Jev Radar

Mostly author-reported. The README states that numbers inside entries are as reported by their authors unless marked reproduced, and grades entries by how the evidence was earned: tier A means the artifact was inspected, such as reading the README or calling the API; tier B is a primary source inspected at description level; tier C is directory-indexed only until a later scan opens it. Point-in-time counts are disclosed as such.

Independent community research, explicitly not affiliated with TypeSafe AI. The maintainer discloses that EveryInfra sells data and search APIs and that this is an interest worth knowing about, and answers it by publishing editorial criteria, confidence grading, and scan logs in the repo. Submissions without primary sources are not accepted. Content is CC BY 4.0 and the data CC0.

The offer is 5,000 keys this round, one per person, issued manually after you star the repo and email the single published address with your GitHub username. The keys come from the maintainer's own quota rather than from TypeSafe and are not affiliated with TypeSafe AI. The repo also ships an access guide covering the waitlist and no-wait alternatives such as gateways and local options.

More of the former than most. The field-tested lessons include a 12-way Choice scoring 0.40 on real bookkeeping data, rows with confidence at or above 0.9 being only 72.2% accurate in one large independent eval, evidence order flipping 5.8% of one verifier's rulings, and junk input being tolerated by LLMs but not by Jev. It also flags a circulating wave of sped-up fake demos and says which repos carry real traces.

Each registry entry carries a verification tier: A (confidence 0.90) means the artifact was inspected directly, by reading the repo README, fetching the live page, reading the original post or calling the API; B (0.75) is a primary source inspected at description level; C (0.60) is directory-indexed only, promoted once opened.

An automated sweep across GitHub, X and five community directories re-runs every three hours, and anything new is verified, graded and committed. The repo says it has received at least one real commit every day since 2026-09-19.

Content is CC BY 4.0 and data is CC0, and the project states it is independent community research, not affiliated with TypeSafe AI. Reported star counts are point-in-time snapshots from 2026-09-19.