Itās a pain, isnāt it?
Using the best available components helps reduce the temperature sensitivity, and getting drivers from a high-quality source (or building them yourself with a consistent method) helps reduce variation between individual units. But even in the best case itās still not as consistent as an e-switch.
Two improvements under discussion at the moment are replacing the OTC with a bigger cap plus a resistor (to make the drain more consistent) or attempting to make the MCU detect power-disconnect events, go into super-low-power mode, and keep running long enough to measure button timings. Both are likely to require hardware changes, and the last one might not even be feasible since a 22uF C1 is probably not big enough to run the MCU for more than a few milliseconds.
For personal use, I just calibrate the OTC individually for each light, using a metronome, with drivers from RMM. That gets them about as close to ideal as currently possible.
On the other end of the spectrum, Mankerās drivers are particularly inconsistent and often pretty far out of spec.