Let’s say I have a dozen temperature sensors, and I calibrate them, so I know that one reads a fraction higher and another a fraction lower (than each other, or some arbitrary baseline).
I can create a template sensor for each real sensor that has an adjustment for the calibration. So, a dozen template sensors each with its own input temp sensor and its own hard-coded calibration adjustment.
The advantages to the compensation integration is it corrects over the entire range of readings, not just at 70F or whatever, and it will also adjust non-linear if you give it 3 or more points of data.
For example, I tried the same experiment with DHT11 and DHT12. An offset didn’t work so I tried an offset and slope. I built a temperature controlled oven and I was able to hold the temperature to +/- 0.1°C (according to my probe). To no avail.
My advice is to look at your OCD vs wallet ratio and proceed from there.