Runs and proof

Most AI tools end a job with a checkmark and your hope. MeshBoard ends it with evidence. This page is what that looks like between pressing Ask and reading the receipt.

The life of an ask

When you send work in, it becomes a task you can watch:

• Open — captured, waiting for the right machine.
• In progress — a machine claimed it; you can see which one, and which attempt it's on.
• In review — the work exists and is being checked before it lands.
• Done / merged — finished, with the artifact and the trail attached.

Attempts are visible, not hidden: if the first machine fails, you see the retry and where it went next. A task that can't proceed says so and names what it needs from you — it doesn't sit silently.

Done means proven

Work doesn't count as finished because a model said it went well:

• Code work lands as a pull request — the diff, the branch, and the link arrive with the task, and verification commands attached to the task must actually pass before it settles.
• Questions come back as an answer with its receipt — which machine, which model, how it verified what it claims.

If verification fails, the task says failed, with the trail that proves it: what ran, where it stopped, what it needs. A false "done" is treated as a worse outcome than an honest failure.

Receipts

Every run leaves one. It names the machine, the model, the token count, and the cost — subscription runs at your subscription's rates, metered runs priced per call, local runs at zero per token.

Receipts live where the work does: open any task at app.meshboard.ai and the receipt is attached to it — the run's trail, artifact links, and cost on one page.

The receipt is the answer to the question every other tool makes you take on faith: what did this actually do, and what did it actually cost?

When something fails

Failures are first-class citizens: they get the same trail as successes. The common ones — a machine offline, an AI account signed out, a stuck run — have standard fixes on the troubleshooting page.