ACE Core 0.4.3 — canonical cognition persistence
Governed Cognition persistence correctness
ACE Core 0.4.3 fixes exact governed-cognition replay across durable database round trips.
A real external Market Intelligence consumer exposed the defect: SurrealDB flexible-object normalization could omit nested optional values, causing content-derived proposal validation to fail closed after restart. Core now persists canonical JSON beside queryable object projections for cognition identities, revisions, heads, proposals, review receipts, selection receipts, and use receipts.
Compatibility
- canonical hash semantics are unchanged
- pre-v176 object-only records remain readable
- schema v176 is additive
- the eleven-tool MCP surface is unchanged
- Core + Intelligence and Domain Pack boundaries are unchanged
Verification
- real SurrealDB proposal → approval → restart → discovery → material use → lifecycle regression
- 7,390-test normal Core gate
- 7,376-test extension-disabled gate
- official lint, security, Canvas, Python, naked-kernel, and Docker checks
- isolated wheel/sdist build and CLI smoke test
GC1 remains active until the clean public 0.4.3 external-consumer journey, failure controls, restart, and attribution evidence are complete.
Full changelog: v0.4.2...v0.4.3