> FeverPhone estimated patient core body temperatures with an average error of about 0.41 degrees Fahrenheit (0.23 degrees Celsius), which is in the clinically acceptable range of 0.5 C
> An early morning temperature higher than 37.2 °C (99.0 °F) or a late afternoon temperature higher than 37.7 °C (99.9 °F) is normally considered a fever, assuming that the temperature is elevated due to a change in the hypothalamus's setpoint.[14] Lower thresholds are sometimes appropriate for elderly people.[14] The normal daily temperature variation is typically 0.5 °C (0.90 °F), but can be greater among people recovering from a fever.
Eww, that's not how it works. The goal isn't to make the average near the limit, since the average of -100 and 100 means 0 average error! The goal is to make sure the measurement have a distribution that gives some reasonable confidence that it's within the limit.
This is PTT: https://www.spcpress.com/pdf/DJW244.pdf
You can’t just return 100 and -100 randomly, since the actual temp is unknown and you’d need to evenly sample around that.
It appears this is moot, because the quote did not reflect the actual value or method used.
My point is, you can’t really get 0.44F avg error unless it’s predictive. The “std dev” of abs(reality - model) is too high