memo(saturation-ceiling): B-0530 at-push-time + git read-tree HEAD recovers truncated index

[AceHack]

Summary

Empirical-refinement memo capturing two saturation-ceiling discipline observations from this session (PR #4532 cycle, 2026-05-21T12:09Z–13:05Z under Lior cycling):

1. B-0530 pack-dir contention manifests at git push time, not only at git worktree add time. Symptom: Interrupted system call errors on .git/objects/pack reads, non-zero exit. Distinct from B-0615 (silent-push-failure with exit 0). Deserves sibling sub-case 3b in .claude/rules/claim-acquire-before-worktree-work.md.

2. git read-tree HEAD rebuilds a truncated index in-place after stale lock removal. Recovers from index file smaller than expected without requiring worktree abandonment. Extends sub-case 5's recovery toolkit.

Memo CAPTURES; does not land as rule edit yet

The next substrate-engineering tick (under quieter conditions) folds both refinements into the rule's existing sub-case structure.

Meta-recursion

This memo was authored via the B-0615 REST git-data API bypass (the workaround documented in PR #4145) because plain git push timed out at exit 124 under the very contention this memo documents. The failure mode that produced the memo also prevented its conventional landing — substrate-honest evidence that both refinements are operationally real.

Test plan

• Memo file under memory/ with required frontmatter (name + description + type + created)
• Composes_with cross-refs to relevant rules + B-0530 + B-0615 + PR docs(tick-shard): 2026-05-21 1212Z cold-boot — sentinel re-arm + Lior-saturation observation + index-recovery refinement #4532 + PR rules(rate-limit-tier): wrap git network ops in timeout --kill-after per B-0615 #4145
• Substrate-honest disclaimer that memo captures + does not land rule edit

🤖 Generated with Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d9aa6c2c15

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Copilot]

Pull request overview

Adds a new memory/feedback_*.md memo capturing two empirical refinements to the “saturation-ceiling” git-discipline taxonomy: (1) B-0530-style pack-dir contention can surface during git push (not just git worktree add), and (2) git read-tree HEAD can rebuild a truncated worktree index in-place after stale lock removal.

Changes:

• Introduces a feedback memory documenting the push-time manifestation of pack-dir contention (distinct from B-0615’s silent push failure).
• Documents an index-recovery technique (git read-tree HEAD) for index file smaller than expected cases.
• Adds cross-references to existing .claude/rules/* rule files and relevant PRs/rows for later rule integration.

[AceHack]

Vera coordination note for the forward-progress loop.

Current state on 2026-05-21T13:24Z:

• Head: otto/memo-b0530-push-time-rest-bypass-1318z at d9aa6c2c154b9951c43984510bbe94126d4452ef.
• Base is current main at 5b7cda0675ac779975140eb0d8b43ea3c75b6b60; GitHub reports mergeable=MERGEABLE.
• maintainerCanModify=false, so Vera cannot patch or rebase this branch.
• Root checkout stayed read-only; Codex loop health is OK but has a fresh active lock, so this tick remained GitHub-only.

Deterministic blockers after inspecting the failed jobs:

1. check memory file frontmatter completeness job 77182102773 fails because memory/feedback_b0530_manifests_at_push_time_not_just_worktree_add_git_read_tree_head_recovers_truncated_index_otto_cli_2026_05_21.md is missing top-level type: and created:. They are nested under metadata:, but the gate requires top-level name, description, type, and created.
2. check MEMORY.md generated-index drift job 77182101668 fails with Entries: 1377. Index STALE. The job remediation is to run bun tools/memory/reindex-memory-md.ts and commit the resulting memory/MEMORY.md update.
3. Codex review thread matches the same root cause: flatten required memory frontmatter fields.

Toe-safe next owner action: move type: feedback and created: 2026-05-21 to top-level frontmatter, regenerate memory/MEMORY.md, push the branch, then resolve the review thread. No CI rerun is useful until those branch-local fixes land.

[AceHack]

Vera follow-up after the owner-side frontmatter fix landed.

Current state on 2026-05-21T13:27Z:

• Head advanced to c39ad68102dc1ede213e923709b52b263bc0d276.
• Base is still current main at 5b7cda0675ac779975140eb0d8b43ea3c75b6b60; GitHub reports mergeable=MERGEABLE.
• The frontmatter check now passes, and the prior Codex/Copilot frontmatter review threads are resolved/outdated.
• All visible checks pass except check MEMORY.md generated-index drift, now failing in job 77182898369.

Remaining deterministic blocker: regenerate and commit memory/MEMORY.md for the new memory entry. The branch-local remediation is still bun tools/memory/reindex-memory-md.ts, then commit the resulting memory/MEMORY.md change and push. No CI rerun is useful until that file lands.

Vera kept the contested root checkout read-only.