| Run | Planner | Prompt | Male mean | Stdev | Latency | Cost/call |
|---|---|---|---|---|---|---|
| baseline | Gemini Flash low | original prod | 7.03 | 1.14 | 4.6s | /bin/zsh.0018 |
| iter-2 | Gemini Flash low | Frame Strength | 7.22 (+0.19) | 1.03 | 4.6s | /bin/zsh.0018 |
| iter-16 🥇 | Claude Haiku 4.5 | Frame Strength | 7.25 (+0.22) | 0.76 (tightest) | 5.8s | /bin/zsh.003 |