Skin temperature: absolute or a deviation?
One of these numbers is a temperature. The other is a difference between two temperatures. They share a unit and nothing else.
Who reports which
| Source | Field | What it is | Typical value |
|---|---|---|---|
| WHOOP | skin_temp_celsius | Absolute skin temperature | ~33 °C |
| Oura | temperature_deviation | Deviation from your own baseline | −0.5 to +1.0 |
| Fitbit | nightlyRelative | Deviation, likewise | −0.5 to +1.0 |
What goes wrong if you merge them
Put both in one series and the axis has to span −1 to 34. On that axis a deviation of −0.3 sits next to zero, which on an absolute scale means freezing; and 33 reads as a deviation of +33 °C, which is not survivable. A chart will render it happily.
A daily average across the two is worse, because it produces a number in the middle of the range that is not obviously absurd — around 16 — and nothing about it announces that it is meaningless.
Why the deviation is the more useful measurement
Absolute skin temperature at the wrist is dominated by things that are not about you: room temperature, whether you are under a duvet, whether your arm was outside it. The signal people actually want — a shift that might precede illness or track a cycle — is a small change against your own norm, and it is buried under those effects.
Reporting the deviation is the vendor doing that subtraction for you against a baseline built from your own history. It also means the number is only meaningful relative to that baseline: a +0.8 from Oura and a +0.8 from Fitbit are not the same claim, because the baselines were computed differently over different windows.
Converting between them
From absolute to deviation you would need that person's baseline, computed the way that vendor computes it. You do not have it. From deviation to absolute you need the same thing.
So this is a case where the conversion is conceptually simple — subtract a baseline — and still not available, because the baseline is a vendor-specific construct that is not in the export. It is a useful contrast with the SDNN and RMSSD case, where no conversion exists even in principle.
What to do
- Keep them as separate quantities. They map to different measurement types in our model for exactly this reason.
- Compare deviations against deviations from the same device.
- Treat an absolute skin temperature as environmental data with a physiological component, not the other way round.