The authors added "sensitivity analyses" excluding Winter break days from their calculation of IRRs with 3- and 7-day lag times between exposure and reported case. This is critical, since school mask mandates don't effect exposure when school is closed. No justification is given for the choice of these lag times, and they are both lower than those used in other studies.

I reproduced the results from this paper (IRR 0.99 [0.92, 1.06] in my attempt to replicate the primary analysis). For the sensitivity analyses with various lag times used to censor Winter break, I also reproduced the result for a 3-day lag (0.98 [0.91, 1.06]) and I also found a similar results for a 7-day lag, as reported (0.96 [0.89, 1.03]).

However, the sum of days from exposure to onset to testing to reporting a test result on the dashboard is unlikely to be as few as 7. Other studies I've found all use longer lag times. Longer lag times give IRRs below one, with one excluded from the 95% confidence interval for lag times between 9 and 14 days:

8 0.93 [0.86, 1.00]
9 0.90 [0.84, 0.98]
10 0.87 [0.81, 0.95]
11 0.86 [0.79, 0.93]
12 0.85 [0.78, 0.92]
13 0.84 [0.77, 0.91]
14 0.86 [0.79, 0.93]