Product

The journey monitor and the evidence record.

We run named AI agents through one declared journey. We record each checkpoint. We repeat the run each week and we report what changed.

We deliver this work by hand today. There is no dashboard, no account, no public API, and no self-service software.

What it is

Four parts, one loop

The product is a loop, not a report. Each part gives the next part its input.

Journey monitor

We declare one journey and the checkpoints that the agent must reach. The declaration comes before the run.
Evidence record

Each run writes one dated record with the checkpoint states, the stop point, and the observation that decided it.
Repeated attempts

We attempt the journey more than one time, and with more than one agent family, before we report a stop.
Change watch

We run the journey each week for one month. A change between two dated records is also a finding.

Control loop

The same five steps run in the same order every time. The order lets a person who was not there check the result.

Declare

We write the goal and the checkpoints, and you approve them.
state: declared
Attempt

A named agent family runs the declared journey.
state: attempted
Record

We mark each checkpoint and we write the observation.
state: recorded
Reproduce

We repeat the run and we compare it with a control journey.
state: reproduced
Watch

We run the journey each week and we date every record.
state: watched

Software

We are open about the state of the software, because a buyer must know what they pay for.

A private runner

The runner holds the journey model, the boundary rules, the evidence writer, and the queue. It runs on our infrastructure.
A fixture check

The runner starts with a test against a built-in fixture site. The check must pass before the runner accepts work.
A person in the loop

An operator reads every record before you see it. The record states what the run observed, not what a model believes.
No public surface

This website exposes no job route, no artifact route, and no control route. There is nothing to sign in to.

Boundaries

These limits are current facts. We will change this page when a limit changes.

It does not give a score, a grade, or a certificate.
It does not prove that every AI agent behaves in the same way.
It does not prove that your site caused a stop. It reports what the run observed.
It does not fix your site. It names the team that usually holds the fix.
It does not test load, and it does not test security.
It does not submit a form, buy a product, hold a slot, or file a claim outside a written scope.

The example record shows every field of the deliverable. Read it, then start a pilot.