That’s very interesting.
I placed mine on a large table, with none of the sensors near the edge, & closed HVAC vents in the room.
From my prior experience with the same sensors on shelves in my family room, I learned that temperatures can vary significantly with sensors fairly close to each other on shelves in an area with human movement & HVAC airflow. And that was nothing compared to the variation when the sensors were further apart in the same room.
You might try your calibration again, in a room with no south-facing windows & no air circulation. Maybe the floor of a garage (assuming the engines in any vehicles there have not been run for at least 2-3 hours). Or maybe a closed bathroom?
ps: After a couple decades of using Celsius in everything (because aviation weather uses Celsius world-wide), I just (two days ago) switched back to (ugh) Fahrenheit, because consumer temperature products these days (including the ecobee) report temperatures in tenths, regardless of Celsius or Fahrenheit. That means I get more resolution (not necessarily accuracy) using Fahrenheit. It’s easier to see small movements in temperature in Fahrenheit (in this case).