| Allowed origins |
The origins a run’s browser may open: the product’s own, and the one setup was redirected to, if any. Everything else is blocked. |
| Cap |
Where a run stops spending on model calls: $0.10 to $50, $2 by default. The last call can take it slightly past. |
| Defect |
A finding that says the product behaved wrongly. It carries steps a stranger can follow, and it is replayed. |
| Estimate |
The likely cost of a run, shown on the Start panel before you start it. |
| Finding |
Something a person reports while using the product: a defect or friction, with a severity and the goal they were pursuing. |
| Focus |
Optional words given to setup to steer the people and goals it proposes. |
| Friction |
A finding that says the product slowed a person down or was unclear. It is not replayed. |
| Goal |
An outcome a person wants to get done in the product. Every person pursues every goal of the plan. |
| Judge |
The model that compares a defect’s claim with what its replay saw and gives a verdict. |
| Model key |
The workspace’s API key from a model provider. Runs are billed to it. |
| Person |
A simulated user: a first name and a brief saying what they are like. The code and the local runner call them personas. |
| Plan |
A project’s people, goals and test accounts — what a run will do. |
| Private beta |
Hosted runs are open only to email addresses on the beta list. |
| Project |
One product under test: its address, name, description and plan. |
| Replay, blind replay |
A fresh agent following a defect’s steps in a new browser, told nothing about what went wrong. |
| Run |
One execution of the plan, with a model, a cap and a number in the workspace. |
| Runner |
The program that carries out a run: the browser, the people, the replays and the judge. Its source is public, and it can run on your own machine. |
| Setup |
Reading the product’s page and proposing a plan. |
| Severity |
How much a finding matters, as the person judged it: low, medium or high. |
| Test account |
A username and password on your product that a person signs in with. |
| Verdict |
The judge’s answer for a defect: confirmed, refuted or inconclusive. A defect whose judge gave no answer could not be judged, and can be judged again after the run if its cap has room; one that never reached the judge is not judged. |
| Workspace |
Where your projects, runs and model key live, and who can use them. |