Cell · cited by the thesis

tower_deepseek-v4-flash_final11_reactive

DeepSeek-V4-Flash on the tower (MT0→MT23w), 3 runs. Cited by the thesis in:

Outcome
Highest floor reached, per runMT17, MT8, MT17
Died0
Final HP, mean2822
Seconds per model call9.1
Parse error rate0.3%
Settings
FloorTower MT0→MT23w
ModelDeepSeek-V4-Flash (deepseek-v4-flash-0731, thinking)
Planningreactive
Promptengine-default-tower-basic
Temperature0
Lookahead depth2
Mechanical overrideson (logged per turn)
Engine settings (environment, 21)
THINKING_MAX_TOKENS=8192
CONVERSATION_HISTORY_TURNS=0
MOTA_OFFER_LEVELUPS=1
MOTA_TOWER_HP_FLOOR=0
MOTA_ORDER_TWO_SIDED=1
MOTA_FIGHT_KEY_FLOOD=1
MOTA_SHOP_ROUTE=1
MOTA_REMOTE_ACTIONS=1
MOTA_WAY_UP=1
MOTA_TOWER_FIDELITY=1
MOTA_PLAN_MEMORY=1
MOTA_NO_KEY_PURCHASE=1
MOTA_WAYUP_COMBINED=1
MOTA_FREE_AFFORD=1
MOTA_NPC_TRADES=1
MOTA_KEY_RESERVE=1
MOTA_AFTER_BATTLE=1
MOTA_BUY_QUANTITY=1
MOTA_VAMPIRE_TAG=1
MOTA_EVENT_WAY=1
LLM_TIMEOUT=300
Runsopen one to replay it turn by turn
#outcomefinal HPHP / optturnsmodel callsreplay
1MT177133–588289replay
2MT8263–22096replay
3MT171069–648311replay
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>_tower_deepseek-v4-flash_final11_reactive, never over the thesis result):

python scripts/run_catalog.py --codes T11DR

Result file: results/reliability/tower_deepseek-v4-flash_final11_reactive.json · updated 2026-10-05