# AI agent check

Measures what an automation tool DOES, not what the browser claims to be. It is the companion to
https://scalebrowser.net/check, which reads the browser itself.

## How to use it

Point an agent at this address and let it work through the task on the page. The task is a short
form: three fields and a button. It is deliberately resistant to shortcuts, so a value written
straight into a field with a script does not complete it. The page needs the events, not the
result.

The run happens twice. Two identical pointer paths are a finding in their own right, because a
hand cannot repeat itself that exactly.

## What it reads

Structural facts about the input, not a score for how human someone is:

- text that arrived without a key ever going down
- a click with no pointer approaching it first
- a pointer that jumps instead of being sampled
- a capital letter typed without a shift key
- pressure that stays identical while a button is held
- a key that reports a press with no press-down before it

In the second depth it also places four measured curves next to published human cohorts: the
speed profile of a movement, the rhythm between keystrokes, the pause before the first input, and
reading time against words on screen. Those are shown as numbers beside their cohort, never as a
verdict.

## What it will also say about us

Some of these are hard for any tool that feeds input through a debugging protocol, including ours.
Where our own browser falls short, the page says so. A measurement that flatters the party running
it is worth nothing.

## Related

- https://scalebrowser.net/check: reads the browser rather than the tool
- https://scalebrowser.net/legal/privacy: what is stored, in full
