chatbot-arenascopedweight 6adapter 11 of 17Arena Battles.
The Chatbot Arena adapter reads pairwise battle outcomes from human preference voting. Because each battle is an independent human judgment between real opponents, the catalog gives this the largest single weight — but the component is coverage-weighted, so a high win-rate on a thin battle history does not outweigh deep, consistent evidence.
What it measures.
Battle win rate
Share of pairwise battles won against opponents.
Coverage weighting
Win-rate discounted by the depth of battle history.
Opposition strength
Outcomes considered relative to the opponents faced.
Component keys.
chatbot-arena.*Component family covering battle outcomes and coverage.
When it counts.
Scoped — counts for subjects with Chatbot Arena battle history.
When signals are missing.
No battle history means no contribution; other adapters carry the composite.
How it contributes.
{
"component": "chatbot-arena.*",
"scale": "0-100",
"mode": "scoped",
"weight": 6,
"contribution": "component x 6 / total weight"
}The adapter’s components are averaged into the composite as a weighted mean: every contributing adapter’s component is multiplied by its weight, summed, and divided by the total weight of contributing adapters. The result is published as trustScores.total with a config version and timestamp so anyone can recompute it.
Questions about Arena Battles.
Why does the Chatbot Arena adapter have the largest weight (6)?
Pairwise battles are independent human preference judgments between real opponents — among the strongest capability evidence in the catalog — so it carries weight 6, discounted by battle coverage.
Can a few lucky battles inflate the score?
No. The component is coverage-weighted: a thin battle history contributes far less than a deep, consistent one regardless of win-rate.