simple_qwen3.6-35b_reactive
Qwen3.6-35B on the simple floor, 5 episodes. Not cited by the thesis (development or earlier run).
| Wins (boss defeated, alive) | 0/5 [0.00, 0.43] |
| Final HP / exact optimum, mean over wins | – |
| Died | 0 |
| Final HP, mean | 523 |
| Seconds per model call | 19.9 |
| Parse error rate | 1.2% |
| Floor | Simple floor |
| Model | Qwen3.6-35B (qwen3.6-35b-a3b, thinking) |
| Planning | reactive |
| Prompt | target_boss_seed.txt |
| Temperature | 0 |
| Lookahead depth | 3 |
| Mechanical overrides | on (logged per turn) |
| # | outcome | final HP | HP / opt | steps | model calls | replay |
|---|---|---|---|---|---|---|
| 1 | not won | 530 | – | 36 | 16 | log not stored |
| 2 | not won | 562 | – | 36 | 17 | log not stored |
| 3 | not won | 507 | – | 36 | 16 | log not stored |
| 4 | not won | 507 | – | 36 | 16 | log not stored |
| 5 | not won | 507 | – | 36 | 16 | log not stored |
In a checkout of mt-ai-lab with an API key set (the result goes to a new file rerun_<time>_simple_qwen3.6-35b_reactive, never over the thesis result):
python scripts/run_catalog.py --codes S1
Result file: results/reliability/_ablation_effcost_numeric/simple_qwen3.6-35b_reactive.json · updated 2026-09-20