complex_qwen3.5-397b_survival_hints_on
Qwen3.5-397B on the complex floor, 5 episodes. Not cited by the thesis (development or earlier run).
| Wins (boss defeated, alive) | 2/5 [0.12, 0.77] |
| Final HP / exact optimum, mean over wins | 95% |
| Died | 3 |
| Final HP, mean | -70 |
| Seconds per model call | 72.2 |
| Parse error rate | 6.9% |
| Floor | Complex floor |
| Model | Qwen3.5-397B (qwen3.5-397b-a17b, thinking) |
| Planning | reactive |
| Prompt | complex_bfs_strategy_survival.txt |
| Temperature | 0 |
| Lookahead depth | 3 |
| Mechanical overrides | off (all engine replacements and assists disabled) |
Engine settings (environment, 2)
MOTA_SUPPRESS_HINTS=0 THINKING_MAX_TOKENS=32768
In a checkout of mt-ai-lab with an API key set (the result goes to a new file rerun_<time>_complex_qwen3.5-397b_survival_hints_on, never over the thesis result):
python scripts/run_catalog.py --codes CS1
Result file: results/reliability/complex_qwen3.5-397b_survival_hints_on.json · updated 2026-09-26