So .. our experiment with downgrading your imps had the result that the reported temperature measurements dropped again, which seems to support the connection with firmware version, and certainly wasn't the result I expected.
However, a colleague from the imp team has a theory why you're seeing these results:
This project causes the imp to regularly wake up, connect to our servers and make a measurement via its ADC and the thermistor. But in a battery powered device, the reference voltage against which the ADC is calibrated may fluctuate a little during the time the imp is waking up and reconnecting, due to background activity particularly by the WiFi radio. This nature of this fluctuation is probably quite repeatable for a given firmware version, but might vary a little between firmware versions, which could explain the offsets you're seeing between 34.11 and 36.7.
The way to test (and potentially fix) this in Squirrel is to introduce a delay between switching on WiFi and making the measurement. See the documentation for
An easier check could be simply to increase the amount of data points in your measurement e.g. via
myThermistor <- thermistor(temp_sns, b_therm, t0_therm, 100, false); // Increase points from 10 to 100.
If these changes affect your readings, without a firmware upgrade, we'll know we're on the right track here.