You may be right the the cause is related to the turbulent air, but realize that the data points in the spreadsheet are 25 seconds apart. It is hard to imagine a vortex that lasts 25 seconds or longer and makes the average 2 deg C hotter or colder than the previous 25 sec.
Maybe the problem is the way this spreadsheet was derived from the raw HP data logger measurements. The logger probably sampled much more frequently than once per 25 seconds. I was assuming that each 25 second spreadsheet row is an average of the last 25 seconds, but maybe the spreadsheet just takes one raw measurements from each 25 second period. That may show the large variation we see, while averaging them (integrating over the entire period) would smooth everything out.
I agree on the smoothing effect due to averaging. JR said (1) that all channels were sampled 20,000 times a second and averaged every 5 seconds. Maybe it is true, maybe not. Who knows.
Anyway, air calorimeters are subject to large temperature excursions and inhomogeneities also due to temperature stratifications caused by buoyancy effects. A similar problem was just described yesterday in a paper presented at ICCF22 in Assisi (2).
Surely the 25 s time scale is much shorter of any time scale due to anything happening inside the reactor.