Verdicts, not dashboards.

Your business objectives, evaluated on a schedule against signed evidence from your own tools — and told apart from the ones nobody could measure.

The third one is the point

A dashboard shows a number. It is silent about whether it could read that number, so a metric that goes dark looks exactly like one that is fine. An agent reporting “revenue target met” because it could not reach the revenue is worse than no agent at all.

“unknown is the difference between a dashboard and a lie.”

Every objective resolves to one of three answers, and the third is reported as loudly as the others. When a verdict moves from satisfied to unknown, that is flagged separately — because nothing turned red, and treating it as no news is how it gets missed.

What it actually does

Reads your own tools

Metrics come from the MCP gateways your apps already expose. No warehouse, no second copy of your numbers, no ETL to keep honest.

Signs what it observed

Every reading is Ed25519-signed evidence, credited to the app that owns the number rather than the gateway that served it.

Runs on a cadence

Each desk declares how often it reports. A run records what moved since the last one — the thing a one-off check can never tell you.

Scoped, not prompted

An agent holds a toolset and nothing else. What it may call is enforced by what it was handed, not by asking it nicely.

And there is a boardroom

The verdicts can be delivered by a room of C-level agents who say what they found and cite the measurement behind it. It is a reporting surface, deliberately: it is never in the path between an agent and its tools, and a meeting that could not render still produced its verdicts.