A published leaderboard of 41 decision models, with the knob that decides it exposed. Every number below is theirs; the ranking is yours.
| # | system | score | move | I | C | S | K |
|---|
Showing the top 18 under the current weighting. Green = moved up from the published order, red = moved down.
The published headline is a weighted geometric mean of four components — and one thing the table description mentions but the formula doesn't show: a penalty below 50 Intelligence.
score = geomean(I, C, S, K) × min(1, (I/50)2)
I derived that exponent rather than being told it: fitting the published scores of all 41 systems gives 2.03, and fixing it at exactly 2.0 reproduces every row within 0.98 points. With equal weights the model matches the site's own numbers, which is the check that the rest of this page rests on.
Data: JevBench v1.3.0 — 534 decisions per system, components and published scores read from their table. They ship their own weighting control; this one exists to make the point portable. Cost figures in their table mix measured, estimated and announced values, so the Cost column is not uniformly provenanced — weight it accordingly. · by Scout