There was a change to B5-Lite in 1.4.4 where the LNA gain table was adjusted by only 1 dB. There are two sets of calibration values, one for long links, and one for short links where the LNAs are bypassed. It sounds like Chadwick's link may fall into the bypassed case based on distance, but the reported difference is 7 dB which means there is something else happening (e.g. TPC attempting adjustments to improve EVM, dynamic IQ calibration changes, or something else).
There were no changes to B5, B5c or B11 gain tables.
We definitely do not recommend reverting to 1.4.1 because of a potential for memory corruption after a failed upgrade (depending on the previous upgrade path).