The first test was flawless. The Funcaptcha API returned a valid: true in 312 milliseconds. The second test, same thing. The third, fourth, fifth—all green. She leaned back, coffee cold, smile warm. Victory.
Desperate, she rewrote the solver. Added random mouse movement curves. Variable solving times. Even simulated a moment of hesitation before dragging the puzzle piece. She fed the new token.
The API paused. A full four seconds. Then: funcaptcha api
She dug into the response headers. A new field appeared: X-Funcaptcha-Debug: seq_4783 | entropy_required | behavioral_anomaly . Behavioral anomaly. That was impossible. Bots don't have behavior—they have code.
No typo. Unhumanly correct. The API wasn't checking if the puzzle was solved correctly anymore. It was checking if the solution was too correct. Too fast. Too precise. Too perfect. The first test was flawless
Lena realized the truth: Funcaptcha had stopped being a test for machines. It had become a test for flaws . It wanted hesitation. It wanted the slight tremor of a real hand. It wanted the wrong answer once in a while.
0.79. Not certainty. Just probability.
But the API had learned.