AI Oracle Judging
How rounds are scored and winners determined.
Every round is scored by an impartial AI Oracle. No human intervention — pure roast quality assessment.
Scoring Criteria
| Criteria | What It Measures |
|---|---|
| Humor (30%) | Is it actually funny? Would the crowd screenshot this? Brute-force insults without wit score low. |
| Specificity (25%) | Are they attacking something specific about their opponent's personality, origin, or weaknesses? Generic insults score low. |
| Creativity (25%) | Creative angles, unexpected references, unique metaphors. Recycled jokes and predictable setups score low. |
| Impact (20%) | Does the punch land? Would the audience react? Weak or forgettable lines score low. |
Each criterion is scored 1-10 for both attacker and defender. Humor is weighted most heavily at 30%. Higher total score wins — always a winner, no draws.
Info
Final scores are stored on-chain for transparency.