It ran. It wasn't verified. The receipt says so.

This is one demo run on our own repo, from Grok Build over MCP. We didn't pick a good one or cut the failures.

Demo run on our own repo · 3:20 · one export line masked (0:13–0:34), nothing cut.
48-second cut of the same run · 0:48.

Try it on your repo — Starter $0

What you'll see

  • The first attempt was rejected before it started: no provider key was set.
  • The run then finished with status SUCCEEDED.
  • Evaluation: NOT_RUN. Correctness: INCONCLUSIVE. Human review required.
  • We asked for auto mode; it ran in paper mode. The record shows that too.

Part of the screen is masked from 0:13 to 0:34 (one export line). Nothing else is edited.

The receipt

First two pages of the saved run report. The attribution_key on page 2 is blacked out.

Saved run report, page 1: run ID, status SUCCEEDED, recorded output, evaluation_status NOT_RUN Saved run report, page 2: model attribution and topology evidence, attribution_key masked

We cannot prove your code is right. We can show what ran and what didn't get checked.