Reliable stabilization is still number one for me because it reduces the amount of footage I have to fix or discard.
If storage and proxy generation are manageable, I would keep the higher-resolution source for shots where reframing matters.
Also wait for the platform's highest-quality processing before comparing. Early playback can make a good upload look much worse.
Agreed. A stable map is usually calmer, and you can use a separate heading arrow if you still need orientation.
That seems like a practical rule. I do not want to use a complex workflow just because it exists.
If you want custom typography and placement, a data-driven desktop overlay is cleaner than trying to crop a pre-rendered dashboard.
I would protect the important subject for both crops when possible, but still decide which version is primary so the shot has intent.
Treat each reframing move like a real camera move: it should reveal something, follow action or solve a composition problem.
Rough cut first unless a shot is so flat that you cannot judge it. There is no benefit in grading footage that will be deleted.
I only use the app-first route when I need a fast social clip. For a longer project I want the original telemetry and styling freedom.