Cell · cited by the thesis

complex_route_qwen3.5-397b_canon

Qwen3.5-397B on the complex floor, 5 episodes. Cited by the thesis in:

Outcome
Wins (boss defeated, alive)3/5 [0.23, 0.88]
Final HP / exact optimum, mean over wins25%
Died1
Final HP, mean8
Seconds per model call90.0
Parse error rate17.7%
Settings
FloorComplex floor
ModelQwen3.5-397B (qwen3.5-397b-a17b, thinking)
Planningreactive
Promptcomplex_bfs_strategy_survival.txt
Temperature0
Lookahead depth3
Mechanical overridesoff (all engine replacements and assists disabled)
Engine settings (environment, 6)
MOTA_SUPPRESS_HINTS=0
THINKING_MAX_TOKENS=32768
CONVERSATION_HISTORY_TURNS=0
MOTA_ORDER_TWO_SIDED=1
MOTA_FIGHT_KEY_FLOOD=1
LLM_TIMEOUT=300
Episodesopen one to replay it turn by turn
#outcomefinal HPHP / optstepsmodel callsreplay
1died-57–3513replay
2not won49–5723replay
3won2232%5120replay
4won1116%5120replay
5won1725%5120replay
Re-run this cell

In a checkout of mt-ai-lab with an API key set (the result goes to a new file rerun_<time>_complex_route_qwen3.5-397b_canon, never over the thesis result):

python scripts/run_catalog.py --codes VQR VQRa VQR5

Result file: results/reliability/complex_route_qwen3.5-397b_canon.json · updated 2026-10-04