Public Policy Debates · v0.1
Score policy debates
on a published rubric.
Online policy fights produce heat without clarity. DebateGrader turns a real debate event into a four-part verdict — premises, rubric, scores, synthesis — and publishes the rubric before the verdict, so observers can challenge the standard, not just the call.
Featured runs
full audit logFive runs selected to span the rubric library and the range of score patterns — from a perfect tie to the widest score gap in the pool. Click any to see the four-part ballot: premises, rubric, per-criterion scores, synthesis.
How it works
01Ingest & extract
Tag @debategrader on a thread, or paste a URL or transcript at /submit. We normalize speakers, then extract each side’s premises, conclusion, and unstated assumptions — with two model variants in parallel as a check.
02Match a rubric — first
The argument structure is matched to a rubric from our curated library. The rubric is published before the verdict, with its match score, so you can challenge the standard. Below the threshold, we generate a provisional rubric and route it to review.
03Score blind, publish thread
Each speaker is scored against the same rubric independently, blind to the other. The output is a four-part thread — premises, rubric, per-criterion scores with justifications, synthesis — with full provenance to the source and a permanent audit-log entry.