fix(salvage): apoptosis transaction + decompile listener-before-reset + audit regression test

[JosephOIbrahim]

Summary

Three correctness fixes salvaged from the now-closed PR #4. PR #4 was closed because master diverged on the banner (ab57275) and tool names (490e3e0) in ways that invalidated its docs commits — but these three code changes are independent of those decisions and remain valuable on master.

Fix	File	Why it matters
Transaction-wrapped apoptosis DELETEs	crates/hippocampus/src/store.rs	The chunked apoptosis loop runs trace DELETE + graph_edges DELETE per chunk. Mid-loop failure could leave traces deleted but their edges still pointing at orphan ids. Wrapping in a single rusqlite::Transaction makes per-chunk DELETEs atomic; commit before VACUUM (which can't run inside an open tx).
on_decompile listener fires before success_count reset	python/harlo/motor/motor_cerebellum.py	Today the listener observes success_count == 0 (post-reset) — the audit trail loses the prior count. Reordering so the listener fires BEFORE the Rule 32 reset means observers capture the pre-reset value while external readers still see 0. Bare except: pass replaced with logger.exception() so listener errors are diagnosable; fire-and-forget control flow preserved.
Regression test for the audit-trail contract	tests/test_motor/test_motor.py	test_decompile_listener_sees_prior_success_count registers a listener that captures pattern.success_count at notification time, sets up a pattern with success_count=42, calls record_failure, and asserts (a) the listener saw 42 (audit) and (b) post-call read sees 0 (Rule 32 literal).

What's intentionally NOT here

• Banner / README / mermaid commits from PR style(branding): Pentagram-aesthetic refresh — banner3 figlet + two-tone-orange README + v6.0-MOTOR docs #4. Master's ab57275 locked the banner to ansi_shadow and 490e3e0 dropped the twin_ prefix; the cosmic banner and twin_* tool names in PR style(branding): Pentagram-aesthetic refresh — banner3 figlet + two-tone-orange README + v6.0-MOTOR docs #4's docs would conflict.
• Anything from PR fix(constitutional): S7 formula + Checkpoint TOCTOU + engine-lock timeout + HotStore commit handling #6 (S7 formula, Checkpoint TOCTOU, _engine_lock timeout, HotStore commit). Those are tracked separately in fix(constitutional): S7 formula + Checkpoint TOCTOU + engine-lock timeout + HotStore commit handling #6.

Test plan

• cargo test -p hippocampus — 42 passed, 0 failed (apoptosis chunk-boundary + transaction behavior preserved)
• pytest tests/test_motor/ — 15 passed, 0 failed (10 baseline + 5 review-driven, including the new prior-count regression)
• Constitutional greps (sleep(, while True, DELETE.*audit, float32, cosine, reasoning_trace in elenchus/verifier.py) — all zero
• No overlap with PR fix(constitutional): S7 formula + Checkpoint TOCTOU + engine-lock timeout + HotStore commit handling #6's diff — independent files, can land in any order

https://claude.ai/code/session_017arHKzx5mTUFiry7JhhRPs

---

Generated by Claude Code

Summary by CodeRabbit

• Bug Fixes

  • Listener error messages are now logged instead of being silently ignored, improving diagnostic visibility.

• Chores

  • Improved reliability and safety for bulk data deletion operations.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 2d0c4399-bcea-4568-aa10-43e6dee84257

📥 Commits

Reviewing files that changed from the base of the PR and between 490e3e0 and b1d320d.

📒 Files selected for processing (3)

• crates/hippocampus/src/store.rs
• python/harlo/motor/motor_cerebellum.py
• tests/test_motor/test_motor.py

---

📝 Walkthrough

Walkthrough

This PR introduces two independent improvements: batching SQL deletions in a single transaction within the storage layer, and refactoring decompile listener notification to fire before success_count reset with proper exception logging.

Changes

Storage Transaction Batching

Layer / File(s)	Summary
Transaction Wrapping crates/hippocampus/src/store.rs	Chunked trace and graph_edges deletions are wrapped in a single unchecked_transaction with per-chunk SQL execution using transaction-scoped calls; VACUUM remains outside the transaction.

Decompile Listener Ordering and Error Handling

Layer / File(s)	Summary
Error Logging Infrastructure python/harlo/motor/motor_cerebellum.py	Module-level logger is introduced and imports are reorganized.
Listener Notification and Reset Ordering python/harlo/motor/motor_cerebellum.py	_decompile method now invokes on_decompile listener before resetting success_count; listener exceptions are logged instead of silently ignored.
Listener Ordering Verification Test tests/test_motor/test_motor.py	New test test_decompile_listener_sees_prior_success_count asserts the listener observes the pre-reset success_count and that reset occurs afterward.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~20 minutes

Possibly related PRs

• JosephOIbrahim/Harlo#3: Implements batching improvements to microglia_apoptosis transaction handling and refines MotorCerebellum decompile listener ordering with enhanced error logging.

Poem

🐰 A rabbit hops through code with care,
Transactions bundled, listeners fair,
Before we reset, we notify with grace,
Errors logged in their rightful place! ✨

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately and specifically describes the three independent correctness fixes included in the PR: transaction-wrapped apoptosis, listener-before-reset ordering, and the regression test.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch claude/pr4-salvage-correctness

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}