thinking-trajectory — L2 evolution synthesis

Framework context

L2 skill answering "how has my thinking on X changed over time." Three views projecting different slices:

• full: semantic + keyword temporal distribution + thinking-stage progression
• velocity: ACCELERATING (recent > 1.5× early) / DECLINING (recent < 0.5× early) / STABLE / INSUFFICIENT (<6 periods)
• first: genesis moment + span in days + first 300ch of the first mention

When to invoke

• User seems stuck repeating a question they've already explored
• User asks "how did I get here" / "when did I first think about X"
• Before a decision, to see if the user's own thinking has accelerated, stalled, or reversed
• Verify claim of "I just started thinking about X" (often untrue in this corpus)

Input

topic: str
view: "full" | "velocity" | "first" = "full"

Output contract

Full view:

## 📈 Thinking Trajectory: {topic}

### Timeline (by month)
2025-08  ███████ 42 msgs  [top_conv_id · date]
2025-09  ███ 18 msgs   [top_conv_id · date]
...

### Thinking stage progression
🔍 exploring (2025-08 to 2025-10) → 💎 crystallizing (2025-11) → 🔧 refining (2026-02) → 🚀 executing (2026-03+)

### First mention
[conv_id · 2025-08-14] > "{300ch preview}"

Velocity view:

## ⚡ Trajectory Velocity: {topic}
📈 ACCELERATING (recent 3 months avg: X, early 3 months avg: Y)
or 📉 DECLINING or ➡️ STABLE or 📊 INSUFFICIENT DATA (<6 periods)

First view: single row — genesis_timestamp, first_title, first_300ch_preview, total span in days, total mention count.

Does NOT do

• Interpret trajectory direction as good/bad (user interprets)
• Predict future mention count
• Conflate "mention" with "understanding" — a topic can peak in mentions while thinking stage remains exploring

Verification checklist

• [ ] Velocity formula thresholds are 1.5× / 0.5× (hard thresholds, not tunable in this version)
• [ ] Stage progression is ordered by stage_order dict: exploring=0 → crystallizing=1 → refining=2 → executing=3
• [ ] First-mention has exact timestamp + title + citation
• [ ] <6 periods → explicit INSUFFICIENT DATA, not silent stable