I've taken a look at your TapTation PT2399 application note and see that you say the circuit design has an average accuracy of 5ms using a MCP41100 digipot.
This seems impossible given that the 100K ohm MCP41100 digipot has a typical resistance tolerance of +/- 30%.
I understand how you arrived at the 5ms value - dividing the 257 step resolution by the device's target resistance, but unfortunately digipots are extremely variable in their actual resistance and can be expected to have a nearly 30% error.
With the PT2399, a 30% error can amount to an enormous error in accuracy.
For example, at 27,600 ohms, the PT2399 datasheet calls for a 342ms delay time.
70% of that value is 19,320 ohms.
Even, very conservatively, at the next value down on the datasheet (21300 ohms), the stated delay time is 273 ms, with a resulting error of 69ms.
Of course this error is even more dramatic when you get up around 100K ohms (> 1 second of delay time) where you can expect an error of over 150ms!
Have you come up with any way to calibrate the output resistance or otherwise account for this typically huge tolerance error?