Everybody always says that mechanical release should be safe. And in theory, if you know exactly what you're doing and know how to check it and do so frequently, then it's safe. But if that's the case, why do you see/hear about so many instances of prelease? Because we don't live in a theoretical world, we live in a practical, highly imperfect one. Even the people who know what they're doing and know to check their release still screw up: https://www.instagram.com/reel/CmZYvREjqxO/?igshid=YmMyMTA2M2Y= (see my comment and Stevie's response.) Two ski buddies of mine both had prelease falls within the same week last summer and both are long-time skiers who said they thought they knew what they were doing in setting their release.
We don't tell people not to wear seatbelts just because theoretically cars are safe ("just don't run into anything!"). -- I admit this is not a great analogy, it's too extreme, but just for a sense.
I think to amend my usual saying, rather than "I don't trust mechanical release," it perhaps should be "I don't trust mechanical release in the hands of bumbling humans." (that includes me, and likely roughly 94.7% of you lol).
Yes, a non-mechanical release can be a problem as well, in that it can not release in a bad fall, and that can certainly cause injury, but I'm more scared of, and have heard more bad experiences with, mechanical prelease falls as opposed to "analogue" non-releasing falls.
However... on both sides of this argument the best anyone can offer right now is haphazard anecdotal evidence, which hardly counts as evidence. Wasn't there a poll a while ago asking people about prelease and injury? Or was that just about hard falls in general? It could be informative to do a poll asking how many people have experienced preleases, how many people have experienced falls in analogue bindings where they erroneously didn't release, and for each segment, how many of those instances led to minor or significant injury. Then there'd actually be some data to prove me/you wrong/right. Although even then, lots of confounding variables are possible - course vs. free, how often ski a year, how many years skiing, etc.