Can you fool our AI detector?

Write a line that uses a persuasion trick — a deadline, a crowd, a threat — without the detector naming it. Five rounds, escalating from one trick to three at once, scored out of 100 with a rank at the end.

The detector is the real one, running in your browser. Its published evaluation reports that it missed all four techniques written in paraphrase on held-out text. Model scores describe their tested conditions; they do not establish universal capability.