Skip to content

docs(research): shadow lesson log for B-0449 collision drift#3054

Merged
AceHack merged 1 commit into
mainfrom
docs/lior-shadow-lesson-3052
May 13, 2026
Merged

docs(research): shadow lesson log for B-0449 collision drift#3054
AceHack merged 1 commit into
mainfrom
docs/lior-shadow-lesson-3052

Conversation

@AceHack
Copy link
Copy Markdown
Member

@AceHack AceHack commented May 13, 2026

Lior antigravity check: document alignment drift from PR #3052.

Copilot AI review requested due to automatic review settings May 13, 2026 22:58
Copy link
Copy Markdown

Copilot AI left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Pull request overview

Adds a short research/lesson-log entry documenting an alignment drift observed in PR #3052: the renumbering fix in #3052 picks B-0449, which is already claimed by a different P1 row (B-0449-bg-services-slice-5-subscriber-agent-design-pass-...), producing a second collision rather than resolving the first.

Changes:

  • New research/shadow-lesson log noting the secondary B-0449 collision introduced by PR #3052.
  • Captures analysis (auto-merge armed despite Otto-CLI flag; no ID-lock/worldview-refresh before pick) and the antigravity action taken (Request Changes on #3052, drift report on Lior bus).

@AceHack AceHack merged commit f48c94e into main May 13, 2026
25 checks passed
@AceHack AceHack deleted the docs/lior-shadow-lesson-3052 branch May 13, 2026 23:08
AceHack added a commit that referenced this pull request May 13, 2026
…rged this session

Records: PRs #3054 + #3055 (Otto-Desktop's shadow-log + archive)
merged. PR #3056 round-1 review surfaced 3 valid Copilot findings
(missing eslint-disable, misleading field name, dead ternary); all
fixed in 7444a05 + threads resolved. Both my PRs back to wait-ci,
threads-clear.

Co-Authored-By: Claude <noreply@anthropic.com>
AceHack added a commit that referenced this pull request May 13, 2026
…p row (#3056)

* feat(bg/audit): duplicate-row-id audit tool + B-0451 substrate-cleanup row

While resolving the B-0444 ID collision (PR #3053), an inline audit
surfaced 12 ADDITIONAL duplicate-ID groups across the backlog
directory. Silently-overwriting substrate state is high-severity
hygiene risk: a consumer of `id: B-0409` gets one of THREE files
depending on load order; every other substrate consumer's implicit
primary-key guarantee is broken.

Changes:

- `tools/bg/audit-duplicate-row-ids.ts` — new audit tool: walks
  `docs/backlog/**/*.md` via `git ls-files`, extracts each frontmatter
  `id:` value, reports any ID appearing in >1 file. Exit code 0 = clean;
  1 = duplicates found.
- `tools/bg/audit-duplicate-row-ids.test.ts` — 14 tests covering id
  extraction, group sorting, real-world patterns (clean substrate,
  pair collision, triple collision, missing-id row skip, sub-row IDs,
  unreadable-file resilience).
- `docs/backlog/P1/B-0451-duplicate-row-id-substrate-cleanup-2026-05-13.md`
  — tracks the cleanup work: lists all 12 collisions, classifies them
  into two patterns (cross-priority namespace bleed + within-priority
  concurrent decomposition), defines the per-collision resolution rule
  (keep the row with external references; renumber the other), and
  outlines CI-wiring as future work.

Empirical findings:
- 559 rows scanned
- 12 collision groups (1 three-way: B-0409; 11 pairs)
- Most pairs are P1-vs-P2 cross-priority bleed (Otto-Desktop vs
  parallel agents filing in overlapping ranges)
- The B-0090.x sub-rows show a within-priority decomposition race
  (Riven's atomic-children sweep vs earlier B-0090 decomposition,
  both 2026-05-10/11)

Tests: 14/14 pass on `tools/bg/audit-duplicate-row-ids.test.ts`.

Co-Authored-By: Claude <noreply@anthropic.com>

* shard(tick): 2317Z — duplicate-ID audit tool surfaces 12 collisions; competing PR closed

Records: PR #3053 (B-0444 collision) merged; PR #3051 (Codex
provenance) merged; competing PR #3052 closed with substrate-honest
comment; new audit tool finds 12 additional ID collisions on main
(1 three-way: B-0409). PR #3056 ships the tool + B-0451 row tracking
the per-collision cleanup work.

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(bg/audit): address Copilot round-1 review on PR #3056

Three Copilot findings resolved:

1. Missing `// eslint-disable-next-line sonarjs/no-os-command-from-path`
   on the `spawnSync("git", ...)` invocation. Added the suffixed-
   rationale comment matching the form used in
   `tools/bg/backlog-ready-notifier.ts`.

2. `rowsScanned` was misleadingly named — it counted only rows with
   an extractable `id:` field, not total files inspected. The tests
   already asserted the smaller count; the CLI's "X rows scanned"
   message therefore under-reported. Renamed to `rowsWithId` and
   updated docstring + CLI message accordingly: "X rows with id
   field, no duplicate IDs".

3. Dead `idToFiles.size > 0` ternary — `reduce` on an empty iterable
   already returns 0. Simplified to a plain spread+reduce.

Tests: 14/14 pass; the `rowsWithId` rename mechanically updates 3
test assertions. Audit tool still reports 12 collision groups on
origin/main (no behavior change beyond the cleaner output).

Co-Authored-By: Claude <noreply@anthropic.com>

* shard(tick): 2339Z — Copilot round-1 on PR #3056 addressed; 13 PRs merged this session

Records: PRs #3054 + #3055 (Otto-Desktop's shadow-log + archive)
merged. PR #3056 round-1 review surfaced 3 valid Copilot findings
(missing eslint-disable, misleading field name, dead ternary); all
fixed in 7444a05 + threads resolved. Both my PRs back to wait-ci,
threads-clear.

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(bg/audit): surface readFileSync errors instead of silently skipping (Codex P2)

Codex P2 on PR #3056: `auditRowFiles` previously caught `readFileSync`
failures with a bare `continue;` — silently swallowing the error and
moving on. That created a false-negative path: if a backlog file was
unreadable (permission, IO error, race with concurrent fs ops), any
duplicate ID inside it never got checked, and the CLI could report
"no duplicate IDs" with the failure hidden.

Fix:

- New `ReadError = { file, reason }` type
- `AuditResult.readErrors: ReadError[]` accumulates per-file failures
  (preserves the original "continue scanning" behavior — see ALL
  problems, not abort on first)
- CLI surfaces read errors with a distinct heading + exits non-zero
  when ANY read error OR duplicate is present
- Success message only prints when both counts are 0

Tests updated (15/15 pass):

- Renamed "unreadable files are skipped without crashing" →
  "unreadable files surface as readErrors (Codex P2: don't silently
  skip)" + assertions on the readErrors[] shape
- Added "readErrors is empty when all files readable" to pin the
  zero-state contract

Co-Authored-By: Claude <noreply@anthropic.com>

* shard(tick): 2358Z — Codex P2 round-2 on PR #3056 (surface read errors); rate-limit-failed CI triaged

Records: PR #3056 CI failures triaged as GitHub-API-rate-limit exhaust
during SARIF upload (not real bugs). Codex P2 round-2 finding addressed:
`auditRowFiles` now accumulates `readErrors[]` and CLI fails non-zero
on any read error or duplicate. 15/15 tests pass. Thread resolved.

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude <noreply@anthropic.com>
AceHack added a commit that referenced this pull request May 14, 2026
…ference card (#3153)

* docs(rules): add ID allocation discipline section to otto-channels reference card

Yesterday's B-0449 collision empirically validated a gap in the channels
reference rule: agents pick monotonically-increasing IDs (B-NNNN backlog
row numbers) by checking on-disk state but not in-flight PRs, race-condition
manifests when peer Otto is filing concurrently.

Adds new section "ID allocation discipline (multi-surface)" after Lane
discipline, requiring BOTH:

1. On-disk check (find docs/backlog → grep B-[0-9]+ → tail)
2. In-flight check (gh pr list --search "B-NNNN")

Empirical anchor: 2026-05-13 collision where Otto on Desktop picked B-0449
for PR #3052, but Otto on CLI had B-0449 in flight via PR #3046 (bg-services
slice 5). Resolved by PR #3053 renumber to B-0450 + PR #3054 shadow lesson log.

Also adds:
- Item 7 to Operational discipline list pointing at new section
- Composes-with entries for PR #3053, PR #3054, refresh-before-decide.md
- /tmp/zeta-otto-id-alloc as example task-specific worktree path

Re-authoring of work lost in yesterday's session crash (was on /tmp/zeta-otto-desktop
which got cleaned up). The substrate gap is still real; the empirical anchor still
holds; the rule update is still valuable.

Co-Authored-By: Claude <noreply@anthropic.com>

* fix(rules): use portable find (drop GNU-only -printf flag) — Codex P2

The on-disk check in the ID allocation discipline section used 'find -printf'
which is GNU-only and fails on BSD find (macOS default). Replaced with the
portable pattern: 'find ... -type f | grep -oE "B-[0-9]+"' — the B-NNNN
pattern only appears once per filename in practice, so extracting from full
paths is equivalent.

Verified portable: ran the new command, output matches expected (top 3 row IDs).

Co-Authored-By: Claude <noreply@anthropic.com>

---------

Co-authored-by: Claude <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants