60 — SB-10 Candidate-State Rehearsal Results (R-4, GCOS keystone)

Doc: 60. Role: Live BEGIN..ROLLBACK rehearsal result for SB-10 (candidate-state store; doc 40 — the convergence keystone). Executes doc 48 Prompt 3. Status: REHEARSAL RESULT — PASS. Author-mode BEGIN..ROLLBACK, ZERO COMMIT, no checked-forever boolean, entry==exit proven. Build NO-GO. Date: 2026-06-01. Entry baseline: doc 57. Depends on SB-12 (snapshot/ruleset) + SB-13 (cursors).

---

60.1 What was rehearsed

• CREATE TABLE governance_candidate_state — GROUP grain (modeled on derived_objects_registry): (group_key, ruleset_version) PK, source_snapshot_ref bigint, candidate_verdict, input_quality_state (doc 33 10 states), recompute_status, dirty/dirty_reason/dirtied_at, scan_time, stale_after, evidence_fingerprint, last_run_id, audit_ref. Deliberately NO is_governed/checked boolean (RR-4 / doc 40 checked-forever anti-pattern).
• CREATE TABLE governance_candidate_object — optional, materialized only for the 4 criteria (indep_authoritative|open_finding|exception|high_risk_write); (candidate_key, ruleset_version) PK.
• CREATE TABLE candidate_scan_run — run ledger.
• Demo rows inserted (group/object/run) then rolled back.

60.2 Transcript (verbatim)

===R4_SB10_PRE===  s=(null) o=(null) r=(null)
BEGIN SET SET SET  CREATE TABLE x3
===R4_INTX_TABLES_VISIBLE===  governance_candidate_state | governance_candidate_object | candidate_scan_run
===R4_PROVE_NO_CHECKED_FOREVER_BOOLEAN_EXPECT_0_ROWS===  (0 rows)
===R4_CANDIDATE_KEY_COALESCE_SAMPLE===
 id |   candidate_key
----+-------------------
  1 | dot_tools:DOT-004
  2 | dot_tools:DOT-005
  3 | dot_tools:DOT-006
===R4_GROUP_GRAIN_DELTA0_MANY_CHILDREN_PER_GROUP===
  collection_name   | children
--------------------+----------
 registry_changelog |    20043
 entity_labels      |     6944
 universal_edges    |     2375
INSERT 0 1 x3
===R4_INTX_ROWS===  cs=1 | co=1 | cr=1
===R4_VERDICT_DECAYING_TRIPLE_clean_now===
 group_key=grp:rehearsal:demo | ruleset_version=gov-rs-rehearsal01 | scan_time=2026-06-01 11:14:27+00 | stale_after=2026-06-08 11:14:27+00 | clean_now=t
ROLLBACK
===R4_SB10_POST===  s=(null) o=(null) r=(null)

Separate-session query_pg after ROLLBACK: all 3 candidate tables NULL.

60.3 The keystone proofs

1. No checked-forever boolean (RR-4): grep of information_schema.columns for {is_governed, checked, is_checked, governed, checked_forever} on governance_candidate_state returned 0 rows. The verdict is the decaying triple (source_snapshot, ruleset_version, scan_time); "clean" is computed as recompute_status='ok' AND NOT dirty AND now() < stale_after (rehearsal showed clean_now=t). dirty is a transient invalidation flag, not a permanent "checked" memory.
2. canonical_address NULL-correction (doc 45 C-2): candidate_key = COALESCE(canonical_address, collection_name||':'||entity_code) resolved to dot_tools:DOT-004… (since canonical is NULL in all rows) — confirms the binding key form, not canonical_address.
3. Group-grain Δrows=0: a single group (e.g. registry_changelog) covers 20,043 children; entity_labels 6,944; universal_edges 2,375. The store holds one group row per (group_key, ruleset_version), never one row per child ⇒ Δtotal rows = 0 at 10^6 scale. Object table materializes only for the 4 criteria.
4. FK-semantics to SB-12: ruleset_version + source_snapshot_ref columns present (hard FK proven in the combined rehearsal, doc 62).

60.4 Entry==Exit

Object	Pre	In-tx	Post (sep. session)	Equal?
governance_candidate_state	NULL	present (1 row)	NULL	✅
governance_candidate_object	NULL	present (1 row)	NULL	✅
candidate_scan_run	NULL	present (1 row)	NULL	✅
birth_registry	unchanged	read-only	unchanged	✅

60.5 Verdict

R-4 SB-10 = PASS. The keystone's three anti-anti-patterns are proven live: no checked-forever boolean, group-grain (Δrows=0), correct candidate-key under the NULL-canonical correction. Reversible/greenfield. Build NO-GO.