://github.com/o/r/pull/8", "owner": "firstmate", "next_action": "diagnose: code or test", "age_seconds": 0, "limit_seconds": 0, "overdue": true, "evidence": "build"}]}
exit=0
$ bin/fm-open-loops.sh --heartbeat --json
{"schema": "fm-open-loops.v1", "generated_epoch": 1791550445, "home": "~/.no-mistakes/worktrees/32d18ed9638d/01M4FKQKCEVXQYBJPMP1A0C6BN/.conditional-live-validation/ledger-live", "complete": true, "rows": [{"id": "0bb7cb632cd5d10acc5c68be", "category": "ready_not_started", "subject": "delivery", "owner": "firstmate", "next_action": "informational: queued and not selected for dispatch", "age_seconds": null, "limit_seconds": 1800, "overdue": false, "evidence": "", "informational": true}, {"id": "4b42e7d0e8233623e839ddeb", "category": "red_check", "subject": "https://github.com/o/r/pull/8", "owner": "firstmate", "next_action": "diagnose: code or test", "age_seconds": 2, "limit_seconds": 0, "overdue": true, "evidence": "build"}]}
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r
{"permissions":{"push":false}}
exit=0
$ bin/fm-open-loops.sh --heartbeat --json
{"schema": "fm-open-loops.v1", "generated_epoch": 1791550445, "home": "~/.no-mistakes/worktrees/32d18ed9638d/01M4FKQKCEVXQYBJPMP1A0C6BN/.conditional-live-validation/ledger-live", "complete": false, "rows": [{"id": "81efc51801d899bfb6a622b7", "category": "coverage", "subject": "ledger stale", "owner": "firstmate", "next_action": "wait for the GitHub quota window to reset; the next run refreshes every row", "age_seconds": null, "limit_seconds": 0, "overdue": false, "evidence": "GitHub core quota low (500 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T13:54:03Z", "informational": true}, {"id": "0bb7cb632cd5d10acc5c68be", "category": "ready_not_started", "subject": "delivery", "owner": "firstmate", "next_action": "informational: queued and not selected for dispatch", "age_seconds": null, "limit_seconds": 1800, "overdue": false, "evidence": "", "informational": true, "stale": true}, {"id": "4b42e7d0e8233623e839ddeb", "category": "red_check", "subject": "https://github.com/o/r/pull/8", "owner": "firstmate", "next_action": "diagnose: code or test", "age_seconds": 2, "limit_seconds": 0, "overdue": true, "evidence": "build", "stale": true}], "stale": true, "stale_reason": "GitHub core quota low (500 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T13:54:03Z", "stale_until_epoch": 1791554043}
exit=0
$ bin/fm-gh-rest.sh get /repos/o/r
{"permissions":{"push":false}}
exit=0
$ bin/fm-open-loops.sh --heartbeat --json
{"schema": "fm-open-loops.v1", "generated_epoch": 1791550448, "home": "~/.no-mistakes/worktrees/32d18ed9638d/01M4FKQKCEVXQYBJPMP1A0C6BN/.conditional-live-validation/ledger-no-prior", "complete": false, "rows": [{"id": "d7b6078e39ab8b786a537a1f", "category": "coverage", "subject": "ledger degraded", "owner": "firstmate", "next_action": "restore the unreadable sources and rerun the reconciler", "age_seconds": null, "limit_seconds": 0, "overdue": true, "evidence": "GitHub core quota low (500 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T13:54:03Z"}], "stale": true, "stale_reason": "GitHub core quota low (500 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T13:54:03Z", "stale_until_epoch": 1791554043}
exit=0
$ bin/fm-open-loops.sh --heartbeat --json
{"schema": "fm-open-loops.v1", "generated_epoch": 1791550448, "home": "~/.no-mistakes/worktrees/32d18ed9638d/01M4FKQKCEVXQYBJPMP1A0C6BN/.conditional-live-validation/ledger-live", "complete": true, "rows": [{"id": "0bb7cb632cd5d10acc5c68be", "category": "ready_not_started", "subject": "delivery", "owner": "firstmate", "next_action": "informational: queued and not selected for dispatch", "age_seconds": null, "limit_seconds": 1800, "overdue": false, "evidence": "", "informational": true}, {"id": "4b42e7d0e8233623e839ddeb", "category": "red_check", "subject": "https://github.com/o/r/pull/8", "owner": "firstmate", "next_action": "diagnose: code or test", "age_seconds": 5, "limit_seconds": 0, "overdue": true, "evidence": "build"}]}
exit=0
$ bin/fm-open-loops.sh --heartbeat --json
{"schema": "fm-open-loops.v1", "generated_epoch": 1791550448, "home": "~/.no-mistakes/worktrees/32d18ed9638d/01M4FKQKCEVXQYBJPMP1A0C6BN/.conditional-live-validation/ledger-live", "complete": false, "rows": [{"id": "81efc51801d899bfb6a622b7", "category": "coverage", "subject": "ledger stale", "owner": "firstmate", "next_action": "wait for the GitHub quota window to reset; the next run refreshes every row", "age_seconds": null, "limit_seconds": 0, "overdue": false, "evidence": "GitHub core quota low (500 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T13:54:03Z", "informational": true}, {"id": "0bb7cb632cd5d10acc5c68be", "category": "ready_not_started", "subject": "delivery", "owner": "firstmate", "next_action": "informational: queued and not selected for dispatch", "age_seconds": null, "limit_seconds": 1800, "overdue": false, "evidence": "", "informational": true, "stale": true}, {"id": "4b42e7d0e8233623e839ddeb", "category": "red_check", "subject": "https://github.com/o/r/pull/8", "owner": "firstmate", "next_action": "diagnose: code or test", "age_seconds": 5, "limit_seconds": 0, "overdue": true, "evidence": "build", "stale": true}], "stale": true, "stale_reason": "GitHub core quota low (500 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T13:54:03Z", "stale_until_epoch": 1791554043}
exit=0

SCENARIO 10: PR advisory readers report all paginated blockers and reviewer evidence even below the sweep floor
Observed result: pass
API responses: POST /graphql=200 POST /graphql=200 POST /graphql=200 POST /graphql=200 POST /graphql=200 GET /repos/o/r/pulls/8/reviews?per_page=100=200 GET /repos/o/r/pulls/8/reviews?per_page=100&page=2=200 POST /graphql=200 POST /graphql=200 POST /graphql=200 GET /repos/o/r/pulls/8/reviews?per_page=100=304 GET /repos/o/r/pulls/8/reviews?per_page=100&page=2=304 GET /repos/o/r/pulls/8=200 GET /repos/o/r/pulls/8/files?per_page=100=200 GET /repos/o/r/pulls/8/files?per_page=100&page=2=200 GET /repos/o/r/commits?sha=bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb&path=dir%2Fa%20b.ts&per_page=100=200 GET /repos/o/r/commits?sha=bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb&path=b.ts&per_page=100=200 GET /repos/o/r/pulls/8=304 GET /repos/o/r/pulls/8/files?per_page=100=304 GET /repos/o/r/pulls/8/files?per_page=100&page=2=304 GET /repos/o/r/commits?sha=bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb&path=dir%2Fa%20b.ts&per_page=100=304 GET /repos/o/r/commits?sha=bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb&path=b.ts&per_page=100=304
$ bin/fm-pr-state.sh https://github.com/o/r/pull/8
CHECKS: none reported yet
REVIEW DECISION: CHANGES_REQUESTED
REVIEW: maintainer CHANGES_REQUESTED
STALE BLOCKING REVIEW: oldreviewer CHANGES_REQUESTED at bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb
exit=0
$ bin/fm-pr-state.sh https://github.com/o/r/pull/8
CHECKS: none reported yet
REVIEW DECISION: CHANGES_REQUESTED
REVIEW: maintainer CHANGES_REQUESTED
STALE BLOCKING REVIEW: oldreviewer CHANGES_REQUESTED at bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb
exit=0
$ bin/fm-pr-reviewers.sh https://github.com/o/r/pull/8
carol	2 recent commits
bob	1 recent commit
exit=0
$ bin/fm-pr-reviewers.sh https://github.com/o/r/pull/8
carol	2 recent commits
bob	1 recent commit
exit=0
$ bin/fm-gh-rest.sh guard
GitHub core quota low (500 of 5000 left, below the 15% floor); network sweep skipped until the window resets at 2026-10-09T13:55:29Z
exit=75
Evidence: Supplemental PR-check deadline: named requests, trigger, and degraded ledger

Source: Supplemental PR-check deadline: named requests, trigger, and degraded ledger

{
  "kind": "supplemental fixture deadline regression (not live)",
  "reader": "pr-checks",
  "attempts": [
    [
      "gh",
      "api",
      "-i",
      "repos/test/project/commits/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/check-runs?per_page=100"
    ],
    [
      "gh",
      "api",
      "-i",
      "repos/test/project/commits/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/check-runs?per_page=100"
    ]
  ],
  "trigger": [
    "gh",
    "api",
    "-i",
    "repos/test/project/commits/aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa/check-runs?per_page=100"
  ],
  "degraded_report": {
    "schema": "fm-open-loops.v1",
    "generated_epoch": 1791550301,
    "home": "~/.no-mistakes/worktrees/32d18ed9638d/01M4FKQKCEVXQYBJPMP1A0C6BN/.focused-fixtures-final-dp37dpk8/world/deadline-pr-checks/home",
    "complete": false,
    "rows": [
      {
        "id": "a88889019e0167917e29b354",
        "category": "coverage",
        "subject": "ledger degraded",
        "owner": "firstmate",
        "next_action": "restore the unreadable sources and rerun the reconciler",
        "age_seconds": null,
        "limit_seconds": 0,
        "overdue": true,
        "evidence": "reconciler: collection exceeded its deadline; PR test/project#0: source command failure"
      }
    ]
  }
}
Evidence: Supplemental PR-state deadline: named requests, trigger, and degraded ledger

Source: Supplemental PR-state deadline: named requests, trigger, and degraded ledger

{
  "kind": "supplemental fixture deadline regression (not live)",
  "reader": "pr-state",
  "attempts": [
    [
      "gh",
      "api",
      "-i",
      "repos/test/project/pulls/0"
    ],
    [
      "gh",
      "api",
      "-i",
      "repos/test/project/pulls/1"
    ]
  ],
  "trigger": [
    "gh",
    "api",
    "-i",
    "repos/test/project/pulls/1"
  ],
  "degraded_report": {
    "schema": "fm-open-loops.v1",
    "generated_epoch": 1791550161,
    "home": "~/.no-mistakes/worktrees/32d18ed9638d/01M4FKQKCEVXQYBJPMP1A0C6BN/.focused-fixtures-43kgnorh/deadline-tuned-world/deadline-pr-state/home",
    "complete": false,
    "rows": [
      {
        "id": "8efdb31b4e572da089b8f471",
        "category": "coverage",
        "subject": "ledger degraded",
        "owner": "firstmate",
        "next_action": "restore the unreadable sources and rerun the reconciler",
        "age_seconds": null,
        "limit_seconds": 0,
        "overdue": true,
        "evidence": "reconciler: collection exceeded its deadline; PR state https://github.com/test/project/pull/0: source command failure"
      },
      {
        "id": "cc9f46b2ae861917142da9db",
        "category": "failed_task",
        "subject": "failed-0",
        "owner": "firstmate",
        "next_action": "recover the failed work or record why it ends",
        "age_seconds": null,
        "limit_seconds": 1800,
        "overdue": true,
        "evidence": ""
      }
    ]
  }
}

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

🔧 **Rebase** - 2 issues found → auto-fixed ✅
  • ⚠️ bin/fm-test-run.sh - merge conflict rebasing onto origin/main
  • ⚠️ tests/fm-contributions.test.sh - merge conflict rebasing onto origin/main

🔧 Fix applied.
✅ Re-checked - no issues remain.

🔧 **Review** - 9 issues found → auto-fixed (4) ✅
  • 🚨 bin/fm-gh-rest.sh:79 - The lowest-remaining guarantee is not atomic. The six parallel contribution reads can both read remaining=800, then stage responses reporting 749 and 750 for the same reset; if the 750 writer renames last, guard permits reads despite having observed quota below the 15% floor. Serialize the entire read/merge/publication at record_rate, and prevent an older reset window from replacing a newer one. Related sites: bin/fm-gh-rest.sh:82 (minimum calculation), :89 (publication), bin/fm-contributions.sh:295-307 (concurrent writers), and tests/fm-gh-rest.test.sh:179-186 (the 'parallel' regression currently executes both requests sequentially).
  • 🚨 bin/fm-gh-rest.sh:193 - The floor is checked once per command, not before every network read. If page 1 records 500/5000 remaining and supplies a next link, the pagination loop still fetches subsequent pages. A contribution wave can likewise finish successfully with low quota recorded, then execute the unconditional GraphQL head check. Enforce the floor immediately before each new request and preserve exit 75 through callers so the existing stale handling runs. Related sites: bin/fm-gh-rest.sh:206 (page fetch and exit-code conversion), bin/fm_open_loops.py:480-483 (separate guard followed by a get without --floor; must propagate QuotaLow), and bin/fm-contributions.sh:235-238 and :309 (REST-only guard leaves the subsequent GraphQL read unguarded).
  • 🚨 bin/fm-gh-rest.sh:134 - A 304 always uses the cached next link and ignores the current response's Link header. Given a cached full comments page with next=null, followed by an appended comment on page 2 and a 304 for unchanged page 1 carrying rel="next", the helper returns only page 1 successfully. Contributions then advances checked_at without discovering the new signal. The new fixture permits exactly this response sequence at tests/fm-gh-rest.test.sh:40-44. Merge current pagination metadata at the shared 304 boundary while retaining cached metadata when omitted. Related sites: bin/fm-gh-rest.sh:148-151 (200 metadata parsing/storage), :208 (traversal), bin/fm-contributions.sh:295-303 and :333-335 (paginated observations), bin/fm_open_loops.py:483, bin/fm-pr-state.sh:137, and bin/fm-pr-reviewers.sh:69 (other paginated consumers).
  • ⚠️ bin/fm-contributions.sh:440 - Quota refusal marks only the current URL, while the next iteration still requires the full observation reserve. With two recently checked PRs, a first observation that spends more than five seconds before detecting low quota leaves fewer than 15 seconds of the default 20-second budget; the second URL is skipped without a stale error. Its snapshot row remains checked until its old timestamp expires, despite the sweep now being unable to verify it. Separate local quota-stale handling from the network-observation reservation and apply it to every remaining live owner without changing observations or checked_at. Related sites: bin/fm-contributions.sh:434 (early break), :255-269 (per-owner marking), and :475 (announcement); bin/fm-contributions.jq:53-60 explains the resulting incorrect freshness.
  • ⚠️ bin/fm-contributions.sh:264 - Announcement suppression compares the entire reason, which includes remaining quota. After a poll announces 700 remaining, another REST reader can record 600 in the same reset window; the next poll treats the changed reason as a new episode and prints another notification. Continuing foreground reads therefore repeatedly surface the same low-quota episode. Update each row's reason as needed, but determine announcement eligibility from the low-quota window rather than exact message equality. Related sites: bin/fm-contributions.sh:267-269 (reason update and announcement flag), :475 (emission), and bin/fm-gh-rest.sh:52-53 (mutable remaining value embedded in the reason).
  • ⚠️ bin/fm-gh-rest.sh:227 - Simplification: the new status command adds an unused public mode. Neither conditional reads nor the shared quota floor requires a separate bucket-dump interface, and no tracked caller uses it. Recommend removing cmd_status, its dispatch at :238, and its help entries at :7 and :23 rather than extending this additional interface.
  • ⚠️ bin/fm-gh-rest.sh:220 - Simplification: guard's new --resource option expands the floor interface beyond the core bucket used by every intended caller. Recording response resource headers remains useful, but configurable non-core guarding is not required by the stated purpose. Recommend removing this option and its help entry at :6; a core-only guard is the narrower sufficient interface.
  • ⚠️ bin/fm-gh-rest.sh:180 - Simplification: -F is an additional, undocumented alias for -f, with no typed-field behavior and no remaining tracked caller. All migrated query-field callers use -f. Conditional REST reads do not require accepting both spellings; recommend removing -F and retaining the documented -f interface.
  • ⚠️ bin/fm-gh-rest.sh:204 - Simplification: the new MAX_PAGES cutoff introduces unrequired truncation semantics. When the limit is reached with a next endpoint still present, the helper emits the collected prefix and exits successfully, without disclosing incomplete data. Conditional reads and quota protection do not require silently limiting result sets. Recommend removing this component, including its definition at :32, rather than preserving and hardening an additional truncation mode.

🔧 Fix applied.
12 issues (4 errors, 8 warnings) still open:

  • 🚨 bin/fm-gh-rest.sh:79 - The lowest-remaining guarantee is not atomic. The six parallel contribution reads can both read remaining=800, then stage responses reporting 749 and 750 for the same reset; if the 750 writer renames last, guard permits reads despite having observed quota below the 15% floor. Serialize the entire read/merge/publication at record_rate, and prevent an older reset window from replacing a newer one. Related sites: bin/fm-gh-rest.sh:82 (minimum calculation), :89 (publication), bin/fm-contributions.sh:295-307 (concurrent writers), and tests/fm-gh-rest.test.sh:179-186 (the 'parallel' regression currently executes both requests sequentially).
  • 🚨 bin/fm-gh-rest.sh:193 - The floor is checked once per command, not before every network read. If page 1 records 500/5000 remaining and supplies a next link, the pagination loop still fetches subsequent pages. A contribution wave can likewise finish successfully with low quota recorded, then execute the unconditional GraphQL head check. Enforce the floor immediately before each new request and preserve exit 75 through callers so the existing stale handling runs. Related sites: bin/fm-gh-rest.sh:206 (page fetch and exit-code conversion), bin/fm_open_loops.py:480-483 (separate guard followed by a get without --floor; must propagate QuotaLow), and bin/fm-contributions.sh:235-238 and :309 (REST-only guard leaves the subsequent GraphQL read unguarded).
  • 🚨 bin/fm-gh-rest.sh:134 - A 304 always uses the cached next link and ignores the current response's Link header. Given a cached full comments page with next=null, followed by an appended comment on page 2 and a 304 for unchanged page 1 carrying rel="next", the helper returns only page 1 successfully. Contributions then advances checked_at without discovering the new signal. The new fixture permits exactly this response sequence at tests/fm-gh-rest.test.sh:40-44. Merge current pagination metadata at the shared 304 boundary while retaining cached metadata when omitted. Related sites: bin/fm-gh-rest.sh:148-151 (200 metadata parsing/storage), :208 (traversal), bin/fm-contributions.sh:295-303 and :333-335 (paginated observations), bin/fm_open_loops.py:483, bin/fm-pr-state.sh:137, and bin/fm-pr-reviewers.sh:69 (other paginated consumers).
  • ⚠️ bin/fm-contributions.sh:440 - Quota refusal marks only the current URL, while the next iteration still requires the full observation reserve. With two recently checked PRs, a first observation that spends more than five seconds before detecting low quota leaves fewer than 15 seconds of the default 20-second budget; the second URL is skipped without a stale error. Its snapshot row remains checked until its old timestamp expires, despite the sweep now being unable to verify it. Separate local quota-stale handling from the network-observation reservation and apply it to every remaining live owner without changing observations or checked_at. Related sites: bin/fm-contributions.sh:434 (early break), :255-269 (per-owner marking), and :475 (announcement); bin/fm-contributions.jq:53-60 explains the resulting incorrect freshness.
  • ⚠️ bin/fm-contributions.sh:264 - Announcement suppression compares the entire reason, which includes remaining quota. After a poll announces 700 remaining, another REST reader can record 600 in the same reset window; the next poll treats the changed reason as a new episode and prints another notification. Continuing foreground reads therefore repeatedly surface the same low-quota episode. Update each row's reason as needed, but determine announcement eligibility from the low-quota window rather than exact message equality. Related sites: bin/fm-contributions.sh:267-269 (reason update and announcement flag), :475 (emission), and bin/fm-gh-rest.sh:52-53 (mutable remaining value embedded in the reason).
  • ⚠️ bin/fm-gh-rest.sh:227 - Simplification: the new status command adds an unused public mode. Neither conditional reads nor the shared quota floor requires a separate bucket-dump interface, and no tracked caller uses it. Recommend removing cmd_status, its dispatch at :238, and its help entries at :7 and :23 rather than extending this additional interface.
  • ⚠️ bin/fm-gh-rest.sh:220 - Simplification: guard's new --resource option expands the floor interface beyond the core bucket used by every intended caller. Recording response resource headers remains useful, but configurable non-core guarding is not required by the stated purpose. Recommend removing this option and its help entry at :6; a core-only guard is the narrower sufficient interface.
  • ⚠️ bin/fm-gh-rest.sh:180 - Simplification: -F is an additional, undocumented alias for -f, with no typed-field behavior and no remaining tracked caller. All migrated query-field callers use -f. Conditional REST reads do not require accepting both spellings; recommend removing -F and retaining the documented -f interface.
  • ⚠️ bin/fm-gh-rest.sh:204 - Simplification: the new MAX_PAGES cutoff introduces unrequired truncation semantics. When the limit is reached with a next endpoint still present, the helper emits the collected prefix and exits successfully, without disclosing incomplete data. Conditional reads and quota protection do not require silently limiting result sets. Recommend removing this component, including its definition at :32, rather than preserving and hardening an additional truncation mode.
  • 🚨 bin/fm-gh-rest.sh:139 - The body served for a 304 is reread from the shared cache, rather than bound to the ETag sent. Concrete concurrent sequence: request A has a delayed 200 containing reviews V2; request B receives V3 and caches E3/V3; another invocation sends E3 and receives 304, but A publishes E2/V2 before that invocation executes line 139. It successfully returns V2 despite GitHub validating V3, potentially omitting the current blocking reviewer. Round 1's cached-pagination-metadata fix left this race, and its metadata rewrite can also attach an old response's links to a replacement entry. Snapshot and validate one cache generation before requesting, serve that snapshot on 304, and prevent metadata updates from modifying another generation. Related sites: bin/fm-gh-rest.sh:116-118 (validation/ETag reads), :141-145 (pagination read/update), :163 (200 replacement); bin/fm-contributions.sh:300 (core), :304-314 (PR read wave), :342-344 (issue read wave); bin/fm_open_loops.py:482 (ledger); bin/fm-pr-state.sh:137 (reviews); bin/fm-pr-reviewers.sh:53, :69, :83 (core/files/commits).
  • ⚠️ bin/fm-contributions.sh:447 - Round 1's quota-stale-tail fix only latches quota degradation after an explicit request refusal. An issue observation can take six seconds, receive below-floor headers on its final single-page comments/events responses, and finish successfully. With 14 seconds left, the next iteration breaks before checking quota, leaving every tail owner's previously fresh row without the promised quota-stale reason. A PR wave that records low quota and exhausts the deadline has the same outcome because forge checks budget before quota. Check the local quota record at the shared poll boundary before budget exits, including after an observation whose budget refusal could hide the quota transition, and mark unmeasured owners without changing their observations or checked timestamps. Related sites: bin/fm-contributions.sh:230 (budget-first refusal), :342-356 (successful final issue wave), :443-454 (quota latch), :456 (budget skip), :258-278 (owner marking); docs/configuration.md:695 states the budget-independent invariant.
  • ⚠️ bin/fm-contributions.sh:270 - Round 1's quota-episode-announcement correction still decides episode eligibility separately for each owner. After one poll announces a low-quota window and marks existing owners, registering another task for the same PR during that window gives the new owner no prior error. The next poll sets QUOTA_ANNOUNCE again and repeats the same notification, even though existing rows already record that resource/reset episode. Determine announcement eligibility from matching saved rows across the sweep before updating individual owners; continue marking new owners and refreshing reasons silently. Related sites: bin/fm-contributions.sh:264-266 (new-owner fallback), :275 (announcement flag), :440 (per-poll initialization), :488 (emission); docs/configuration.md:695 promises one announcement per resource/reset window. tests/fm-contributions.test.sh:1239-1256 covers remaining-value and window changes but not late-owner registration.

🔧 Fix applied.
✅ Re-checked - no issues remain.

  • ⚠️ tests/assets/gh-http-shim.sh:34 - When gh-backend fails, the preceding ! makes $? zero, so the shim exits successfully instead of propagating the failure. For example, both advisory fixtures exit 91 on unexpected requests, but their installed HTTP shim turns that into exit 0. Capture the original status using an unnegated conditional and an else branch. Related consumers: tests/lib.sh:556 installs the shim; tests/fm-contributions.test.sh:145, tests/fm-pr-state.test.sh:68, and tests/fm-pr-reviewers.test.sh:62 use it. The production helper also checks HTTP status, so this is a test-double fidelity defect, not a demonstrated production bypass.

🔧 Fix applied.
1 warning still open:

  • ⚠️ tests/fm-open-loops.test.sh:1168 - The pr-checks and pr-state deadline cases can pass without reaching their named reader. With fresh state, fm-gh-rest first invokes the fault-injected cat for its cache snapshot, recording attempt 1. After gh successfully returns the PR list, Round 1's response-locking change invokes cat again during lock acquisition, recording attempt 2 and signalling the collector's deadline. The assertions accept this degraded report and never verify that a check or individual PR-state request occurred. Thus these regressions could still pass if deadline propagation through those readers were broken. Install the cat fault stub only for the questions case, and require the targeted failing request to be reached before accepting the deadline result. Related sites: tests/fm-open-loops.test.sh:1116 (both affected cases), :1159-1163 (untargeted attempt counting and alarm), :1174-1179 (false-positive acceptance); bin/fm-gh-rest.sh:128 (snapshot cat), :114 (lock acquisition).

🔧 Fix applied.
✅ Re-checked - no issues remain.

✅ **Test** - passed

✅ No issues found.

  • Live validation: ✅ go - 12 of 12 scenarios driven live against the product
Scenario Result Live Evidence
Repeated REST reads reuse 304 bodies while changes and corrupt caches fetch fresh data ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/01-repeated-rest-reads-reuse-304-bodies-while-changes-and-corrupt-caches-fetch-fresh-data.json
Appending and removing pages behind an unchanged first page produces complete results ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/02-appending-and-removing-pages-behind-an-unchanged-first-page-produces-complete-results.json
Concurrent delayed reads cannot substitute another body or pagination for a validated ETag ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/03-concurrent-delayed-reads-cannot-substitute-another-body-or-pagination-for-a-validated-etag.json
Header-derived quota guards enforce boundaries and stop pagination without partial output ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/04-header-derived-quota-guards-enforce-boundaries-and-stop-pagination-without-partial-output.json
A contribution poll rechecks unchanged PRs conditionally and still publishes new maintainer signals ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/01-a-contribution-poll-rechecks-unchanged-prs-conditionally-and-still-publishes-new-maintainer-signals.json
Low-quota polling preserves every owner and suppresses repeat announcements including late registrations ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/01-low-quota-polling-preserves-every-owner-and-suppresses-repeat-announcements-including-late-registrations.json
After quota reset a contribution poll refreshes the observation and clears its stale error ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/02-after-quota-reset-a-contribution-poll-refreshes-the-observation-and-clears-its-stale-error.json
Quota crossed in a completed read wave blocks GraphQL and marks unmeasured budget-tail owners ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/03-quota-crossed-in-a-completed-read-wave-blocks-graphql-and-marks-unmeasured-budget-tail-owners.json
The open-work ledger holds all dated rows at low quota and clears stale coverage after reset ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/08-the-open-work-ledger-holds-all-dated-rows-at-low-quota-and-clears-stale-coverage-after-reset.json
Reviewer discovery preserves mapped authors and encoded paths across conditional rereads ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/09-reviewer-discovery-preserves-mapped-authors-and-encoded-paths-across-conditional-rereads.json
PR state retains blocking reviews on 304 and reflects later reviewer decisions ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/04-pr-state-retains-blocking-reviews-on-304-and-reflects-later-reviewer-decisions.json
Six concurrent contribution REST reads complete within the existing five-second bound ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/01-six-concurrent-contribution-rest-reads-complete-within-the-existing-five-second-bound.json
  • gh --version — installed gh 2.101.0.
  • bash tests/fm-gh-rest.test.sh — initial 240-second command bound expired; rerun with a 1200-second bound completed successfully.
  • python3 .live-validation/drive.py — exercised real CLI commands against a disposable HTTPS GitHub-compatible endpoint; resumed after the initial runner bound expired.
  • env LIVE_RESUME=1 LIVE_RUN_SUFFIX=-continued python3 .live-validation/drive.py
  • env LIVE_ONLY=contributions,blockers LIVE_TRACE=1 LIVE_RUN_SUFFIX=-diagnostic python3 .live-validation/drive.py — diagnosed read deadlines and corrected disposable GraphQL responses.
  • env LIVE_ONLY=contributions,stale_owners,waves,blockers LIVE_RUN_SUFFIX=-corrected python3 .live-validation/drive.py
  • env LIVE_ONLY=timingdiag,contributions LIVE_BINARY_PATH=1 LIVE_RUN_SUFFIX=-native python3 .live-validation/drive.py
  • env LIVE_ONLY=contributions,stale_owners,waves LIVE_DIRECT=1 LIVE_BINARY_PATH=1 LIVE_RUN_SUFFIX=-direct python3 .live-validation/drive.py
  • env LIVE_ONLY=stale_owners,waves,timingdiag LIVE_DIRECT=1 LIVE_BINARY_PATH=1 LIVE_RUN_SUFFIX=-stateproof python3 .live-validation/drive.py
  • env LIVE_ONLY=contributions,waves LIVE_DIRECT=1 LIVE_BINARY_PATH=1 LIVE_RUN_SUFFIX=-stabilized python3 .live-validation/drive.py
  • env LIVE_ONLY=concurrentbounds LIVE_DIRECT=1 LIVE_BINARY_PATH=1 LIVE_RUN_SUFFIX=-comparison python3 .live-validation/drive.py — compared actual six-way gh and helper reads under the existing five-second bound.
  • env LIVE_ONLY=contributions,recovery,waves LIVE_DIRECT=1 LIVE_BINARY_PATH=1 LIVE_RUN_SUFFIX=-acceptance python3 .live-validation/drive.py — completed contribution acceptance using direct loopback HTTPS and workspace-local symlinks to real installed tools, without bypassing product deadlines.
  • Published acceptance-scenarios.json from the observed successful scenario evidence; stopped local services and ran rm -rf -- .live-validation.

✅ No issues found.

  • Live validation: ✅ go - 10 of 10 scenarios driven live against the product
Scenario Result Live Evidence
Read unchanged and changed paginated resources with safe cache fallback ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-01.json
Concurrent 304 serves its validated generation without rewriting another generation ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-02.json
HTTP failure and fresh or cached pagination refuse safely below the quota floor ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-03.json
Concurrent quota responses retain the minimum and reject older windows ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-04.json
Poll unchanged PRs conditionally and discover a new maintainer signal on an appended page ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-05.json
Quota-stale polling keeps observations, suppresses late-owner repeats, and recovers after reset ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-06.json
A completed low-quota issue remains measured while budget-skipped tail owners become stale ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-07.json
A low-quota PR wave blocks GraphQL and marks every unmeasured tail owner ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-08.json
Ledger keeps all published rows stale under quota refusal and refreshes after reset ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-09.json
PR advisory readers report all paginated blockers and reviewer evidence even below the sweep floor ✅ pass live ~/.no-mistakes/evidence/01M4FKQKCEVXQYBJPMP1A0C6BN/live-scenario-10.json
  • git diff --stat 0a787b17347d5207ee9af3b710dbadc737b19126 2463a3c0e4f563f1c62b08839ec393dca1d3ccce and focused runtime diffs.
  • python3 .conditional-live-validation/drive.py: real fm-gh-rest.sh and installed gh against a disposable loopback HTTPS API; cache reuse, changed data, pagination, corrupt-cache fallback, concurrent replacement, HTTP errors, and quota boundaries.
  • python3 .conditional-live-validation/consumers.py: real contribution polling, appended-page signals, durable wakes, late-owner announcement suppression, and reset recovery.
  • python3 .conditional-live-validation/consumers.py after correcting disposable URL ordering and service-delay isolation.
  • python3 .conditional-live-validation/wave-retry.py: seeded prior observations, delayed REST wave, GraphQL refusal, and preservation of every unmeasured owner's observation and timestamp.
  • python3 .conditional-live-validation/readers.py: real ledger publication, conditional reads, retained stale rows, missing-prior degradation, reset recovery, and mid-pagination refusal.
  • python3 .conditional-live-validation/advisory-retry.py: re-drove PR advisories after correcting the disposable GraphQL schema and isolating gh caches.
  • bash tests/fm-gh-rest.test.sh with worktree-local temporary storage.
  • bash tests/fm-pr-reviewers.test.sh, including backend exit-status propagation through the HTTP shim.
  • Executed only the four deadline-reader cases from tests/fm-open-loops.test.sh:1115-1189 in isolated throwaway runners; verified named requests, matching deadline triggers, and persisted degraded JSON. Short-timeout setups expired before the check reader; longer disposable timeouts reached and passed the exact assertions.
  • bash -x <fixture-copy>/bin/fm-gh-rest.sh get '/repos/test/project/pulls?state=open&per_page=100' --floor --paginate --slurp: diagnosed the deadline fixture's prerequisite.
  • python3 .conditional-live-validation/collect-evidence.py && rm -rf .conditional-live-validation; confirmed no disposable validation directories remained.
🔧 **Document** - 1 issue found ✅
  • ⚠️ docs/fm-test-isolation-proof.md:168 - Recorded concurrency evidence predates the newly admitted fm-gh-rest suite and shared REST-reader changes. The validation phase must refresh the pr-forge family proof and record its results; generating execution evidence is outside this document phase.

✅ No issues found.

🔧 **Lint** - 1 issue found → auto-fixed ✅
  • ⚠️ linter found issues (exit code 1)

🔧 Fix applied.
✅ Re-checked - no issues remain.

✅ No issues found.

✅ **Push** - passed

✅ No issues found.

✅ No issues found.

@MrGTV-love
feat: conditional REST reads and a GitHub quota floor for the ledger …
0ba202c 
…and contributions sweeps

Add bin/fm-gh-rest.sh: GET with If-None-Match from a per-URL ETag cache under
state/ (a 304 serves the cached body and is not counted against the rate
limit), recording X-RateLimit-* from every response. Route fm-contributions,
fm_open_loops, fm-pr-state and fm-pr-reviewers REST reads through it.

Below the floor (FM_GH_RATE_FLOOR_PERCENT, default 15) the contributions poll
keeps each row's last observation marked stale with the reset time, and the
ledger republishes its last rows marked stale instead of publishing partial
rows.
@MrGTV-love
no-mistakes(review): Fix quota races, request guards, pagination, and…
6a52002 
… stale reporting
@MrGTV-love
no-mistakes(review): Fix conditional cache generations and quota stal…
daf0712 
…e episodes
@MrGTV-love
no-mistakes(document): Clarify GitHub conditional-read documentation …
83b7390 
…ownership
@MrGTV-love
no-mistakes(lint): Add missing ShellCheck source directive for wake l…
ccd5afb 
…ibrary
@MrGTV-love
no-mistakes(review): Preserve HTTP shim backend failures with behavio…
465c6d3 
…ral regression coverage
@MrGTV-love
no-mistakes(review): Isolate ledger deadline fixtures and verify name…
2463a3c 
…d-reader alarms
@MrGTV-love
no-mistakes(document): Correct GitHub read guarantees and deadline do…
505614c 
…cumentation
@MrGTV-love
MrGTV-love force-pushed the fm/fm-github-conditional-reads branch from 97f644d to 505614c Compare October 9, 2026 13:22
@MrGTV-love MrGTV-love changed the title fix: reduce GitHub quota usage with conditional REST reads fix: add conditional GitHub REST reads and quota-aware sweeps Oct 9, 2026
@MrGTV-love
no-mistakes(ci): Fixed the reported CI failure by moving contribution…
772cf37 
…s-observer unregistration to the shared bounded-watcher fixture, preventing unrelated observer wakes from preempting PR-security scenarios. Removed redundant per-case cleanup and documented the isolation. The previously failing self-merge scenario and focused transition, retry, and concurrent-publication scenarios pass. Check-unregister tests, shell syntax, and documentation audience checks pass. The scoped CI runner progressed through the reported failure and subsequent retirement/authority cases, but the full security suite remains red at a later, separate teardown fixture failure: it records a nonexistent project, preventing nested-worktree ownership verification. That fixture and teardown implementation are unchanged from the base commit; they were left untouched. Production behavior was not changed
@MrGTV-love

MrGTV-love commented Oct 10, 2026

Copy link
Copy Markdown
Owner Author

Superseded by #88, which merged.

@MrGTV-love MrGTV-love closed this Oct 10, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

No reviews

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

1 participant

@MrGTV-love