Hello Gianluca,

> What causes this anomalous behavior of the PEC tool? Is it a bug in
> the software?

No, it's not a bug. The PEC table has about 1000-2000 entries, the
exact number depends on gear ratio and teeth number.
Let's assume the table has 1000 entries, and the worm wheel has 180
teeth. That means the learning period is about 480 seconds long.
During the learning period the FS2 stores which keys you pressed, and
in the above case every 0.5 seconds a value is written in the PEC
table. If you press a key shorter than 0.5 seconds, it's possible
that it's not stored at all. If you press a little longer than 0.5
seconds, it's possible that it's stored two times. With other words
you have a quantization error in the stored data.
With slower correction speed, the corrections become longer and this
quantization error becomes less evident.
I recommend that you set the "rate 1" parameter smaller than 0.25x.

Michael