It would be interesting if I'm seeing a combination of the onboard recalibration coupled with a temperature 'switch'. The one thing I think we're not sure about is the actual maximum recalibration capabilities of the sensor. If Sensirion is only claiming a maximum of 3%, do we really know that the recalibration can go much higher with age? Wouldn't Sensirion mention that in the data sheets?
Sensirion only mentions that prolonged exposure to humidity above 80% may result in degradation of overall sensor performance. I doubt every sensor takes the same trajectory since each is subjected to different environments and potential contamination. But I think that statement by the manufacturer provides plenty of space for the imagination without making their lawyers sweat.
As I said, I only intuited the aging process was to blame because I noticed the problem seemed to worsen with time. But it's entirely possible that the length of time it was out of calibration was purely a function of varying weather patterns and not age. At a certain point I began documenting how long that error period lasted. The minimum was ~6 hours and the maximum was ~48 hours.
This is how it would work. Moist air would move in. It would stay in place for half a day or longer. Then as the humidity would start dropping, sensor B would stay pinned in the 90s rather than dropping into the 80s. If the actual humidity stayed in the 80s for a day, then the sensor would run wet for a day. But if it dropped into the 70s then it was only a matter of time until it returned to accuracy. The drier the air, the shorter the interval until recalibration occurred. Assuming a 75% humidity air mass (measured at 2 meters) hovers indefinitely afterward, my memory tells me it would take, on average, about 12 hours for the sensor to recover. YMMV.
As an aside, since the switch and/or recalibration, I've been tracking amazingly close on every parameter to the airports & the Drop. It's like a different unit. It's too bad there isn't an 'on demand' recalibration that the user could initiate.
Yes, when my sensor B is on, it's dead on. It's a night and day difference.