Records
Expectimax over sixteen worlds against greedy
Registration bbb0214b-63cf-46a3-b5a9-a1a7d322770e (study expectimax/beats-greedy), 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 7a66da0, research commit a000925.
Command:
cd server && cargo run --release -p armies-arena --bin matchup -- --red 'expectimax:{"samples":16,"depth":2}' --blue greedy --games 200 --seed 0 --rules training --threads 32 --moves --out ../research/runs/bbb0214b-63cf-46a3-b5a9-a1a7d322770e/expectimax-vs-greedy.json
python3 scripts/paired_contrast.py runs/bbb0214b-63cf-46a3-b5a9-a1a7d322770e/expectimax-vs-greedy.json| Player | Wins | Draws | Losses | Effective | 95% Wilson | Mean plies | ms/move |
|---|---|---|---|---|---|---|---|
| expectimax | 137 | 4 | 59 | 0.695 | 0.628 to 0.755 | 137 | 9.809 |
| greedy | 59 | 4 | 137 | 0.305 | 0.245 to 0.372 | 137 | 0.001 |
| ending: flag_captured | 191 | ||||||
| ending: no_moves | 5 | ||||||
| ending: quiet_limit | 4 |
Per-seed paired contrast for expectimax (score as Red plus score as Blue minus one): mean 0.390, 95% normal interval 0.240 to 0.540 over 100 seeds (58 positive, 24 split, 18 negative). Expectimax won 69 as Red and 68 as Blue. Draw share 2.0%. Wall time 4.6 s on 32 threads.
Provenance
Archived JSON: runs/bbb0214b-63cf-46a3-b5a9-a1a7d322770e/expectimax-vs-greedy.json (SHA-256 e509424b75a9aadd3a11f4ed61b656a02e406192c08df8aa5273d32783aecd8d), 1.4 MB because every game keeps its move list, so it is not copied next to this record. Games planned 200, attempted 200, completed 200, interrupted 0. No deviation from the registration; the budget of 2,000 ms per move was met with a mean of 9.8 ms.