hygiene(#268+): pr-preservation drain-log for #433 (#226 first cascade)

[AceHack]

Summary

Otto-268 follow-on: drain-log for PR #433 — 3-finding first cascade to #226 (memory reconciliation algorithm design v0). Captures three orthogonal algorithm-correctness improvements.

#226 → #433 → #434 walks the algorithm spec through three iterative cascade waves; this 433-drain-log.md covers the first wave.

Per Otto-250 (PR review comments + responses + resolutions are high-quality training signals).

Coverage — three orthogonal algorithm-correctness fixes

• chain_head-liveness vs priority-tie-break distinction (P1): two invariants conflated at high level distinguished — Invariant 2 (chain_head liveness) vs Invariant 6 (priority-tie-break-for-winner fallback).
• Paired-delimiter-vs-raw-character normalization rule (P1): "Strip markdown formatting" reformulated as paired-delimiter unwrap (preserves _internal_var + stray-backtick; unwraps _text_ + **text**).
• MEMORY.md dedup-by-source_path (P1): multiple typed facts from same source file → ONE index row synthesized from highest-priority fact's title + description.

Pattern observations

1. Spec-correctness findings on algorithm designs benefit from per-invariant orthogonal-checks reasoning. Codex catches invariant-conflation reliably.
2. Paired-delimiter-vs-raw-character is a normalization-rule precision class — same shape as craft: fourth module — semiring-basics (recipe-template anchor) #206's K-relations subset-vs-superset precision error.
3. Index-rendering dedup is its own correctness class. Every aggregation rule needs an explicit dedup-key + dedup-strategy.
4. Algorithm-design specs benefit from multiple cascade waves: research: memory reconciliation algorithm design v0 (Amara Determinize 3/5, L-effort design) #226 → drain(#226 follow-up): priority tie-break + strip-paired-delimiters + dedup-by-source_path #433 → drain(#226 follow-up 2): CC log schema + populated-state alignment #434 walked through three iterations, each catching a different class of correctness gap.

Test plan

• All 3 algorithm-correctness findings preserved with severity + outcome class.
• research: memory reconciliation algorithm design v0 (Amara Determinize 3/5, L-effort design) #226 → drain(#226 follow-up): priority tie-break + strip-paired-delimiters + dedup-by-source_path #433 → drain(#226 follow-up 2): CC log schema + populated-state alignment #434 cascade-wave arc documented.
• Drain-log shape matches existing docs/pr-preservation/*-drain-log.md template.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

Adds a PR-preservation drain log for PR #433, capturing the first post-merge cascade of algorithm-spec correctness findings for the memory-reconciliation design work introduced in #226.

Changes:

• Introduces docs/pr-preservation/433-drain-log.md documenting 3 post-merge findings (priority tie-break vs chain_head liveness, paired-delimiter stripping rule, and MEMORY.md dedup-by-source_path).
• Records pattern observations tying these findings to recurring “training-signal” classes in the drain corpus.
• Captures a final-resolution summary pointing to the fixing SHA.

[Copilot]

The log says this is a “full record” of the Otto-250 training signal, but the threads don’t include stable provenance fields (e.g., thread ID and/or file:line) nor the verbatim original review comment + reply that earlier PR-preservation drain logs capture. Consider adding at least (a) the GitHub review thread ID and (b) the parent-file location (path + line/heading), and ideally quote the original comment/reply verbatim, so future readers can trace each finding back to the exact discussion context.