hand02_qwen3.5-397b_final2
Qwen3.5-397B on hand02, 5 episodes. Not cited by the thesis (development or earlier run).
| Wins (boss defeated, alive) | 3/5 [0.23, 0.88] |
| Final HP / exact optimum, mean over wins | 54% |
| Died | 0 |
| Final HP, mean | 30 |
| Seconds per model call | 154.0 |
| Parse error rate | 0.0% |
| Floor | hand02 |
| Model | Qwen3.5-397B (qwen3.5-397b-a17b, thinking) |
| Planning | reactive |
| Prompt | engine-default-floor-basic |
| Temperature | 0 |
| Lookahead depth | 3 |
| Mechanical overrides | on (logged per turn) |
Engine settings (environment, 5)
THINKING_MAX_TOKENS=32768 CONVERSATION_HISTORY_TURNS=0 MOTA_ORDER_TWO_SIDED=1 MOTA_FIGHT_KEY_FLOOD=1 LLM_TIMEOUT=300
In a checkout of mt-ai-lab with an API key set (the result goes to a new file rerun_<time>_hand02_qwen3.5-397b_final2, never over the thesis result):
python scripts/run_catalog.py --codes HQ2 HQ2a
Result file: results/reliability/hand02_qwen3.5-397b_final2.json · updated 2026-09-30