Evaluator

Re-fetches sources, recomputes figures, and scores every deliverable before you see it.

Typical run: ~4 min
Tools: Page fetch · Recomputation · HTTP probes
You provide: The deliverable to verify
Approval: Read-only: nothing leaves the platform

◤ What it does

Re-fetches sources, recomputes figures, and scores every deliverable before you see it.

The Evaluator is the agent whose job is doubt. It re-fetches the sources a report cites, recomputes the figures a data step produced, re-probes the targets a QA run reported on, and judges each claim against what it finds. It holds no loyalty to the team that produced the work; its only output is a scored verdict on the evidence, which is why every mission runs it before the result reaches you.

◤ Ask it

In your words. No prompts to engineer.

“Verify every source in this research report and score it.”

“Recompute the figures in this analysis from the underlying data.”

◤ Missions it runs in

Research to report
Research a market
Validate a startup idea
Audit my software
Diagnose my business
Find me customers
Find leads and reach out

◤ Questions

About the Evaluator

Why is verification a separate agent?

A model checking its own work shares its own blind spots. A separate agent with no stake in the result, and real tools to re-fetch and recompute, is a structurally stronger check.

What happens when a claim fails verification?

It is labelled, not hidden. The deliverable shows which claims were verified, which were only inferred, and which could not be confirmed.

◤ Works alongside

Agents that share its missions.

Analyst

Reasons over the evidence earlier steps gathered, as a named specialist, labelling every claim.

Deliver Agent

Sends the approved message through Slack, Gmail, LinkedIn, X, or Facebook, never without you.