Records
OLSS-II with a two-strategy portfolio at 200 ms against expectimax over sixteen worlds
Registration 4ffa3c2f-c818-4d24-828a-b3d8fc49d73f (study subgame-solving/opponent-portfolio), run 2026-09-19 on AMD Ryzen AI Max (Strix Halo), 32 threads, 125 GB unified memory, Radeon 8060S via ROCm, Linux 6.18. Server commit e3107a0, research commit 9588223 (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 'olss2 n2 200ms=kluss:{"mode":"olss2","samples":64,"expansions":10000,"cfr_every":64,"portfolio":[{"policy":"greedy"},{"policy":"self","parameters":{"temperature":0.3}}]}' --blue 'expectimax:{"samples":16,"depth":2}' --games 200 --seed 0 --rules training --threads 8 --record-dir ../research/runs/4ffa3c2f-c818-4d24-828a-b3d8fc49d73f/records --out ../research/runs/4ffa3c2f-c818-4d24-828a-b3d8fc49d73f/matchup.json --tag 4ffa3c2f-c818-4d24-828a-b3d8fc49d73f
cd research && python3 scripts/paired_contrast.py runs/4ffa3c2f-c818-4d24-828a-b3d8fc49d73f/matchup.json| Player | Wins | Draws | Losses | Effective | 95% Wilson | Mean plies | ms/move |
|---|---|---|---|---|---|---|---|
| olss2 N=2 (greedy, self 0.3; 10,000 expansions) | 76 | 0 | 124 | 0.380 | 0.316 to 0.449 | 147 | 681.656 |
| expectimax (16 worlds, depth 2) | 124 | 0 | 76 | 0.620 | 0.551 to 0.684 | 147 | 8.374 |
| ending: flag_captured | 192 | ||||||
| ending: no_moves | 8 |
Per-seed paired contrast for the olss2 player (score as Red plus score as Blue minus one): mean minus 0.240, 95% normal interval minus 0.390 to minus 0.090 over 100 seeds (20 positive, 36 split, 44 negative). The olss2 player won 46 games as Red and 30 as Blue. Wall time 1,294 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 190 ms for the olss2 side at 10,000 expansions and about 9.5 ms for the matched expectimax cell.
Decision rule
The hypothesis (the olss2 player holds against expectimax, interval above zero) is not supported: the contrast interval minus 0.390 to minus 0.090 lies entirely below zero. The portfolio player beats full kluss at this budget but still loses to the plain depth-2 expectimax that beat full kluss in the previous study.
Provenance
Archived JSON: runs/4ffa3c2f-c818-4d24-828a-b3d8fc49d73f/matchup.json (SHA-256 3ee817fb6652dc1be278c1ddf1469be26faeb0ea940bfd91c129a44e2674c29a), with per-game moves in the arena record files under the same directory. Games planned 200, attempted 200, completed 200, interrupted 0. The process that launched this cell was ended from outside while the matchup was in flight; the matchup process itself was not interrupted and wrote all 200 game records and the aggregate JSON. No deviation from the registration.