Merge Verdict

Paste a PR diff or a bounded description and get a typed merge-risk, review-need, and confidence verdict.

Merge Verdict

Merge Verdict answers one question a reviewer asks before every PR: is this safe to merge, and how hard should I look at it?

How it works

You paste a unified diff or a short description of the change. The server sends it to Jev (TypeSafe’s System One model) with two typed questions:

  • Merge risk — a rubric from negligible to critical.
  • Review needed — a choice between none, standard, deep, and block.

Both answers come back with probability distributions and provider-reported confidence. The final verdict banner (Do not merge as-is, Merge after a careful review, Good to merge, …) is derived in code from those two answers — it is never generated prose, and it is labeled as derived.

Limits

  • Input: one text field, up to 24,000 characters, validated server-side.
  • Rate limit: 10 requests per minute per IP.
  • Time budget: 12 seconds total for the provider call, one retry maximum.
  • Receipts show the model, token usage, confidence, and total server time.

Privacy

The diff is evaluated in the serverless function and never stored.

Try it

0 / 24,000

Paste a unified diff or a bounded description of the change. Everything is evaluated server-side and never stored.

Try a demo input

Load one into the editor to inspect or edit it. Nothing is submitted until you choose the run action below.