* First point is that it is too sensitive to visible light around. But even if I switch the light off...

* second is that it is only sensitive to a huge model of the eye - a big lens and a palm over it. I do get a reliable peak out of that system, but it is a small change in a quite big background signal. As for the eye behind the eyepiece - simply nothing.

So, back to thinking.

Update:

channel1 (yellow) - speaker drive voltage, which is more-or-less proportional to motion speed (half of the whole period is displayed).

channel2 (cyan) - signal from phototransistor, AC coupled. There should be a peak somewhere around the center, but there is no.