It’s not about anyone having an “anti-” axe. You don’t need an axe to grind to be a bad statistician. You just have to be too lazy to record diverse response properly, and that’s enough to end up forcing a very conservative result.

It seems to me that it’d be simple enough to electronically record both the inputs as they were submitted (minus spoiled fields) and the sanitised probabilistic interpretation of the input, and make both available. Presumably if they’re being systematic about it, the data is being input as it is on the forms and then sanitised by computer rather than by some data input worker’s preference, and I’m pretty sure the storage space cost increase of holding both datasets would be negligible (compared with budget of project).