cognitive-patterns — L2 self-knowledge surface

Framework context

L2 meta-skill. Aggregates pattern/approach/tone frequencies across summaries, separately counts breakthrough-only occurrences of each pattern. Synthesizes insight: "breakthroughs most associate with {top_pattern} and tend to happen when {top_tone}."

Unlike tunnel-state (what is the state), this is WHO is the thinker.

When to invoke

• User asks "when do I do my best thinking"
• User asks "what happens right before breakthroughs"
• Self-reflection / coaching context
• Before alignment-check, when user wants to know if the decision style matches past breakthroughs

Input

domain: str | None   # if None, aggregates across whole corpus

Output contract

## 🧬 Cognitive Patterns {domain or '(all domains)'}
**Conversations**: N · **Breakthroughs**: M

### Cognitive patterns (top 10)
- {pattern}: count (💎 N of them led to breakthroughs) [N citations available]
...

### Problem-solving approaches (top 10)
- {approach}: count (💎 N breakthroughs)
...

### Emotional tones (top 8)
- {tone}: count (💎 N breakthroughs)

### Breakthrough insight
Breakthroughs most associate with **{top_pattern}** cognitive pattern and tend to happen when you're feeling **{top_tone}**.

Does NOT do

• Claim causation — correlation only
• Prescribe a pattern as "better"
• Include raw conversation text (this is a meta-view)

Fallback

No summaries → basic activity stats (msgs/questions/sources/months), with footer "Cannot determine cognitive patterns without summaries — showing activity stats".

Verification checklist

• [ ] Breakthrough counts are separate from total counts
• [ ] Each pattern row notes "citations available" — caller can drill down via search-summaries
• [ ] Insight is a synthesis of top_pattern + top_tone (deterministic, not LLM-generated)
• [ ] Empty DB → explicit "no data found"