> In 1851, the German physician Carl Reinhold August Wunderlich obtained millions of axillary temperatures from 25,000 patients in Leipzig, thereby establishing the standard for normal human body temperature of 37°C or 98.6 °F (range: 36.2–37.5°C [97.2- 99.5 °F]) (Mackowiak, 1997; Wunderlich and Sequin, 1871).

The adjective "axillary" means "relating to the armpit", so armpit vs. oral doesn't explain it.

Presumably many patients have abnormal temperatures, so to figure out the normal temperature from patient data would require some kind of filtering.

A couple ways to do that come to mind. One would be to only use patients who are there for things like an annual checkup.

Another way would be to include everyone and then do a graph of x = temperature, y = number of patients with that temperature. I'd expect there to be a peak centered at normal temperature, and another peak at whatever the average fever temperature is.

If whatever method he used was not able to exclude from the "normal" group people who are just starting to get sick and are at the stage where their temperature has started to rise but they are not aware that they are getting sick, he'd get a slightly too high result.

36.6°C is the more modern number:

> A compilation of 27 modern studies, however (Sund-Levander et al., 2002), reported mean temperature to be uniformly lower than Wunderlich’s estimate. Recently, an analysis of more than 35,000 British patients with almost 250,000 temperature measurements, found mean oral temperature to be 36.6°C, confirming this lower value (Obermeyer et al., 2017).