ops(0-0-0): batch 2 — 8 files SAFE (81 lines) + 1 NEEDS_HUMAN_DECISION (12 lines)

docs/0-0-0-readiness/CLASSIFICATION.md

@@ -98,6 +98,24 @@ Per `docs/active-trajectory.md` strict bucket taxonomy: classification requires
 
 **Batch 1 result: 2 of 2 files SAFE_TO_RESET_LFG_SUPERSEDES.** No forward-sync needed.
 
+### Batch 2 (2026-04-29T12:05Z)
+
+Per `docs/active-trajectory.md` strict bucket taxonomy: classification requires named AceHack content + named LFG equivalent + named reason. The 8 SAFE entries below each have LFG-only commits that are deliberate improvements (perf fix, doctrine update, role-vs-name compliance, technical-accuracy fix, or strict-superset extension); the AceHack-only `+` lines are older shapes superseded by those LFG decisions. The 1 NEEDS_HUMAN_DECISION entry is a content-divergence class that hard-reset alone cannot resolve.
+
+| File | Diff +/- | Status | Evidence |
+|---|---|---|---|
+| `.github/codeql/codeql-config.yml` | +6/-13 | **SAFE_TO_RESET_LFG_SUPERSEDES** | LFG has 3 commits AceHack lacks: `bb0227e` (java-kotlin matrix + scan AlloyRunner.java; removes `tools/alloy/**` from paths-ignore with detailed rationale), `817e59c` (B-0073 obj/bin refinement), `1e09fef` (xunit→xUnit Copilot review fix on PR #701). AceHack +6 lines = older shape that re-included `tools/alloy/**` in paths-ignore — directly contradicts LFG's deliberate paths-ignore decision. |
+| `.github/workflows/memory-index-duplicate-lint.yml` | +8/-5 | **SAFE_TO_RESET_LFG_SUPERSEDES** | Comment-only diff. AceHack `+` lines add persona-name attribution (citing an external AI reviewer + a dated technical-review reference + a PR number) to a CI workflow file. Per the role-vs-name rule (`docs/AGENT-BEST-PRACTICES.md` closed-list), workflow files are current-state surfaces — persona-names belong on history surfaces only. LFG version uses role-ref-style comment ("an index with duplicate entries is a discoverability defect") which is the rule-compliant form. |
+| `tools/hygiene/audit-memory-index-duplicates.sh` | +8/-7 | **SAFE_TO_RESET_LFG_SUPERSEDES** | Same pattern as the workflow file above — comment-only diff. AceHack `+` adds persona-name attribution (citing an external AI reviewer + a dated technical-review reference) to a tool script. Tool scripts are current-state surfaces; LFG's role-ref version is rule-compliant. |
+| `src/Core/Shard.fs` | +9/-12 | **SAFE_TO_RESET_LFG_SUPERSEDES** | LFG has `aa5395b sync: AceHack→LFG merge-needed batch (#649)` + `9373755 fix(audit): comprehensive HashCode.Combine sweep + SplitMix64 refactor (Otto-281 follow-up)`. LFG version uses `EqualityComparer<'K>.Default.GetHashCode(key)` (non-boxing, null-safe) per Copilot P1 review on LFG #649 (boxing on every call for value-type 'K → GC regression on hot paths). AceHack version still uses `match box key with \| null -> 0 \| boxed -> boxed.GetHashCode()` (explicit boxing every call). LFG is the perf-fixed form. |
+| `docs/AUTONOMOUS-LOOP.md` | +9/-18 | **SAFE_TO_RESET_LFG_SUPERSEDES** | LFG has 5 commits AceHack lacks, including `8aa4adc feat(tick-history): land Option B per-tick shard-file transport (task #276 — maintainer-delegated call) (#724)` (landed 2026-04-29T02:04:38Z). LFG describes the **current Option B per-tick shard mechanism** at `docs/hygiene-history/ticks/YYYY/MM/DD/HHMMZ.md`. AceHack still describes the legacy single-table append. LFG is the current doctrine. |
+| `tools/setup/macos.sh` | +11/-16 | **SAFE_TO_RESET_LFG_SUPERSEDES** | Two changes: (1) shellcheck source-directive comment — LFG has detailed rationale (multi-line explanation of why SC1091 fires + why source= directive resolves it); AceHack has terse 1-line version. (2) Setup script comments — LFG uses role-refs ("the maintainer's standing 'just install everything' framing"); AceHack adds persona-name attribution (citing the human maintainer + a round number) and persona-name attribution (citing an external AI reviewer + a PR number). Both AceHack changes are role-vs-name violations on a current-state surface. LFG version is more detailed AND rule-compliant. |
+| `tools/hygiene/fix-markdown-md032-md026.py` | +16/-157 | **SAFE_TO_RESET_LFG_SUPERSEDES** | LFG has `f57d683 fix(hygiene): MD032 auto-fixer skips YAML frontmatter (don't break composes_with: lists) (#703)` + `515db64 feat(hygiene): fix-markdown-md032-md026.py — recurring pattern extracted to substrate primitive (Otto-346 in action) (#542)`. LFG file is 376 lines vs AceHack 235 lines: LFG = strict superset (fence detection at line 60 on both sides + YAML frontmatter handling that AceHack lacks). AceHack +16 lines are docstring/comment shifts on the fence-detection block (logic identical between sides; LFG just has more code before/after). |
+| `tools/setup/common/curl-fetch.sh` | +14/-44 | **SAFE_TO_RESET_LFG_SUPERSEDES** | LFG has 2 commits AceHack lacks (`5298114` infra clean-additive batch + `dfb49e5` 63-file forward-port). Two changes in the comment block: (1) attribution — LFG: "Reviewers confirmed:" (role-ref); AceHack: persona-name attribution citing an external AI reviewer + a PR number (role-vs-name rule violation on a current-state surface). (2) retry-math accuracy — LFG: `--retry 5 — up to 5 retries (6 total attempts including the initial try, per curl(1))` (correct per curl manual); AceHack: `--retry 5 — five attempts total` (technically inaccurate). LFG is more accurate AND rule-compliant. |
+| `docs/hygiene-history/loop-tick-history.md` | +12/-12 | **NEEDS_HUMAN_DECISION** → maintainer chose option (c) | Mutual divergence with two distinct cases. **Truly unique to AceHack** (9 timestamps): 2026-04-28T04:08/04:18/04:33/05:01/05:23/05:44/05:50/07:15/08:50. **Truly unique to LFG** (9 timestamps): 2026-04-27T23:58, 2026-04-28T14:43/17:47/19:41/19:50, 2026-04-29T00:34/00:53/00:58/01:00. **Present on both forks with content drift in trailing portion** (1 timestamp, 2026-04-21T17:28): the row exists on both sides but differs enough to appear as `+/-` pair in `git diff`. All AceHack-only rows are pre-Option-B (Option B shard-mode landed 2026-04-29T02:04:38Z on LFG via PR #724) and were correctly appended to the legacy single-table file at the time. **Maintainer call (2026-04-29)**: option (c) — migrate AceHack-only rows to per-tick shard files under `docs/hygiene-history/ticks/2026/04/{21,28}/` first, then hard-reset. Rationale: preserves evidence in the modern Option B format, avoids accepting loss, makes loop-tick-history.md content-identical between forks, converts the only NEEDS_HUMAN_DECISION item into a preservation-forward cleanup. **Reclassifies to SAFE_TO_RESET_LFG_SUPERSEDES once option-(c) migration PR lands.** Pre-existing 9 LFG-only rows are out of scope for this PR (separate follow-up to migrate THOSE to shards lives on the LFG side). |
+
+**Batch 2 result: 8 of 9 files SAFE_TO_RESET_LFG_SUPERSEDES (81 lines), 1 file NEEDS_HUMAN_DECISION (12 lines).**
+
 ### Calibration batch reclassification (older "ALREADY-COVERED" label)
 
 The 5 calibration-batch entries above use the prior taxonomy's "ALREADY-COVERED" label, which under the strict bucket rules (named AceHack content + named LFG equivalent + named reason — see `docs/active-trajectory.md` Hard-reset-safety classification) corresponds to **SAFE_TO_RESET_LFG_SUPERSEDES** (4 entries with semantic supersession evidence) or **ALREADY_RESOLVED** (where AceHack-only content is zero). Re-tabulated for the strict-bucket ledger:

docs/active-trajectory.md

@@ -136,35 +136,39 @@ For the text-file ledger, the conceptual computation is straightforward: `git di
 
 Verified 2026-04-29T10:43Z: the 5 binary-classified files in the current diff have status `D` (LFG-only), so `binary_acehack_only_files = 0` and `binary_modified_or_renamed_files = 0` in this specific round.
 
-Current ledger (computed 2026-04-29T11:32Z):
+Current ledger (last updated 2026-04-29T12:13Z, post-Batch-2):
 
 ```text
 potential_loss_lines  = 273    all AceHack-only +lines (would be erased on hard-reset)
-classified_safe_lines = 134    semantic evidence in BUCKET 2 (SAFE_TO_RESET_LFG_SUPERSEDES)
-unsafe_lines          = 0      no NEEDS_FORWARD_SYNC or NEEDS_HUMAN_DECISION
-unclassified_lines    = 139    HEURISTIC_LFG_DOMINATES — pending per-file semantic inspection
+classified_safe_lines = 215    semantic evidence in BUCKET 2 (SAFE_TO_RESET_LFG_SUPERSEDES)
+unsafe_lines          = 12     1 NEEDS_HUMAN_DECISION (loop-tick-history.md mutual divergence)
+unclassified_lines    = 46     HEURISTIC_LFG_DOMINATES — pending per-file semantic inspection
 ```
 
-Composition of `classified_safe_lines = 134`:
+`potential_loss_lines = 273` was computed 2026-04-29T10:25Z via `git diff --numstat refs/remotes/origin/main..refs/remotes/acehack/main` and remains canonical at the time of the most recent batch-PR open (the AceHack and LFG main tips have not advanced relative to each other in a way that touched the divergent files — see "Drift cadence observation" earlier; LFG advanced via #837 but #837 only touched docs in `docs/0-0-0-readiness/` which are not in the AceHack-vs-LFG file set). Re-compute on next batch open if either tip moves.
+
+Arithmetic sanity check: `273 = 215 + 12 + 46` ✓ (per the multi-AI review discipline — verify mechanically, do not trust the math because it "looks plausible").
+
+Composition of `classified_safe_lines = 215`:
 
 - 9 infra files (97 lines): see "9 infra files" table above. SAFE_TO_RESET_LFG_SUPERSEDES with named per-file evidence.
 - 5 calibration-batch files (28 lines, 2026-04-28): MEMORY.md (11) + codeql_umbrella (12) + doc_class_mirror_beacon (1) + CURRENT-aaron (2) + CURRENT-amara (2). Originally labeled "ALREADY-COVERED" in older taxonomy; under strict bucket each has named evidence in `docs/0-0-0-readiness/CLASSIFICATION.md` → SAFE_TO_RESET_LFG_SUPERSEDES.
 - Batch 1 (9 lines, 2026-04-29T11:32Z): SECURITY.md (4) + validate-agencysignature-pr-body.sh (5). See `docs/0-0-0-readiness/CLASSIFICATION.md` Batch 1 table for named evidence per file.
+- Batch 2 (81 lines, 2026-04-29T12:05Z): codeql-config.yml (6) + memory-index-duplicate-lint.yml (8) + audit-memory-index-duplicates.sh (8) + Shard.fs (9) + AUTONOMOUS-LOOP.md (9) + macos.sh (11) + fix-markdown-md032-md026.py (16) + curl-fetch.sh (14). See `docs/0-0-0-readiness/CLASSIFICATION.md` Batch 2 table for named evidence per file. Common pattern: LFG version is either rule-compliant (role-refs vs persona-name violations on current-state surfaces), more accurate (correct retry-math on curl-fetch.sh), the perf-fixed form (Shard.fs non-boxing comparer), the current doctrine (AUTONOMOUS-LOOP.md Option B shard-mode), or strict superset (fix-markdown-md032-md026.py YAML frontmatter handling).
+
+Composition of `unsafe_lines = 12` (1 file, NEEDS_HUMAN_DECISION):
+
+```text
+12  docs/hygiene-history/loop-tick-history.md  (mutual divergence — 9-10 unique tick rows on each fork)
+```
+
+Per Batch 2 evidence: each fork has unique tick rows for periods when the loop ran independently on that fork. All AceHack-only rows are pre-Option-B (Option B shard-mode landed 2026-04-29T02:04:38Z on LFG via PR #724). Maintainer call is needed on whether to (a) accept the loss on hard-reset, (b) forward-sync to LFG first, or (c) migrate to per-tick shard files first then hard-reset. Recommended path: (c) — preserves evidence in modern format AND makes the file content-identical between forks.
 
-Composition of `unclassified_lines = 139` (11 files):
+Composition of `unclassified_lines = 46` (2 files):
 
 ```text
 38  .github/workflows/budget-snapshot-cadence.yml
-16  tools/hygiene/fix-markdown-md032-md026.py
-14  tools/setup/common/curl-fetch.sh
-12  docs/hygiene-history/loop-tick-history.md
-11  tools/setup/macos.sh
- 9  src/Core/Shard.fs
- 9  docs/AUTONOMOUS-LOOP.md
- 8  tools/hygiene/audit-memory-index-duplicates.sh
  8  memory/project_laptop_only_source_integration_scratch_sqlsharp_features_or_designs_high_priority_2026_04_27.md
- 8  .github/workflows/memory-index-duplicate-lint.yml
- 6  .github/codeql/codeql-config.yml
 ```
 
 ### Hard-reset signoff gate (strict)

docs/hygiene-history/ticks/2026/04/29/1207Z.md

@@ -0,0 +1 @@
+| 2026-04-29T12:07:27Z | claude-opus-4-7 | 43c91583 | 0/0/0 batch-2 classification tick post computer-restart. Resumed from compacted-conversation state. Recovery sequence: (a) /tmp/zeta-clean-2026-04-29/lfg fresh clone wiped by macOS reboot — recreated via `git clone` + `git fetch acehack`, frozen with `gc.auto=0` + `reflogExpire=never`; (b) local repo synced to LFG main = `6f8df33` (#837 squash-merge); (c) merged feature branch deleted; (d) per-file batch-2 evidence (gathered pre-restart) re-verified against fresh clone — 9 of 9 files match the pre-restart classifications. Batch 2 PR (#838) opened with: 8 SAFE_TO_RESET_LFG_SUPERSEDES (codeql-config.yml + memory-index-duplicate-lint.yml + audit-memory-index-duplicates.sh + Shard.fs + AUTONOMOUS-LOOP.md + macos.sh + fix-markdown-md032-md026.py + curl-fetch.sh = 81 lines) + 1 NEEDS_HUMAN_DECISION (loop-tick-history.md = 12 lines, mutual divergence in pre-Option-B tick rows, recommended path: migrate AceHack-only rows to per-tick shard files first then hard-reset). Ledger transition: 134 → 215 classified_safe; 0 → 12 unsafe; 139 → 46 unclassified. Arithmetic verified: `273 = 215 + 12 + 46` ✓ (per Amara 2026-04-29 mechanical-verification discipline). Common batch-2 evidence pattern: AceHack-only `+` lines are either persona-name attribution on current-state CI/tool/code surfaces (role-vs-name rule violation under the closed-list per `docs/AGENT-BEST-PRACTICES.md`), older shapes superseded by deliberate LFG decisions (paths-ignore on codeql-config, perf fix on Shard.fs comparer, Option B shard-mode doctrine on AUTONOMOUS-LOOP.md), or technical-accuracy regressions (curl-fetch retry-math). Lane discipline preserved per Amara 2026-04-29 review: classify files, reduce unclassified_lines, record named evidence. No LOST recovery, no tick-3 promotion, no gate-runner script. Cron `43c91583` alive. | [PR #838](https://github.com/Lucent-Financial-Group/Zeta/pull/838) | **Resume-protocol working as designed**: the load-state file (`docs/active-trajectory.md`) prevented re-derivation of state after computer-restart. Read-once at session-start gave: ledger numbers, fresh-clone path + recovery procedure, gate conditions, classification rule, recommended resume command. The resume cost was minimal (~2 min to re-establish state vs the ~30+ min audit-spiral that would have happened without the load-state file). **Observation**: the `/tmp` wipe is a known macOS behavior, but the fresh-clone-recreation procedure was self-evident from active-trajectory.md (clone → fetch acehack → fsck → freeze) — the procedure WAS in the doc, not just in agent memory. This validates the design choice to put procedure-as-text in the load-state file. |