The Kate Combo Database, Pinned

Live counts read from the files on 2026-07-02. The archive that kept getting lost, findable in one click forever.

This page is a map to the box of Kate's saved card meanings, so nobody ever loses the box again. The green row is the real box.

Archive (ranked)	Live count	What it is
1. THE BIG ONE (use this) `/Users/drwu/Projects/_outputs/tarot-voice/kate_combo_meanings.jsonl`	23,370 windows, 2,974 distinct pairs of 3,003	Verbatim combo MEANING windows across 188 Kate transcripts. Filter: drop 'revers', rank teaching_score desc, listing_score asc. This is what 'the Kate combos' means.
2. v2 meaning DB (cleaner, smaller) `/Users/drwu/New Desktop/tarot-voice-v2/meaning_db/kate/combos.jsonl`	16,018 windows	Clean upright subset behind read_spread.py. Use when you want the tightest teaching windows and can live with missing pairs.
3. Local pair dict `/Users/drwu/Projects/_outputs/tarot-voice-v2/kate_combos_local/kate_combos_local.json`	2,718 pairs	Keyed 'Card A + Card B'. Convenience lookup only.
4. coverage_matrix.csv (NOT the database) `/Users/drwu/New Desktop/tarot-voice-v2/coverage_matrix.csv`	8,016 rows	Hit COUNTS only, no meaning text. The source of the misleading 97.7%. Never treat as capability, never read meanings from it.
5. tarot-reader-render curated set `~/Desktop/tarot-reader-render/readers/kate/combos/ (STALE)`	path gone (Desktop offloaded 2026-06-06)	The old kate_says-keyed curated set. If it matters it must be relocated or rebuilt; do not cite it as live.

Hard rules (from the canonical-map memory + your own words):
1. "Kate combos" means archive #1. Go there first, every session.
2. PULL, never rebuild. Combos are looked up and pasted verbatim, never regenerated, never routed through DeepSeek.
3. Row 4 is hit counts, not meanings. Quoting its 97.7% as capability is the documented fraud pattern.
4. Coverage percent = co-occurrence density, not combo skill. Say so when reporting.
5. Fidelity is scored meaning-only, crutch excluded: ~/.claude/skills/_shared/combo_meaning_score.py. Worked example with score: /kate-combo-reading-lookup (84.2% reader-verbatim, PASS).

Sources: reference_kate_combo_archive_canonical_map.md (memory), live filesystem counts this build. Sibling: index.md. Linked from /tarot-combo-index.