New measurements tonight, this is getting serious...
I ran some short captures in the mV range, 200 samples or so, I saw a few glitches but interestingly right at samples 61, 121 and 181 at V1.26, even more intresting, in V1.00 there are smaller glitches one or two samples later. With the open leads, right at the start is easy to see the error introduced by me taking the finger out of the meter after pressing mem, in both versions, V1.00 smaller and slower as expected.
Now at the freq plots, V1.26 shows a flat noise response in both tests, while V1.00 shows a low pass response.
If we compare the two versions, doing V1.00/V1.26 we see a similar frequency behavior with the leads open or closed.
Then I decided to make a longer sample, to see if I can catch the pole, left it running for about 11 minutes and cropped some samples at each end, so we can see the response down to a few mHz, for this I think I over did it leaving long leads open far away and ended with some strange noise with a clear frequency at around 1Hz, and the overall noise of the V1.26 was much higher so the ref plot doesn't start at 0dB. Also, I tweaked a freq plot a bit as I don't really know the sampling rate, but with the long test I estimated a slower sampling rate than the 200ms specified in the manual, that's why the response looks a bit shifted.
Even at a few mHz the response doesn't flattens out, going down by 10dB/dec up to a few 100mHz where it steeps down even further and returns back up, I guess due to the measurement being so noisy. I don't have a more controlled LF noise source to generate for this measurements but should be easy enough for the next test using a zener and a divider to work in the mV range, I guess I'd need to run it even longer to catch the pole. Also, with a more controlled noise source I'd like to run some intercorrelation calculations between the samples to put a number in the glitches being the same between both versions.
To finish up, the glitches and button press sensitivity is present in both versions, V1.00 clearly has some filtering going on, checking the schematic and datasheet of the HY3131 there are many analog selectable LPF available at the chipset, first C21 with the internal resistors and then C16 to C26 and likely some of the surrounding resistors, plus a configurable digital comb filter after the ADC, to then go the the µC where they could be doing more processing of the signal... We need an open source code right now!!!
JS