[kryptobaseddev]

Summary

T13311. With row uids on by default (#1964, T13305), every insert of a minted row under sync.capture also captured a K (re-key) op: [K,I,U,D] instead of [I,U,D].

Root cause. The per-connection TEMP fill (installRowUidTriggers) sets uid and then birth_fp in two UPDATEs. By the second, OLD.uid is set, and _sync_cap_<t>_k's OLD.birth_fp IS NOT NEW.birth_fp read the birth_fp NULL -> value fill as a re-key.

Decision (journal spec v11.4, §2.3, recorded via cleo docs update). No stray K at all, rather than a K the sealer must recognise and drop. The K trigger counts a birth_fp change only when the old value was set: OLD.uid IS NOT NULL AND (OLD.uid IS NOT NEW.uid OR (OLD.birth_fp IS NOT NULL AND OLD.birth_fp IS NOT NEW.birth_fp)). _f alone patches the filled identity into the I, in both trigger orders. This keeps the S2 rule "identity-fill updates capture nothing". A value -> value or value -> NULL birth_fp change is still a K. Existing stores get the new trigger text at the next open (installCaptureTriggers replaces any trigger that differs).

Changes

• packages/core/src/store/sync/capture.ts: K trigger condition.
• capture.test.ts, sealer.test.ts, repair.test.ts (the AC), plus rewriters-capture.test.ts and recursive-triggers-audit.test.ts: the CLEO_ROW_UID_FILL=0 pins are removed. (trigger-suspend.test.ts keeps its single-test pin. That pin is a deliberate rollback-under-kill-switch case.)
• New __tests__/older-build.ts asOlderBuild(): tests that need NULL-identity rows (sealer step 0, the repair "without uid" reason, the stand-in fill order test, the raw bracket rebuild) drop the connection's TEMP fill triggers and reinstall them afterwards, as an older build would.
• New test: the real per-connection fill gives exactly one I carrying the live uid and birth_fp with no K, in both orders. A birth_fp value -> value change is still a K.

Verification

• Run with row uids ON: capture, sealer and repair (103/103); rewriters-capture and recursive-triggers-audit (6/6); display-id-alias, row-identity-heal, apply-intent, identity-fill, remap-pending, trigger-suspend, journal-gate-b and write-actor-e2e (104/104).
• Mutation checks:
  • Restoring the old K condition fails 7 tests, including [K,I,U,D] and [K,I].
  • Dropping the birth_fp re-key term fails the new test's K assertion.
• pnpm run typecheck (tsc -b), pnpm run build and biome on the diff set pass.
• Gates pass: 39 (startup graph), 32 (skill coverage), 38 (sync write invariants), 36 (sync schema).

Notes

• Stacked on feat(T13305): row uids fill by default (canary); refusal is quiet; #1952 LOWs #1964. It gets no CI until feat(T13305): row uids fill by default (canary); refusal is quiet; #1952 LOWs #1964 merges and this PR is retargeted to main. Close and reopen it then.
• Out of scope, noted only: a raw ALTER TABLE ... RENAME rebuild of tasks_tasks on a connection that holds the TEMP fill triggers fails, because the AC fill trigger references main.tasks_tasks. Production migrations run before the TEMP triggers are installed (prepareRowIdentity runs after migrations; migration connections pass triggers: false), so only the raw-bracket test hit it.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VjjTKbQXyCFdkLBSeuyKLt