I repeated the test as you indicated. Also, in order to be more reproducible I reset the default settings. The setup is a single probe in channel A connected to the probe compensation signal. No external electronics are used. (This was not how I discovered the problem, I was just trying to show an easily reproducible scenario). To setup the picoscope, I did the following:
1) Setup hardware with 1 probe in channel A connected to compensation signal, ground clip connected to trigger ground. Probe in X10 mode.
2) Load PicoScope 6.
3) File-->Startup Settings-->Reset Startup Settings
4) Shutdown PicoScope
5) Load PicoScope again (to get startup settings)
6) File-->Startup Settings-->Load Startup Settings
7) Select 500 us/div, x1, 10 kS, Channel A at Auto, AC. Channel B,C,D are off.
Trigger on "Auto", A, Rising, 0V, 50%
9) Single measurement on channel A of "Cycle Time", "Whole Trace". All other parameters for measurement default.
10) Take readings while varying the number of requested samples.
My results at different number of samples while measuring the compensation signal on AC mode:
REQ ACTUAL RESULTS(Value, Min, Max, Average)
10k 9926 1 ms, 999.9 us, 1 ms, 1000 us
1M 625,005 1 ms, 750 us, 1 ms, 1 ms (Min varies between 750 us, 873 us and 1 ms)
2M 1,250,005 1 ms, 1000 us, 1 ms, 1 ms
3M 2,500,005 ((blank)), ---, ---, ---, --- (Readings go blank, or indicate "---")
5M 5,000,005 ((blank)), ---, ---, ---, --- (Readings go blank, or indicate "---")
If you put the trigger in single mode, so you see each trace, and request 1 MS, then the cycle time indicates 750 us some of the time. The waveform looks perfect on the display, the measurement calculation is sometimes wrong. The value is not wrong all the time. It gets it wrong about 1 of 5 tries. The value when incorrect is either 750 us or 873 us, so it is not a small error, it is a gross error. Also, when the cycle time reads 873 us, if you add a "High Pulse Width" measurement, it reads 406.5 us (normal value is 508.1 us).
So, in conclusion, it appears that it works differently in AC mode, but still gives bad readings. It still fails completely at requested samples of 3 Meg and higher, and only occasionally at 1 MS request. Regardless, it means that if I use the measurement function, I have to verify any values with a manual measurement before trusting Picoscope's value.