I don't think it has anything to do with what we know "now". It's just paying attention to the fact that the signal photon hitting the screen causes a collapse that affects the state of the idler photon. Which then explains the data via the collapsed state depending on the position of the hit, and one of the possible idler measurements being in a basis perpendicular to those variations. All quantum interpretations give the right answer for this experiment, and very few of them invoke retrocausation, therefore the experiment clearly doesn't require retrocausation.

I don't even think the delayed choice eraser is a "quantum" paradox. It involves quantum particles, but they're really just there for flair. They're not crucial. You can apply the same confusion to a classical experiment. Set up some basic correlation between A and B, with A revealed first and then a choice to reveal B or an unrelated C. Then describe the situation so badly that it sounds like choosing to measure B vs C is changing the probability distribution of A backwards in time (since if you condition on B you'll see the correlation vs A, but conditioning on C shows no correlation).

To be fair, almost everything in quantum is poorly named. That's how they attract funding.

Except for "quantum supremacy", which is the best name in the entire multiverse.