V2 P1: Outcome-driven memory updating — ledger + usefulness scores

[cursor]

Background

V2 Enhancement 5: memory quality improves/degrades from post-action outcomes. Schema stub exists in V004.

Problem Statement

No action ledger; trust score is overloaded; Outcome linker (#63) not yet a closed feedback loop.

Proposed Solution

• Action ledger (append-oriented): agent, task, memory used, decision, action, outcome, trace_id
• Separate scores: source_trust, memory_confidence, usefulness_score, freshness_score, outcome_reliability, action_confidence
• Outcome events update usefulness; never erase historical evidence
• MCP cortex_record_outcome
• Coordinate with / supersede implementation detail of Phase 8: Implement outcome linker (decision → metrics/incidents) #63

Acceptance Criteria

• High-impact actions traceable to memory + policy inputs
• Failed outcomes can challenge a procedure
• Successful outcomes increase procedure confidence
• Historical evidence preserved

Related

#63 Phase 8 outcome linker

Plan ref

todo p1-later — V2 Enhancement 5