Skip to content

hosted · classifier · typesafe/jev-1.13 via OpenRouter

Jev 1.13

Answers a typed placement question over the candidate ids, a yes/no hold question and a risk score.

Rating

Score
81,825
±95%
±25,834
Rank
#4 of 9

Mean final score over 30 public seeds with gravity paused. The ± is a 95% t-interval across seeds. Beats GPT-5.6 Luna on 23/30 seeds, Qwen 3.7 Flash on 28/30 seeds, GPT-4o mini on 30/30 seeds and Random legal on 30/30 seeds. No evidence of a difference from DeepSeek V4 Flash. Loses to Greedy heuristic, Dellacherie-style and Two-ply search.

Greedy heuristic−192,382−217,776 to −165,6510-0-30
Dellacherie-style−187,919−212,491 to −159,7501-0-29
Two-ply search−185,850−209,300 to −159,3811-0-29
DeepSeek V4 Flash+10,121−20,913 to +44,43114-0-16
GPT-5.6 Luna+39,913+14,844 to +67,63523-0-7
Qwen 3.7 Flash+62,470+37,535 to +88,10228-0-2
GPT-4o mini+79,171+56,493 to +105,15030-0-0
Random legal+81,398+58,619 to +107,16130-0-0

Per opponent: mean score difference over the shared seeds, the 95% bootstrap interval of that difference (lower to upper, not symmetric), then wins-draws-losses by seed.

Reliability

IQ

30/30 games · 8,423 calls · 100% answered · p50 360 ms · p95 470 ms · $0.014 per 100 decisions

Blitz

not run

Latency covers completed calls only, measured from the question leaving the harness to the answer arriving. In Blitz the deadline is the next gravity step, so the share answered in time depends on the level as much as on the brain.

Replays

IQ · gravity paused

seed-0130,814 · 61 lines · top-out →seed-0220,024 · 50 lines · top-out →seed-0338,938 · 72 lines · top-out →seed-0422,536 · 52 lines · top-out →seed-0510,390 · 32 lines · top-out →seed-0650,306 · 82 lines · top-out →seed-07157,384 · 148 lines · top-out →seed-0811,072 · 33 lines · top-out →seed-0919,518 · 48 lines · top-out →seed-1039,066 · 69 lines · top-out →seed-11100,732 · 118 lines · top-out →seed-12249,898 · 187 lines · piece cap →seed-13166,538 · 154 lines · top-out →seed-1482,374 · 104 lines · top-out →seed-15109,586 · 120 lines · top-out →seed-16117,492 · 129 lines · top-out →seed-17189,304 · 163 lines · top-out →seed-18146,888 · 143 lines · top-out →seed-1960,344 · 88 lines · top-out →seed-20279,660 · 195 lines · piece cap →seed-2141,916 · 74 lines · top-out →seed-2260,276 · 90 lines · top-out →seed-2372,974 · 94 lines · top-out →seed-2442,642 · 72 lines · top-out →seed-2572,502 · 100 lines · top-out →seed-2638,092 · 69 lines · top-out →seed-2741,584 · 73 lines · top-out →seed-2880,138 · 104 lines · top-out →seed-2983,468 · 105 lines · top-out →seed-3018,302 · 45 lines · top-out →

Notes

Footnotes
Elo-style rating1545 · 1488 to 1599
Calibration (Brier)0.108 over 8,405 forecasts
Preparation p50 / p953.4 ms / 4.8 ms
IQ counts0 invalid · 0 stale · 0 errors · 0 timed out · 0 unreachable · 0 retries
Holds0 IQ
Adapter version3.0.0
Modeltypesafe/jev-1.13 via OpenRouter
Model settingsmodels.ts ↗

The Elo-style number is a penalised Bradley-Terry fit over IQ seed points, centred at 1500. It is a secondary view of the same games and does not enter the ranking. Calibration is the Brier score of the optional risk forecast against top-out within ten pieces. Preparation is the harness's own time to build the question, excluded from latency.