See where a simulated visitor completes, pauses, or stops.

Set a task and persona, then run the journey in a browser. A separate grader checks recorded actions and links friction to screenshots.

Simulated browser journeys

  1. One defined task per path
  2. Separate grader checks observations
  3. Six persona styles, from newcomer to screen-reader user

View the workflow

A persona's completion claim is not the grade.

The grader checks recorded observations. A simulated task result does not represent how real visitors behave.

  • Grader isolation

    Separate grade

    Independent check

    The grader never sees the persona's own claim.

  • External copy mode

    Exact domain allow-list

    Read-only

    Non-GET requests, form submission, and credential typing are blocked for an external target.

Measure one path before running a round.

A mission names the target, the persona's constraints, and the actions that count as done.

  • Choose a mission

    Define the target and select a persona card with context, communication style, constraints, and a definition of done.

  • Measure one path

    An independent DONE grade is required before a path is measured; a full round is capped at the measured step count plus five.

  • Run the browser

    Personas use accessibility snapshots and take one browser action per turn.

  • Review the report

    A separate grader checks the recorded actions and returns a grade, friction notes, and step screenshots.

A report tied to the steps the browser took.

Each round returns a grade, friction notes and step screenshots.

  • Round report

    See grade, friction, and evidence.

    Each round can return JSON and standalone HTML with a grade, friction notes, and screenshots for the steps.

  • Measured path

    Keep the proposed path reviewable.

    After a grader-confirmed probe, the tool can propose a measured-path patch for a person to review.

Choose a local or staging target.

A writable run requires a signed-in Codex CLI on the machine running the test. External public sites use exact allow-listed copy mode. Review screenshots before sharing because images can contain typed contact details.

  • Writable run

    Prepare a loopback target.

    Writable tests are limited to loopback targets in the documented setup.

  • External copy mode

    Allow-list the exact domain.

    Cross-origin navigation is fenced; the site's own styles and scripts still load.

Choose one journey to test.

Prepare a local or staging copy and choose one job for the browser to attempt. This preview does not launch a browser test.

  1. Name the task

    Choose one job, such as finding a guide or contacting the business.

  2. Set the constraints

    Choose the persona and define what counts as completion.

  3. Save the report

    Keep the run output beside the job description for review.