Yep - 4.7 is in its beta releases, and has been branched out, so new developments (in/over the master branch) register as 4.8. We are still actively back-porting changes to the 4.7 branch though, once they’ve been merged in and determined as relevant/important for version 4.7.
That’s not fully correct - ArduSub automatically re-calibrates its barometer offset whenever it is lower pressure than the previous calibration, under the assumption that it is out of the water / at the surface.
If the sensor is somehow outputting continually decreasing pressure values then the calibration offset will drift lower over time - the real question then is where are those values coming from?
This is a point I am wondering about as well, since it seems to imply there are measurement spikes that get more severe over time. A DataFlash (.bin) log from the autopilot would allow us to see the sensor values at a higher frequency, which would help to get a better sense of the nature and extent of the values. Ideally that would be from an extended duration of testing, so we can also see whether there’s any ongoing drift in the values, and whether the noise level is consistent.
I would note that the sensor is designed for use in water, so operating it in air may have unexpected / poorly defined behaviour. Could you try repeating your test with the sensor head in water (even if it’s just sitting in a cup), and seeing whether the reading still fluctuates in a similar way?