The Correctness Tax
A correction is cheap. It happens on the institution's own schedule, costs a few column inches or a database update, and closes on the institution's own terms. None of that tells you anything about what the institution will do for the person who was right, once someone outside it decides to make that person pay for being right.
That's the actual test, and it's a different one from the test everyone reaches for. "Does this place admit when it's wrong" is a low bar. Every functioning institution clears it eventually, because the alternative is worse for the institution: a standing error is a liability that compounds, while a correction is a liability you pay off once and file away. Correcting the record is bookkeeping. It's the institution talking to itself.
The harder case is the one where the institution was never wrong. Someone reports something true, the report holds up, and a third party with power over that person, an employer, a landlord, a government office, a hospital, decides the report was the problem and comes after the reporter for making it. At that point the institution that received the accurate report has a choice that has nothing to do with correcting anything. There's no error to fix. The only question left is whether it will spend anything to defend the person who told it the truth.
Most institutions don't, because the two systems were never built to talk to each other. The correction process exists to protect the institution's own credibility, so it only activates when the institution's own output is wrong. It has no hook for "our output was right, and the source is now being retaliated against for giving it to us." That's not a bug anyone had to design out. It's just never been in scope, because building it in scope is expensive in a way that running a correction isn't.
What it would actually cost is the interesting part. Standing behind a reporter after someone punishes them means picking a side against whoever did the punishing, and that someone usually has leverage the institution wants to keep: continued access, a working relationship, a seat at the next briefing, a source inside the very body doing the retaliating. Protection is a vote against that access. A correction never asks the institution to burn anything. Protection might ask it to burn the exact relationship the reporting depended on in the first place.
The tax recurs
Call the price the punishing party imposes on the reporter the correctness tax. It's not paid once. Every person who might report something true next is watching who covered the last person's bill, and every institution that let the bill go unpaid is teaching its next source the actual terms of the deal: being accurate is fine right up until it's inconvenient for someone with power over you, at which point you're on your own.
That's the part a track record of corrections can't paper over. An institution can run an immaculate correction desk for twenty years and still be the kind of place that goes quiet the moment one of its sources gets fired for telling it the truth. Those are not the same muscle. One is administrative honesty about the institution's own file. The other is a willingness to take a hit on someone else's behalf, with nothing in it for the institution except the fact that it was the right thing to do.
So don't grade an institution on whether it corrects itself. Grade it on the one case where correction isn't even on the table, because it was never wrong, and the only thing left to protect is the person who made sure it didn't have to be.