v0.4.3 — Confidence Scores + Brain Pages
Nexus Memory v0.4.3
🧠 New Feature: Confidence Scores in Recall
Every memory returned by recall() now includes a composite confidence score:
| Field | Description |
|---|---|
confidence |
0.0-1.0 composite score (belief trust > provenance confidence > default 0.7) |
confidence_label |
Human-readable: very high / high / medium / low / very low |
evidence_count |
Number of corroborating sources |
lifecycle_status |
canonical / deprecated / rolled_back |
valid_until |
Expiry timestamp (if TTL set) |
How composite confidence works:
- If memory has a belief trust score → use that
- Else if provenance has confidence → use that
- Else default 0.7
- +0.05 per evidence source (max +0.2 boost)
This means agents can now tell their users: "I found this memory with high confidence (0.85, backed by 3 sources)" - instead of just returning text with a relevance score.
📦 Upgrade
cd ~/nexus-memory && git pull && pip install -e .Backwards compatible - existing memories without trust/evidences use provenance confidence.