News → Polymarket re-ranker · case explorer

How each model judged every case

Pick a news item to see the candidate markets it retrieved, the hand label for each, and how gpt-5.4-mini, qwen3.5-9b and gemma3:1b scored it. Each cell is shaded by whether the model's gate decision (score ≥ 0.50 + top-5) matched the truth.