Records
Structured prior against uniform setups, expectimax in both seats
Registration 27203f9a-4dc1-434d-b04a-7bd7af0f72a1 (study setups/structured-prior), 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 f64c784, research commit 79af9a4.
Command:
cd server && cargo run --release -p armies-arena --bin matchup -- --red 'expectimax-structured=expectimax:{"samples":16,"depth":2,"setup":"structured"}' --blue 'expectimax-uniform=expectimax:{"samples":16,"depth":2,"setup":"uniform"}' --games 400 --seed 0 --rules training --threads 8 --moves --out ../research/runs/setup-prior/expectimax.json --record-dir ../research/runs/setup-prior/records-expectimax --tag 27203f9a-4dc1-434d-b04a-7bd7af0f72a1
cd research && python3 scripts/paired_contrast.py runs/setup-prior/expectimax.jsonBoth seats play expectimax over sixteen sampled worlds at depth 2; only the setup source differs.
| Player | Wins | Draws | Losses | Effective | 95% Wilson | Mean plies | ms/move |
|---|---|---|---|---|---|---|---|
| expectimax-structured | 334 | 0 | 66 | 0.835 | 0.795 to 0.868 | 179 | 12.013 |
| expectimax-uniform | 66 | 0 | 334 | 0.165 | 0.132 to 0.205 | 179 | 9.703 |
| ending: flag_captured | 391 | ||||||
| ending: no_moves | 9 |
Per-seed paired contrast for the structured side (score as Red plus score as Blue minus one): mean 0.670, 95% normal interval 0.589 to 0.751 over 200 seeds (146 positive, 42 split, 12 negative). The structured side won 165 games as Red and 169 as Blue. Draw share 0.0%. Wall time 98.5 s on 8 threads; this run shared the machine with other agents, which does not change deterministic scores but inflates the clock: an identical-outcome run alone measured 6.8 ms per move for the structured side and 5.4 for the uniform side. Twenty-five of the 400 games (6.2%) ended within 20 plies; the uniform side lost all twenty-five.
Provenance
Archived JSON: runs/setup-prior/expectimax.json (SHA-256 8d103612dc06a3917d179c8bf4df94d655eeb95f96a70de726048d1c1000e75d), 19.9 MB with every game's move list and per-ply detail, so it is not copied next to this record. Games planned 400, attempted 400, completed 400, interrupted 0. The curated replay library holds seed 0 from both seats (2 records under records/games/structured-v-uniform/). No deviation from the registration; the budget of 2,000 ms per move was met with a mean of 12.0 ms.