SubMiner/docs/subagents/agents/codex-duplicate-kiku-20260221T043006Z-5vkz.md

codex-duplicate-kiku-20260221T043006Z-5vkz

• alias: codex-duplicate-kiku
• mission: Fix Kiku duplicate-card detection/grouping regression for Yomitan duplicate-marked + N+1-highlighted cards
• status: completed
• start_utc: 2026-02-21T04:30:06Z
• last_update_utc: 2026-02-21T10:07:58Z

Intent

• Reproduce bug where clear duplicate cards no longer trigger Kiku duplicate grouping.
• Add failing regression test first (TDD).
• Patch duplicate detection logic with minimal behavior change.

Planned Files

• src/anki-integration/duplicate.ts
• src/anki-integration/duplicate.test.ts (or nearest duplicate-detection tests)
• docs/subagents/INDEX.md
• docs/subagents/collaboration.md
• backlog/tasks/task-94 - Fix-Kiku-duplicate-detection-for-Yomitan-marked-duplicates.md

Assumptions

• Duplicate signal should still come from Anki duplicate search + Yomitan/N+1-derived fields used in note content.
• Regression likely from term/readings normalization/query escaping mismatch.

Outcome

• Root cause: candidate-note exact-check only resolved the originating field name (Expression or Word), so duplicates failed when candidate note used the opposite alias.
• Added regression test first (RED): Expression current note vs Word candidate with same value returned null.
• Implemented minimal fix: candidate resolution now checks both aliases (word and expression) before exact-value compare.
• GREEN: targeted duplicate test passed; related anki-integration test passed.
• User follow-up repro showed remaining miss when duplicate appears only in alias field search results.
• Added second RED test for alias-query fallback.
• Implemented query-stage alias fallback: run findNotes for both alias fields, merge note ids, then exact-verify.
• GREEN after follow-up: duplicate tests + anki-integration test pass.
• User reported still failing after first follow-up.
• Added third RED regression: source note containing both Expression (sentence) and Word (term) only matched duplicates via Word; previous logic missed this by using only one source value.
• Implemented source-candidate expansion: gather both Word and Expression source values, query aliases for each, dedupe queries, then exact-match against normalized set.
• GREEN: duplicate tests (3/3) + anki-integration test pass.
• Image-backed repro indicated possible duplicate outside configured deck scope.
• Added fourth RED regression: deck-scoped query misses, collection-wide query should still detect duplicate.
• Implemented deck fallback query pass (same source/alias combinations without deck filter) when deck-scoped pass yields no candidates.
• GREEN: duplicate tests (4/4) + anki-integration test pass.
• User confirmed fresh build/install still failed with 貴様 repro.
• Added fifth RED regression: field-specific queries return no matches but plain text query returns candidate.
• Implemented plain-text query fallback pass (deck-scoped then global), still gated by exact word/expression value verify.
• GREEN: duplicate tests (5/5) + anki-integration test pass.
• Added runtime debug instrumentation for duplicate detection query/verification path:
  • query string + hit count
  • candidate count after exclude
  • exact-match note id + field
• No behavior change from instrumentation; build + tests still green.
• User requested logging policy update: prefer console output unless explicitly captured, and persistent logs under ~/.config/SubMiner/logs/*.log.
• Updated default launcher/app mpv log path to daily file naming: ~/.config/SubMiner/logs/SubMiner-YYYY-MM-DD.log.
• Typecheck green.
• Found observability gap: app logger wrote only to stdout/stderr while launcher log file only captured wrapper messages.
• Added file sink to src/logger.ts so app logs also append to ~/.config/SubMiner/logs/SubMiner-YYYY-MM-DD.log (or SUBMINER_MPV_LOG when set).
• Verified with typecheck + build.

Files Touched

• src/anki-integration/duplicate.ts
• src/anki-integration/duplicate.test.ts
• backlog/tasks/task-94 - Fix-Kiku-duplicate-detection-for-Yomitan-marked-duplicates.md
• docs/subagents/INDEX.md
• docs/subagents/collaboration.md
• docs/subagents/agents/codex-duplicate-kiku-20260221T043006Z-5vkz.md

Handoff

• No blockers.
• Next step: run broader gate (bun run test:fast) when ready, then commit.