Floor

Simple floor

The 11 × 11 boss floor of the published solver's test maps: the floor the observation was developed on, and the thesis's running example.

13 HPexact optimum: the most HP left when the boss falls
13 HPanswer of the published solver
1winning sets of cleared monsters and doors (1 optimal)
11 × 1112 decisions on the optimal route · 412 search states

Optimal route

Route on the map
HP470final 13
ATK / DEF9 / 4
Keys0 / 0 / 0yellow / blue / red
Step0 / 2112 decisions
0
1
2
3
4
5
6
7
8
9
10
0
24
1
114
2
114
3
56
4
55
5
182
55
6
24
7
55
56
8
9
114
182
114
10
!
Original game tiles (24-floor sample tower, mota-js). Number on a monster = HP it costs to kill now (green: winnable, red: not). Hover any tile for its stats.
Exact route · 13 HP

Every decision clears a door or fight node; items and fights that cost nothing are taken automatically ("auto"). 412 states expanded.

Optimal route against the models' wins

184 model episodes won this floor; 2 reach the optimum of 13 HP. Below, monsters and doors only, the exact optimal route next to the most common winning trajectory (178 wins clear the same monsters and doors in the same order), as played in this Qwen3.5-397B episode (final HP 2).

turn 15 The trajectory leaves the optimal route here: Fight Slime Lord at (1,9) where the route takes Bat (3,5). The state value (the most HP still reachable at the boss) falls from 13 to 2 HP.

Open the episode at turn 15 to read the model's reasoning there.

Exact optimal route12 decisions
#clearsHP after
1Black Slime (6,4)415
2Green Slime (5,6)391
3Black Slime (3,7), yellow door (4,7)336
4yellow door (7,0), yellow door (8,0), Green Slime (9,0)320
5Bat (3,5)451
6yellow door (7,5)526
7Black Slime (9,5)566
8Slime Lord (1,9)589
9Red Slime (10,7), Slime Lord (9,9)726
10Slime Lord (2,1), Slime Lord (4,2)682
11Bat (5,9)601
12Boss (5,10)13
Most common winning trajectory178 episodes
turnactionHP aftervalue
2Fight Black Slime at (6,4)41513
3Fight Green Slime at (5,6)39113
4Open yellow door at (4,7)39113
5Fight Black Slime at (3,7)33613
9Open yellow door at (7,0)33613
10Open yellow door at (8,0)33613
11Fight Green Slime at (9,0)32013
15Fight Slime Lord at (1,9)leaves the route5202
18Fight Bat at (3,5)5882
22Open yellow door at (7,5)6632
24Fight Black Slime at (9,5)7032
27Fight Red Slime at (10,7)7542
28Fight Slime Lord at (9,9)7152
31Fight Slime Lord at (4,2)6932
32Fight Slime Lord at (2,1)6712
35Fight Bat at (5,9)5902
36Fight Boss at (5,10)22

Model episodes on this floor

By model
modelepisodeswonbest final HP
Qwen3.5-397B22310710
DeepSeek-V4-Flash1562013
Qwen3.6-35B963313
Llama-3.1-8B320–
Qwen3.8-27B271010
GLM-5.3-Flash15142
Mistral-Medium-3.550–

All stored episodes on this floor, every configuration together; the thesis reports them per cell (see Results).

Wins (184)
modelfinal HPof optimumcell
Qwen3.6-35B13100%simple_qwen3.6-35b_final2_la2replay
DeepSeek-V4-Flash13100%gepa_result_gs2_dsv4flash_fullreplay
Qwen3.5-397B1077%simple_qwen3.5-397b_basic_k1k2_hist0replay
Qwen3.8-27B1077%simple_qwen3.8-27b_final2_plan5_t06replay
Qwen3.5-397B1077%simple_qwen3.5-397b_final2_la3_t10replay
DeepSeek-V4-Flash431%simple_deepseek-v4-flash_blockergraphreplay
Qwen3.5-397B215%simple_qwen3.5-397b_basicreplay
Qwen3.5-397B215%simple_qwen3.5-397b_legacy_strategic_controlreplay
Qwen3.5-397B215%simple_qwen3.5-397b_legacy_strategic_controlreplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_order_awarereplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_order_awarereplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_order_awarereplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_order_awarereplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_order_awarereplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_order_awarereplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_order_awarereplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_order_awarereplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_order_awarereplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_order_awarereplay
GLM-5.3-Flash215%simple_glm-5.3-flash_basic_order_awarereplay
GLM-5.3-Flash215%simple_glm-5.3-flash_basic_order_awarereplay
GLM-5.3-Flash215%simple_glm-5.3-flash_basic_order_awarereplay
GLM-5.3-Flash215%simple_glm-5.3-flash_basic_order_awarereplay
GLM-5.3-Flash215%simple_glm-5.3-flash_basic_order_awarereplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_d1d4replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_d1d4_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_d1d4replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_d1d4_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_d1d4replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_d1d4replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_d1d4_t06replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_d1d4replay
GLM-5.3-Flash215%simple_glm-5.3-flash_basic_d1d4replay
GLM-5.3-Flash215%simple_glm-5.3-flash_basic_d1d4replay
GLM-5.3-Flash215%simple_glm-5.3-flash_basic_d1d4replay
GLM-5.3-Flash215%simple_glm-5.3-flash_basic_d1d4replay
GLM-5.3-Flash215%simple_glm-5.3-flash_basic_d1d4replay
Qwen3.8-27B215%simple_qwen3.8-27b_basic_d1d4replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_k1k2_16kreplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_la2replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_k1k2_16k_hist0replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_la2replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_k1k2_16k_hist0replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_k1k2_16k_hist0replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_k1k2_16k_hist0replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_k1k2_16k_hist0replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_k1k2_16k_hist0replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_k1k2_16k_hist0replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_k1k2_16k_hist0replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_k1k2_16k_hist0replay
Qwen3.6-35B215%simple_qwen3.6-35b_basic_k1k2_16k_hist0replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0_compactreplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0_compactreplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0_compactreplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0_compactreplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0_compactreplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0_compactreplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0_compactreplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0_compactreplay
Qwen3.5-397B215%simple_qwen3.5-397b_basic_k1k2_hist0_compactreplay
Qwen3.5-397B215%simple_qwen3.5-397b_final_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final_la3replay
Qwen3.6-35B215%simple_qwen3.6-35b_final_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la3replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la3replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la3replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la3replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la2replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la2replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la2replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3replay
Qwen3.6-35B215%simple_qwen3.6-35b_final2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la2replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3replay
Qwen3.6-35B215%gepa_result_cgepa_simple_35breplay
Qwen3.6-35B215%gepa_result_cgepa_simple_35breplay
Qwen3.6-35B215%gepa_result_cgepa_simple_35breplay
DeepSeek-V4-Flash215%gepa_result_gs_dsv4flash_fullreplay
DeepSeek-V4-Flash215%gepa_result_gs_dsv4flash_fullreplay
Qwen3.6-35B215%gs_35b_fullreplay
DeepSeek-V4-Flash215%gs_dsv4flash_fullreplay
Qwen3.6-35B215%gs_35b_fullreplay
Qwen3.5-397B215%gs_397b_fullreplay
Qwen3.5-397B215%gs_397b_fullreplay
DeepSeek-V4-Flash215%gs2_dsv4flash_fullreplay
Qwen3.8-27B215%simple_qwen3.8-27b_final2_reactivereplay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_final2_reactivereplay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_final2_reactivereplay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_final2_reactivereplay
Qwen3.8-27B215%simple_qwen3.8-27b_final2_reactivereplay
Qwen3.8-27B215%simple_qwen3.8-27b_final2_reactivereplay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_final2_reactive_t32replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_plan2replay
Qwen3.8-27B215%simple_qwen3.8-27b_final2_reactive_t32_t06replay
Qwen3.8-27B215%simple_qwen3.8-27b_final2_reactive_t32_t06replay
Qwen3.8-27B215%simple_qwen3.8-27b_final2_reactive_t32_t06replay
Qwen3.8-27B215%simple_qwen3.8-27b_final2_plan2_t06replay
Qwen3.8-27B215%simple_qwen3.8-27b_final2_plan2_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3_t06replay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_final2_reactive_t32_t06replay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_final2_reactive_t32_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3_t10replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3_t10replay
Qwen3.5-397B215%simple_qwen3.5-397b_final2_la3_t10replay
GLM-5.3-Flash215%simple_glm-5.3-flash_final2_la3replay
GLM-5.3-Flash215%simple_glm-5.3-flash_final2_la3replay
GLM-5.3-Flash215%simple_glm-5.3-flash_final2_la3replay
GLM-5.3-Flash215%simple_glm-5.3-flash_final2_la3replay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_effnet_t32replay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_effnet_t32replay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_canonreplay
Qwen3.6-35B215%simple_qwen3.6-35b_canonreplay
Qwen3.5-397B215%simple_qwen3.5-397b_canonreplay
Qwen3.5-397B215%simple_qwen3.5-397b_canonreplay
Qwen3.5-397B215%simple_qwen3.5-397b_canonreplay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_canon_planmemreplay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_canon_planmemreplay
Qwen3.5-397B215%simple_qwen3.5-397b_canonreplay
Qwen3.5-397B215%simple_qwen3.5-397b_canonreplay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_canon_t06replay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_canon_t10replay
DeepSeek-V4-Flash215%simple_deepseek-v4-flash_canon_t10replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_planmemreplay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_planmemreplay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_hp60replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_hp60replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_t10replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_t10replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_t10replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_t10replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_t10replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_t06replay
Qwen3.5-397B215%simple_qwen3.5-397b_canon_t06replay