Rate the response against its context. Model identities, judge scores, and author expectations are hidden. This is rubric development, not a final benchmark.
This file works offline. Ratings stay in this browser and are downloaded as JSON. Nothing is sent to a server.
I can independently assess Irish grammar, meaning, and naturalness (required for the Irish track).
Choose “cannot judge” when context or expertise is insufficient. Explain ambiguous tasks or rubric problems in notes. Do not consult other judges or the source ledger before finishing.