Records
Structured prior against uniform setups, rollout in both seats
Registration 73c92dcd-e644-4cee-8995-883d7c9fb0ff (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 'rollout-structured=rollout:{"setup":"structured"}' --blue 'rollout-uniform=rollout:{"setup":"uniform"}' --games 400 --seed 0 --rules training --threads 8 --moves --out ../research/runs/setup-prior/rollout.json --record-dir ../research/runs/setup-prior/records-rollout --tag 73c92dcd-e644-4cee-8995-883d7c9fb0ff
cd research && python3 scripts/paired_contrast.py runs/setup-prior/rollout.jsonRollout plays with the registry defaults (eight worlds, four random playouts of 24 plies per candidate); only the setup source differs.
| Player | Wins | Draws | Losses | Effective | 95% Wilson | Mean plies | ms/move |
|---|---|---|---|---|---|---|---|
| rollout-structured | 322 | 8 | 70 | 0.815 | 0.774 to 0.850 | 251 | 6.864 |
| rollout-uniform | 70 | 8 | 322 | 0.185 | 0.150 to 0.226 | 251 | 6.199 |
| ending: flag_captured | 371 | ||||||
| ending: no_moves | 21 | ||||||
| ending: quiet_limit | 8 |
Per-seed paired contrast for the structured side (score as Red plus score as Blue minus one): mean 0.630, 95% normal interval 0.551 to 0.709 over 200 seeds (138 positive, 52 split, 10 negative). The structured side won 155 games as Red and 167 as Blue. Draw share 2.0%. Wall time 83.8 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 4.8 ms per move for the structured side and 4.4 for the uniform side. Twenty-nine of the 400 games (7.2%) ended within 20 plies; the uniform side lost all twenty-nine.
Provenance
Archived JSON: runs/setup-prior/rollout.json (SHA-256 695b41c77b7d0d359c9ce46870671bed5436f64f94f4c45f481e3e263b48bbb7), 26.8 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 6.9 ms.