For AI teams
Human ratings of AI output, with how sure each rater was.
The Puck is how every answer on PuckSurveys is given: drag it along an arc from No to Yes and let go. It records where the rater landed, how long they waited before moving, how much they went back and forth, and how long it all took. A click gives you the answer. The puck also tells you how sure they were.
For teams that collect human judgments on model output: evals, preference data for training, annotation and red-team review.
One phone, one screen.
The laptop shows the conversation and the turn being rated. The rater answers on their phone.
What a rated set looks like.
Five questions on one model reply, answered by 134 example raters. Each bar shows where they landed and how sure they were.
Replay the raters.
Every rater's answer plays back at once, pauses and changes of mind included.
- Live Eval
A rater answers on their phone while a laptop or projector shows what they are rating and which question they are on. Good for a room watching one rater work through a conversation or an image.
- Video
Raters keep the puck moving while a clip plays. Their traces line up on the clip's timeline: the average, the middle half of raters, and the second it went worst.
- Export
A CSV with one row per rater per item: each answer, the pause before it, the back-and-forth, the time taken, and whether they skipped.
Try it.
Nothing you do in the demos is saved.
Talk licensing.
founder@pucksurveys.com