Let's say a state actor had access to a whole pile of tax returns and wanted to manipulate them to change the conclusions ProPublica would draw. The state actor changes half the data points. Let's say ProPublica was able to check an average of 3 data points on those 50 individuals they reviewed. The data point could have either been manipulates or untouched. I'd model this like a coin flip and say that if ProPublica didn't find the manipulation after checking 150 data points, it's like flipping 150 heads in a row or 2^150 or basically impossible.

Firstly we have no idea how many of the 50 individuals data were public or not. All public data can be discounted since the state actor can just copy it.

Secondly, for the private data, the definition of ‘private’ is unspecified. It really just means not part of a published record. If propublica has access to it, then why couldn’t someone else?

I agree that if there were 150 separate sources with data not disclosed anywhere else, it would be impossible to guess.

But that’s just a made up scenario.

There could be many correct records that are public, and one or two that are private but available to (or even provided through another channel of) the state actor.

As long as the fake records are not part of the public or private data propublica already has, there would be no way to verify them.

This of course assumes that propublica’s list of records itself is kept securely.

Indeed you have proven my point - which is that they don’t give us enough information to do anything except blindly trust them.