sql-representative-and-interface-boundary

Date: 2026-09-12. Status: implemented.

Decision. For the SPD SQL setup and the cross-table comparison, select the newest life per full key across both user types, then prefer live records, whole/A, and newest introduction per ordinary postcode. Ties and missing A receive no geography. Retain last-known deleted records, including named small-user link targets, with current flags. Missing dates and years without an edition have distinct statuses. Accept ordinary input postcodes only. SSPL already selects latest whole records upstream. Python keeps its separate current/as-of, own-record geography policy; no Python/SQL parity is claimed.

Reason. This ordering and complete deleted-record retention are explicit project policies, not a claim to reproduce PHS's unpublished selection details. Do not silently change existing Python consumers or revive an older small-user record when the newest life is an unlinked large user. New SIMD editions and year mappings require separate review.