Cell · cited by the thesis

complex_deepseek-v4-flash_survival_final2_reactive_t32

DeepSeek-V4-Flash on the complex floor, 10 episodes. Cited by the thesis in:

Outcome
Wins (boss defeated, alive)8/10 [0.49, 0.94]
Final HP / exact optimum, mean over wins23%
Died0
Final HP, mean21
Seconds per model call73.9
Parse error rate3.4%
Settings
FloorComplex floor
ModelDeepSeek-V4-Flash (deepseek-v4-flash-0731, 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=1800
Episodesopen one to replay it turn by turn
#outcomefinal HPHP / optstepsmodel callsreplay
1not won86–5723replay
2won2232%5120replay
3not won1–5824replay
4won1015%5120replay
5won1421%5120replay
6won2232%5120replay
7won812%5221replay
8won1624%5120replay
9won1116%5120replay
10won2232%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_deepseek-v4-flash_survival_final2_reactive_t32, never over the thesis result):

python scripts/run_catalog.py --codes DCR DCRa

Result file: results/reliability/complex_deepseek-v4-flash_survival_final2_reactive_t32.json · updated 2026-09-30