Nothing on the sheet yet.
521 candidates are queued. The crawler judges a batch every 30 minutes; hand Jev a link to go first.
Things people built with Jev, found by a crawler and judged by Jev itself: is it real, how cool is it, and how much does Jev <3 it. Every cent that cost is on the tape.
Nothing on the sheet yet.
521 candidates are queued. The crawler judges a batch every 30 minutes; hand Jev a link to go first.
A crawler searches GitHub, Hacker News, Bluesky and (when paid for) X every 30 minutes. Each candidate goes to Jev, TypeSafe AI's System One model, with one request asking the questions on the right, which run in parallel. Jev returns probabilities, not prose.
Jev judges, code decides. An item passes when all three yes/no gates clear the pass mark you pick above (default 0.7). Between 0.4 and the mark it is borderline and gets a cross-ticked rule. Below 0.4 it is thrown out but stays inspectable at the foot of the sheet. Top ranks by Jev's own taste: 35% how cool Jev finds it, 30% how much Jev likes it, 15% evidence of a working build, 12% freshness, 8% attention. Changing any of this never re-asks Jev.
Costs are list prices, logged when spent. Jev: $0.042 per million input tokens, output free. X API pay-per-use: $0.005 per post read, $0.01 per user read. GitHub, Hacker News and Bluesky reads cost nothing.