How scoring works

The point

The judge does not decide who is right about the world. It decides who won the round: who better met the burden, answered the other side, and left fewer holes in the transcript.

If the model picked the “correct” opinion, ELO would become a popularity contest for whatever the model already believes. Debloon scores argument performance only.

Four scored criteria (0–4 each)

Clash

35%

Did they answer the opponent's actual arguments, especially the strongest one? Unanswered material claims count as conceded. Two speeches that never touch is a loss for both on this row.

Reasoning

30%

Valid chain from claim to conclusion, internally consistent, with some weighing (why this point beats that one). Fallacies, contradictions, and moving the goalposts cost points here.

Support

20%

Specific, on-point backing. Vague "studies show" and invented citations get no credit. Unverifiable facts are ignored, not rewarded.

Resolution

15%

Stay on the exact topic as written. Aff has to make a real case for the claim. Neg can win by defeating that case. Off-topic rants score 0.

Side score = (clash×0.35 + reasoning×0.30 + support×0.20 + resolution×0.15) × 25 → each debater lands on 0–100.

How a winner is picked

Each side is scored independently from the transcript first — no backfilling.
8+ point margin = clear win. 3–7 = close win. Under 3 = draw (better for ELO than a coin flip).
Tiebreak: if Aff never made a real case for the resolution (resolution score 0), Neg wins by burden even on a numeric tie.

Hard fouls (after scoring)

Fabricated quote, source, or statistic: that claim is void; −15 if it was load-bearing.
Harassment, doxxing, or hate: forfeit.
Pure sabotage / clock-wasting with no argument: forfeit.
Strawman or quote-mine: clash score cannot exceed 2.

The dropped-argument rule keeps judging honest: the judge cannot invent a reply a speaker did not make.

What we never score

Word count, SAT vocabulary, confidence, politeness, humor, or whether the judge agrees with your position. Concise can beat long. Running out of clock is a strategy choice, not a foul.

What Debloon is

Debloon is a structured debate platform: you set a resolution, share an invite, argue in timed rounds, and get a transcript-backed verdict plus ELO that tracks skill over time — not agreement with the AI.

Start a debate →