The following video is a good summary of the struggle between local and non-local physics:
Note that the initial paradox of contradicting observers (explained at 1:40 in the video) is based on local perception. The second observer sees the change in gravity before the sun disappears, which shouldn't be possible.
But, if you take non-local perception (remote viewing) into account, the paradox is resolved. A remote viewer would still see the sun disappear first. Modern science ignores this possibility, as non-local perception is rejected.
Unfortunately, taking non-local perception into account leads to another problem:
If the many worlds explanation is true, then non-local perception would mean that there is no way of telling the difference between the worlds without a measurement that proves what world we are in.
Let's take 3I/Atlas as an example. Farsight saw a distinct outcome (3I/Atlas will be attacked). But that outcome is not attached to a specific world within the many worlds continuum. It's just one of many possible outcomes. So, how can we know that this outcome will happen in our world?
Short answer: We can't.
When formulating the target, being specific about what you want to know limits the scope of possible worlds. For example, if you look for 3I/Atlas when closest to Earth, all the worlds where there's no 3I/Atlas, or where it has been destroyed before that point, are out of scope. But the problem is: When looking at the future, you can't be specific enough to limit the target to one single world, as the tasking would have to take all remaining possibilities into account.
Fortunately, this is not true when looking at the past. A tasking that includes formulations like "in our timeline" would limit the target to our world, because what happened in that world already has been decided.
But there's still a "little" problem: What does the tasker mean when saying "our timeline"? As consciousness can switch between local and non-local perception, I suspect that there has to be an agreement between tasker and viewer about what world to look at.
I think it works most of the time because that agreement is made implicitly.
But it doesn't solve the problem of the many worlds when it comes to future targets. The only implicit information the viewer gets is that the tasker wants to look at the future that will most likely happen. But "most likely" still can mean anything. It's an extrapolation based on the present moment.
Courtney already spoke about this in the Spotlight titled "Science of Forecasting Future Events".
Conclusion: We cannot predict the future, we can only approximate. It's not only the intention of the tasker that matters, but also the knowledge about how to set a target that limits the possible worlds. The more the tasker knows about the past and present, the better the approximation.