Pretty sure they can tell the method used when disengaging fsd/ap. So they would know if it was manually turned off or if the system lost enough info and shut it down. They should be able to tell within a few seconds if accuracy the order of events. I can't imagine a scenario that wouldn't be blatantly obvious where the tesla was able to determine an accident was imminent and shut off fsd/ap wroth enough time to "blame it on the driver". What might be possible is that the logs show fsd shut off like a millisecond before impact/event and then someone merely reported that fsd was not engaged at the time of the accident. Technically true and tesla lawyers might fight like hell to maintain that theory, but if an independent source is able to review the logs, I don't see that being a possibility.

They supposedly also have a threshold, like ten seconds - if FSD cuts out less than that threshold before the accident, it’s still FSD’s fault

That would require their self driving algorithm to actually detect an accident. I doubt it's capable of doing so consistently.

On a related note, getting unstuck from something like train tracks is a pretty significant hurdles. The only real way is to back up IF turning onto the tracks wasn't a drop down of the same depth as the rails. Someone who is caught off guard isn't going to be able to turn a passenger car off the tracks because the rails are tall and getting an angle with the wheels to get over them isn't really available.

So while in a perfect world the driver would have slammed on the brakes immediately before it got onto the tracks, getting even the front wheels onto the tracks because they weren't fast enough may have been impossible to recover from and going forward might have been their best bet. Depends on how the track crossing is built.

It’s more about when they don’t release it/only selectively say things that make them look good and staying silent when they look bad.