Score an end-user AI app or feature across 11 criteria, 1 to 5. Complete it by hand, or let an LLM judge draft the scores for you to review.
0
of 55 points · avg - / 5
Not scored yet
Scores save in this browser automatically
How to read it. The average per criterion (out of 5) is the fairest quality read while scoring is partial; the point total and the ring fill both count every active criterion. Drop the AI-tagged criteria if the app is not model-powered - use the toggle in each of those rows to exclude it.
0
of 55 points · avg - / 5
Not judged yet
You can override any score the judge gives
Runs locally only. This tab loads an in-browser model or calls a provider API, which the shared artifact sandbox blocks - use the local file or localhost. Your API key is stored only in this browser and sent only to the provider you pick; it is never uploaded anywhere else. The model judges the content you paste - it cannot open the URL on its own.
-
within-1 agreement
Score the same app on both the Human evaluation and LLM as judge tabs, then come back here to see how closely the judge agrees with you.
Compares the two tabs criterion by criterion
Validate the validator. An LLM judge is only trustworthy where it agrees with a human. This view measures that agreement per criterion so you can see where the judge runs harsh or lenient, and which criteria are too fuzzy for it to score reliably - the core idea from Who Validates the Validators (Shankar et al., 2024).