Memory Policy v2: explicit remember/update/ignore/reject decisions

[aurexav]

Summary

Introduce a first-class memory policy engine with explicit decision outcomes.

Problem

Current ingest outcomes are not fully normalized into a consistent policy taxonomy across pipelines.

In Scope

• Define ingestion outcomes:
  • remember
  • update
  • ignore
  • reject
• Persist decision metadata in audit paths.
• Add config controls by note type, scope, and confidence signals.

Out of Scope

• Training a new model for policy decisions.
• UI policy authoring tooling.

Deliverables

• Decision model and persistence updates.
• Config schema and validation.
• Determinism-focused tests.

Acceptance Criteria

• Every ingest result includes a policy decision category.
• Policy decisions are queryable from history/audit data.
• Invalid policy config fails validation.
• Regression tests confirm deterministic behavior.

Dependencies

• Ingestion: evidence-bound entities/relations for add_note and add_event #49

Implementation Checklist

• Decision enum and mapping.
• Audit schema updates.
• Config schema and docs.
• Deterministic tests.

Done When

• Ingestion behavior is explicit, auditable, and policy-driven.

[aurexav]

Shipped in commit 0c1d7dd on main.

Verification:

• cargo make checks
• cargo make test-integration
• cargo make e2e

What landed:

• policy_decision
• memory.policy.rules
• memory_ingest_decisions