KB-AD6D
05 — Data-Quality Classification (all 37 pivots)
2 min read Revision 1
registries-pivotdata-qualityclassificationcleandirtyambiguous37-pivots2026-06-03
05 — Data-Quality Classification (Objective D)
All 37 live pivots classified vs governed truth (from read-only rehearsal).
Buckets:
- clean (21): PIV-002,009,003,007,008,011,010,014,207,004,005,006,012,013,104,105,015,017,018,019,201 — already governance-aligned.
- dirty noncanonical comp (3): PIV-204(material),205(product),206(building) — all catalog=atom; fix comp via DOT, move drill-level to pivot_kind/drill_level.
- mismatch-vs-governed (7): PIV-001,101,102,106,202,203 (catalog labeled molecule/compound but governed atom) + 021 (checkpoint_support atom vs governed molecule).
- ambiguous needs-human-naming (2): PIV-103,016 (species='species' over entity_species; comp atom vs governed meta).
- missing classification (1): PIV-020 (_uncategorized, inactive).
- structurally-null/matrix (3): MTX-L1-OVERVIEW, MTX-L2-ATOM, MTX-TEST (species null defensible; move matrix out of composition_level into pivot_kind).
- orphan/phantom: 0.
- blocked by L2 gate: ownership of any group only; classification mapping NOT blocked.
The 7 mismatch + 3 dirty are all catalog (or checkpoint_support) where RP composition_level encodes pivot drill-depth not entity composition (Category error #2): clean on species, dirty only on composition semantics.
RP-side fix list (via DOT, Đ26 — no manual INSERT): fix PIV-204/205/206 comp; retire PIV-020 + MTX-TEST; resolve species='species' (PIV-016/103) set comp=meta; decide catalog drill-depth policy; split registry_group into domain_facet + pivot_kind.