An AI agent tested a practice form and wrote the bug report you will see on the right. You get the same
form with the same planted bugs. Test it, then mark each item of the report: a real bug as described, a
real bug with the wrong steps or cause, not a bug, or the same bug as another item.
The report is the agent's own text; only its item numbers were dropped. The agent had about 40
browser actions and the instruction to test the form against its requirements.
The key was built by reproducing every item on this same layout of bugs.
One submission per session. The result gives the right mark for every item and the reason.
Use made-up names, addresses and passwords only. Accounts keep only salted hashes and are deleted
after a day.
Starting creates a session. The site keeps your marks for the check and nothing you type
into the form. No cookies, no recording of clicks or keys.
Preview: the form loads when you start
The form the agent tested
The same page the agent saw: the form or the API console, its requirements and its
release notes.