chore: disable memory-lint Phase B.5 (trial rolled back)#128
Merged
Conversation
The 2026-05-17→05-28 trial (ADR-069) found Phase B.5 cross-file contradiction lint did not earn its keep: - 0 contradictions detected across 8 nightly runs (~1500 candidate evaluations) - Structurally MISSED the one real cross-type contradiction that arose: the candidate filter uses "same type" as a match signal, so a user-file vs project-file contradiction can't reach the >=2-signal threshold. Cross-type drift is exactly the most dangerous class. - The pre-existing narrative "Noted for Review" digest (predates this trial) caught that case instead. Flip LINT_PHASE_B5_ENABLED to false. Phase C still applies confidence/ revisit_if frontmatter (forward-compatible, harmless). Narrative digest retained — it's the part that works. Retrospective: reference/tasks/workspace-txyu/retrospective.md (private). ADR-069 -> SUPERSEDED.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Roll back the memory-lint Phase B.5 trial introduced in #124. Single-line flag flip:
LINT_PHASE_B5_ENABLED=true→false.Trial result (ADR-069, 2026-05-17 → 2026-05-28, ended day 11 of planned 30)
Decisive finding: a real cross-type contradiction arose (a
user-type file claimed an activity was active while the relatedproject-type file was marked SETTLED). Phase B.5 missed it — its candidate filter uses "same type" as a match signal, so cross-type pairs can't reach the ≥2-signal threshold. Cross-type drift is the most dangerous class, and the filter is structurally blind to it. The pre-existing narrative "Noted for Review" digest (predates this trial) caught the case instead.Conclusion: the algorithmic cross-file lint found nothing in ~1500 evaluations and missed the one real case, while burning LLM judgment calls on up to 36% of all file pairs nightly. The free, pre-existing narrative digest does the actual work. Rolled back.
What changes
LINT_PHASE_B5_ENABLEDflag →false(skill skips Phase B.5; Phases 0/A/B/C/D run as before)confidence/revisit_iffrontmatter still applied by Phase C (forward-compatible, harmless, kept)Test plan
grep -q 'LINT_PHASE_B5_ENABLED=false' .claude/skills/memory-consolidation/SKILL.mdmemory/lint-stats.jsonlappend, no Pending Review writesNotes
The trial harness (feature flag + stats file + reminder + pre-defined rollback criteria) worked exactly as intended — reached a confident rollback in 11 days with zero damage. Reusable pattern for future experiments.
🤖 Generated with Claude Code