Background
Not all memories are equally certain. A preference observed once should carry less weight than one reinforced 20 times. Hindsight attaches confidence scores (0.0–1.0) to opinion-type memories.
Proposed behavior
remember(content, confidence=1.0) — explicit confidence on store- Confidence updated automatically when the same fact is reinforced across sessions
- Confidence factored into retrieval ranking (higher confidence = higher score)
- Decay over time if not reinforced (ties into temporal decay feature)
Changes needed
- Add
confidence field to storage schema
- Update retrieval scoring to weight by confidence
- Logic to detect and reinforce existing memories (upsert with confidence bump)
Background
Not all memories are equally certain. A preference observed once should carry less weight than one reinforced 20 times. Hindsight attaches confidence scores (0.0–1.0) to opinion-type memories.
Proposed behavior
remember(content, confidence=1.0)— explicit confidence on storeChanges needed
confidencefield to storage schema