Cell · cited by the thesis

simple_deepseek-v4-flash_final2_oneshot_plan

DeepSeek-V4-Flash on the simple floor, 5 episodes. Cited by the thesis in:

Outcome
Wins (boss defeated, alive)0/5 [0.00, 0.43]
Final HP / exact optimum, mean over wins–
Died0
Final HP, mean512
Seconds per model call18.9
Parse error rate0.0%
Settings
FloorSimple floor
ModelDeepSeek-V4-Flash (deepseek-v4-flash-0731, thinking)
Planningreactive
Promptengine-default-floor-basic
Temperature0
Lookahead depth3
Mechanical overrideson (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
MOTA_PLAN_STEPS=60
Episodesopen one to replay it turn by turn
#outcomefinal HPHP / optstepsmodel callsreplay
1not won482–361replay
2not won564–361replay
3not won420–361replay
4not won539–361replay
5not won556–361replay
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>_simple_deepseek-v4-flash_final2_oneshot_plan, never over the thesis result):

python scripts/run_catalog.py --codes DSPO

Result file: results/reliability/simple_deepseek-v4-flash_final2_oneshot_plan.json · updated 2026-09-29