Evidence before examples

What belongs here

One record describes one capability, not one social post or prompt. A public repository, demo, video, benchmark, cookbook, or approved write-up must support the claim. Discord and X can lead us to a project, but a login-only message alone is not enough to publish it.

Evidence levels

E1 · Author-reported

A public claim with a stable artifact, not independently inspected in depth.

E2 · Artifact-verified

An editor inspected the public repository, demo, video, or benchmark artifact.

E3 · Editor-reproduced

An editor independently ran the project or evaluation and recorded the conditions.

Featured records must be E2 or E3. Metrics remain marked Reported unless we reproduced them independently.

Official and Community

Official means TypeSafe itself published the example on a TypeSafe-owned site or in its GitHub organization. Community means a third party published the project, even if TypeSafe later reposted or linked to it. Official examples are capped at 30% of the directory.

How records are written

We identify the bounded judgment Jev owns, the deterministic work code still owns, the public outcome, and the limits of that evidence. We do not copy source code, screenshots, long quotations, or private community messages. Model confidence is not treated as correctness.

Who maintains this

One editor and a small collection pipeline. New candidates arrive from public GitHub activity, X posts, community Discord channels, and the submission form; each candidate is scored for how much it genuinely demonstrates about Jev before any verification effort is spent. Publishing runs in small batches — at most three new records per round — so that verification depth does not degrade as the index grows. Nothing enters the directory without passing the five gates described on the methodology page, and the full dataset behind the index is published as machine-readable JSON so every count and claim on this site can be recomputed by anyone.

The index re-verifies on a rolling basis rather than in annual audits: repositories move, demos expire, and authors publish better results. When a source's claims change, the record changes with it or comes down, and the corrections policy describes exactly how to trigger that.

What this site is not

Not affiliated with, endorsed by, or speaking for TypeSafe AI. Not a rankings service — nothing here can be bought, and there are no sponsored records, affiliate links, or paid placements. Not a benchmark: we publish what artifacts demonstrate and label author-reported numbers as reported, which is why the methodology page spends as much space on what the system cannot confirm as on what it can.

Names and independence

“TypeSafe,” “TypeSafe AI,” and “Jev” refer to the company and product documented at TypeSafe's official sites. The bracketed typesafeai.app wordmark, palette, and editorial ledger design are deliberately separate from official TypeSafe branding. No endorsement is implied.