Victor's measurements provide a filter for which Alice+Bob photons to look at.

When Victor chooses entanglement, that filter is applied, and Alice and Bob find correlation in their measurements (as they now know which photons have correlation, and they ignore the ones that don't). When Victor does not choose entanglement, a different filter is applied, and Alice and Bob find that those particular photons which match Victor's detections are not correlated.

It is as if you have seemingly random data, and then Victor hands you an extra piece of decoding information, "Use photons 1, 3, 7, 8, 11, ...". When you include only those in the set, you find the results expected.

Yes, if the detector did not transfer it's information to the macroscopic domain fast enough, then it is possible for an occasional event to be "undone" in QM. But, once the detector has registered the result in a macroscopic way, there is no going back, and no changing the past.

This experiment could have sent Victor's signal to the Moon and back, and the results would have been the same.

The past is NOT altered by this experiment. What Victor's measurements do, is change which sets of Alice and Bob's measurements to include in the experiment. Only detections by both Victor and Alice/Bob are going to be included in the experiment. If Victor does a delayed choice, he doesn't change Alice and Bob's results. What he does is change which photons he detects, which acts as a filter to which photons from Alice and Bob to include in the experiment. It is this filtering effect which preserves the Heisenberg Uncertainty Principle.

As with all delayed choice experiments, the past is not being changed by the future. The article would have you believe that Alice and Bob, without any information from the future Victor, can determine if their measurements are correlated or not. This is not true. If it were, then the future can affect the past. For these experiments, photons from the various detectors are checked for time correlation, to identify that the measurements occur for the same pair of entangled photons. Only measurements by all three participants, with the appropriate time delay, are declared to be "a measurement".

Lots of photons are measured by Alice and Bob, that are not measured by Victor. Why? It is Victor's "delayed choice" of whether or not to entangle the photon that determines which events Victor measures. He will not measure anything that could give him more information about the entangled system than is allowed by the Heisenberg Uncertainty Principle.

So, Victor's choice changes which photons, as measured by Alice and Bob, are going to be included in the test.

The future does not change the past. But, information from the future can change what information we can extract from the past.