"User engagement" is not a single, well-defined, easily measurable thing.
For one thing, it's easy for A/B tests to show something that looks good in the short term, but has longer-term negative effects. If your goal is to iterate quickly and stop your test as soon as you think it shows a positive result, you'll never see that.
But more to the point, there's no guarantee that the "user engagement" metrics Twitter cares about are actually a proxy for what users really want. If it takes me more time and more clicks to find the information I want -- enough to be annoying, but not quite enough to drive me away from the site -- that probably gets counted as "engagement".
I think something like this circular logic is actually useful, and in some sense true. It might better phrased as "companies which usually make bad financial decisions die, so the decisions of a company which hasn't died are probably financially beneficial." Further qualifiers could be added to account for longevity of a company (we might put more faith in a company which has spent longer not dying), earning reports, and the fact that a good decision isn't necessarily beneficial (just probabilistically so given available knowledge), but the basic reasoning seems valid.
You could apply this reasoning to literally anything, because every organisation / nation state / what have you has at some point and in some form considered the consequences of their actions. History shows that internal evaluation is an imperfect science. I suspect you know this and only reserve this extremely circular reasoning for cases which you already agree with. It would be easier to admit you've already formed an opinion and don't care to change it under any circumstance, rather than weakly appeal to Twitter's own authority to justify their actions.
It is not just an appeal to Twitter’s authority. It’s an informed opinion based off working in a similar field and conversing with others in this field. So unless you can come up with some concrete data supporting your points, instead of just attacking mine, you are unlikely to change my opinion.
For one thing, it's easy for A/B tests to show something that looks good in the short term, but has longer-term negative effects. If your goal is to iterate quickly and stop your test as soon as you think it shows a positive result, you'll never see that.
But more to the point, there's no guarantee that the "user engagement" metrics Twitter cares about are actually a proxy for what users really want. If it takes me more time and more clicks to find the information I want -- enough to be annoying, but not quite enough to drive me away from the site -- that probably gets counted as "engagement".