Part of the reason was to avoid making the curve artificially easy, then practising a touch that does not transfer well to another instrument or an acoustic piano. I used the Yamaha response as a practical target because I expect its factory tuning to be a safer starting point than simply optimizing for the easiest feel.

Here is the final curve I am using:

Velocity = [0, 1, 6, 7, 15, 30, 60, 90, 120, 127; 0, 1, 3, 4, 14, 48, 91, 117, 127, 127]

I measured attack level from matched MIDI/audio renders across C2, C3, C4 and C5, fitted each register separately, then compared the pooled response curves. I also ran a dedicated low- and high-velocity sweep before simplifying the curve to the points above.

With Pianoteq SK-EX at 54 dB Dynamics and 280VC at 50 dB, this curve tracks the CLP CFX and Bösendorfer internal responses closely through the main playable range. The two small low-end points, 6→3 and 7→4, are there because Pianoteq has a large discrete jump between its velocity 3 and 4 layers.

The plot below shows the result. On the left, the remapped Pianoteq curves closely follow the two internal Yamaha curves. The right panel is only a reminder that bass/treble voicing is a separate issue from the global velocity curve.