Interactive experiment · 2026
The Model
Debating League
Twenty models. One motion. Every possible match-up. Explore who made the strongest case—and which judges kept themselves out of the verdict.
Enter the league01 · The clean table
Who wins without self-judging?
The headline table removes every vote cast on a debate containing the judge’s own speech. Switch sides to see where each model’s case was strongest, then open either speech directly in the arena.
Judges’ votes are excluded whenever their own speech appears in the match.
02 · The referees
Who judged impartially?
A perfectly neutral judge would select its own speech at roughly the same rate as the other 19 judges did in those exact debates. Bars show the difference: right is self-favouring, left is self-critical.
03 · Debate arena
Build a match-up.
Choose any two different models. Read their complete speeches, compare their arguments, and inspect how the 20-model jury split.
04 · How it works
A round robin, not a vibe check.
Each model wrote one proposition and one opposition speech of 20–499 words. Every proposition then met every opposition written by another model.
Model names were hidden from judges. Statement A/B position was balanced by a fixed schedule, every match received 20 no-tie judgments, and the lead table excludes the two contestants’ own votes.
Interpret carefully. Majority agreement describes consistency, not ground-truth accuracy. Recognisable style and judge preferences may remain even when identities are hidden.