Cell · cited by the thesis

complex_qwen3.8-27b_survival_final2_maxsteps_once_t06

Qwen3.8-27B on the complex floor, 3 episodes. Cited by the thesis in:

Outcome
Wins (boss defeated, alive)0/3 [0.00, 0.56]
Final HP / exact optimum, mean over wins–
Died0
Final HP, mean2
Seconds per model call912.2
Parse error rate0.0%
Settings
FloorComplex floor
ModelQwen3.8-27B (qwen3.8-27b, thinking)
Planningreactive
Promptcomplex_bfs_strategy_survival.txt
Temperature0.6
Lookahead depth3
Mechanical overridesoff (all engine replacements and assists disabled)
Engine settings (environment, 8)
MOTA_SUPPRESS_HINTS=0
THINKING_MAX_TOKENS=32768
CONVERSATION_HISTORY_TURNS=0
MOTA_ORDER_TWO_SIDED=1
MOTA_FIGHT_KEY_FLOOD=1
MOTA_PLAN_STEPS=60
MOTA_PLAN_NO_REPLAN=1
LLM_TIMEOUT=1800
Episodesopen one to replay it turn by turn
#outcomefinal HPHP / optstepsmodel callsreplay
1not won2–581replay
2not won2–581replay
3not won2–581replay
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_qwen3.8-27b_survival_final2_maxsteps_once_t06, never over the thesis result):

python scripts/run_catalog.py --codes SQCMO

Result file: results/reliability/complex_qwen3.8-27b_survival_final2_maxsteps_once_t06.json · updated 2026-09-30