Records

The pool's exploiter against the pool: the exploitability proxy

Registration ca2fd11a-ff2c-40d5-b5e6-f8a2370e5962 (study setups/selection-by-self-play), 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 0eb994c, research commit 19c62a6. The run shared the machine with other agents' jobs, which inflates the per-move clocks but not the deterministic scores.

Command:

cd server && target/release/matchup -- --red 'pool-exploiter=expectimax:{"samples":16,"depth":2,"setup":"B43546BFB3/3652194B37/362B25225B/4X62872782"}' --blue 'pool=expectimax:{"samples":16,"depth":2,"setup":"pool:../research/content/assets/setup-pool.json"}' --games 400 --seed 0 --rules training --threads 8 --out ../research/runs/setup-selection/cell-exploiter-v-pool.json --record-dir ../research/runs/setup-selection/records-exploiter-v-pool --tag ca2fd11a-ff2c-40d5-b5e6-f8a2370e5962

Both seats play expectimax over sixteen sampled worlds at depth 2. The red side always plays the fixed arrangement the exploit procedure found against the pool (procedure seed 20260920, 16,368 games); the blue side draws uniformly from the selected 32-entry pool. The evaluation seeds 0 to 199 are disjoint from every procedure seed.

PlayerWinsDrawsLossesEffective95% WilsonMean pliesms/move
pool-exploiter22681660.5750.526 to 0.6232974.614
pool16682260.4250.378 to 0.4742975.039
ending: flag_captured271
ending: no_moves121
ending: quiet_limit8

Per-seed paired contrast for the exploiter side: mean 0.150, 95% normal interval 0.043 to 0.257 over 200 seeds (79 positive, 71 split, 50 negative). The exploiter won 105 games as Red and 121 as Blue. Draw share 2.0%. No game ended within 20 plies. Wall time 73.7 s on 8 threads; the budget of 2,000 ms per move was met with a mean of 4.6 ms.

This is the pool's exploitability proxy. The prior's proxy, the same measurement against the prior's exploiter, reads 0.458 (0.364 to 0.551) (record): the difference between the two proxies is minus 0.308 with a 95% interval of minus 0.450 to minus 0.166, so the pool is the less exploitable source and the study's rule for this pair of cells is met.

Provenance

Archived JSON: runs/setup-selection/cell-exploiter-v-pool.json (SHA-256 f1e5e5d1d1ebcc6964468c3e987bdc384b8046dc19409920486274255724cb51); per-game records under runs/setup-selection/records-exploiter-v-pool/. Games planned 400, attempted 400, completed 400, interrupted 0. The curated replay library holds seed 0 from both seats under records/games/setup-pool/. No deviation from the registration.