Records
KLUSS at the 200 ms budget against expectimax over sixteen worlds
Registration 516cedd8-8bf2-4ae8-b287-c0ab50ce019c (study subgame-solving/kluss-vs-expectimax), run 2026-09-18 on AMD Ryzen AI Max (Strix Halo), 32 threads, 125 GB unified memory, Radeon 8060S via ROCm, Linux 6.18. Server commit 3dcb590, research commit 2b0100a (registration files as committed; this record added afterwards). Other agents shared the machine during the run; per-seed results are deterministic regardless of load.
Command:
cd server && cargo run --release -p armies-arena --bin matchup -- --red 'kluss 200ms=kluss:{"samples":64,"shadows":4,"expansions":5000,"cfr_every":64}' --blue 'expectimax:{"samples":16,"depth":2}' --games 200 --seed 0 --rules training --threads 8 --record-dir ../research/runs/516cedd8-8bf2-4ae8-b287-c0ab50ce019c/records --out ../research/runs/516cedd8-8bf2-4ae8-b287-c0ab50ce019c/matchup.json --tag 516cedd8-8bf2-4ae8-b287-c0ab50ce019c
cd research && python3 scripts/paired_contrast.py runs/516cedd8-8bf2-4ae8-b287-c0ab50ce019c/matchup.json| Player | Wins | Draws | Losses | Effective | 95% Wilson | Mean plies | ms/move |
|---|---|---|---|---|---|---|---|
| kluss (5,000 expansions) | 53 | 1 | 146 | 0.268 | 0.211 to 0.333 | 176 | 415.405 |
| expectimax (16 worlds, depth 2) | 146 | 1 | 53 | 0.733 | 0.667 to 0.789 | 176 | 6.442 |
| ending: flag_captured | 192 | ||||||
| ending: no_moves | 7 | ||||||
| ending: quiet_limit | 1 |
Per-seed paired contrast for kluss (score as Red plus score as Blue minus one): mean -0.465, 95% normal interval -0.607 to -0.323 over 100 seeds (14 positive, 26 split, 60 negative). Kluss won 25 games as Red and 28 as Blue. Mean tree size 121,500 nodes per kluss decision. Wall time 945 s on 8 threads. Time per move is as played in the arena under load; the calibration bench (single thread, middlegame positions, other jobs sharing the machine) read 219 ms at 5,000 expansions, 168 ms on a quieter repeat.
Decision rule
The hypothesis (kluss beats expectimax at this budget) is not supported: the contrast interval -0.607 to -0.323 lies entirely below zero.
Provenance
Archived JSON: runs/516cedd8-8bf2-4ae8-b287-c0ab50ce019c/matchup.json (SHA-256 cbf0a9677974301ff4a92762d9aec20d165e7a698d83c2aefa6bcc86cee89cf3), with per-game moves in the arena record files under the same directory. Games planned 200, attempted 200, completed 200, interrupted 0. The seed-0 records of both seats are in the replay library under records/games/kluss-vs-expectimax/. The kluss contestant is labelled kluss 200ms in the records so the two cells of the study have distinct ids. No deviation from the registration.