Memory Flow: Conversation to Long-Term Understanding

How salience filtering determines what moves from working memory (active-context) to episodic snapshots to long-term semantic understanding.

Memory Flow: Conversation to Long-Term Understanding How salience filters determine what moves from working memory to permanent storage Conversational Context Current session dialogue Questions, answers, decisions being discussed right now Working Memory (active-context.md) Current projects, pending decisions, active threads Updated every session, holds what matters RIGHT NOW Salience Filter Does this moment have: • Emotional weight (breakthrough, frustration, delight) • Decisional significance (chose X over Y, here's why) • Pattern shift (new understanding, approach change) High salience Low salience Not captured Transient conversational noise, doesn't persist Episodic Memory Specific moments captured • "Feb 15: Chose Vector name" • "Breakthrough on catalogs" Pattern Recognition Over multiple sessions: Similar moments cluster, patterns emerge Semantic Memory General understanding • "User prefers spatial thinking" • "ADHD requires Focus Shepherd" • "Catalogs beat full-load" Example flow: Session: "We should use catalogs instead of loading everything" Working memory: Adds "catalog optimization" to pending decisions Salience check: HIGH (decisional significance, efficiency breakthrough) → Capture this moment Episodic: "Mar 5: Decided catalog system, 94% reduction achieved" Cataloged with emotional tag: breakthrough, pride Semantic (over time): "On-demand loading is core to efficiency" General principle, no longer tied to specific moment Salience is what prevents memory bloat. Not everything needs to persist. Emotional weight + decisional significance = worth remembering long-term.

Memory in Continuity Bridge follows a natural cognitive pattern: not everything that's said becomes long-term knowledge. Conversational context flows into working memory (active-context.md), where it's held temporarily. The salience filter—emotional weight, decisional significance, pattern shifts—determines what gets captured as episodic memory. Over time, patterns across multiple episodic moments consolidate into semantic memory: general understanding about the user, their preferences, and what works. This mirrors how human memory actually functions: high-salience experiences become stories, stories reveal patterns, patterns become wisdom.