Method

Declare, attempt, record, reproduce, watch.

We use the same five steps in the same order for every run. The order makes each stop checkable by a person who was not there.

Public research stops before form data. A customer test uses the written scope that you approve.

The five steps

One order, every run

Declare

You state the journey that earns money. We write the goal and the checkpoints in your words, and you approve the text before a run starts.
state: declared
Attempt

A named agent family runs the declared journey from a clean session. We record the identity that we declare to your site.
state: attempted
Record

We mark each checkpoint reached or stopped. We write the observation that decided the mark, and we keep the page state that supports it.
state: recorded
Reproduce

We repeat the run, and we run a control journey on a fixture site. A stop becomes a finding only when the control passes and the stop repeats.
state: reproduced
Watch

We run the journey each week for one month. Each record carries its date, and each record names what changed.
state: watched

Completion

A completion rule must be written before the run. Without a rule, a result is an opinion.

Completion criterion: You name the page state that means success, for example a basket with the correct total or a free slot with the correct service.
Reached: The agent showed the state that the checkpoint asks for.
Stopped: The agent did not reach the next checkpoint, and the record holds the observation that shows it.
Boundary stop: The agent reached our stop rule, for example the payment step. This is a planned stop, not a failure of your site.
Control result: The same agent configuration ran our fixture journey. A pass shows that the agent worked at that time.
Human review: An operator reads the record and the evidence before you receive it.

Attribution

Attribution is the hard part of this work. We keep the claim inside the evidence.

We say what the run observed

The record names the agent family, the goal, the checkpoint, and the observation. You can repeat the steps.
We name a likely fix owner

Content, front end, platform, or policy. The word likely stays in the record.
We do not blame the site alone

An agent can fail for its own reasons. That is why we use a control journey and more than one attempt.
We do not speak for all agents

A result covers the agent families and the dates in the record. It says nothing about other agents.

Tell us the journey and the completion criterion. We reply with the written scope and the boundary.