Let me expand further on the postdiction vs. prediction issue. This is actually something that may be relevant for an experiment like mine, when we’re trying to measure a quantity at a precision that no one has done before.
Suppose a detector with method A finds a statistically significant signal of new physics (say a phenomena called X). The scientists make a convincing case that they understand the physics of their detector well (including accounting for sources of systematic errors), even if their detector has never been operated at this sensitivity before. However, because this is a very radical result if true, not everyone is fully convinced. Perhaps there are some unknown systematic errors that detector A overlooked.
A few months later, another detector with a different method B also finds a signal of X that matches A’s findings. Similar with A, detector B uses well-understood physics, even if the detector has never operated at this sensitivity before.
A year later, the same phenomenon happens with detector C (with yet another method), which also detects signal X.
What is the “proper” epistemic attitude towards X, A, B, and C? First, I think most would agree that it is rational to believe that X is real, as it has been detected by three independent methods.
Now, if we only accept prediction and not postdiction (as Sabine argues), then it seems that when detector B detected X, that only increased our confidence in the correctness of detector A’s result. Since detector B only “postdicted” X, we have no idea if B is seeing anything real. It is only when detector C detects X that we become confident of detector B. And even then, we would not be sure in the confidence of detector C, only A and B.
Furthermore, after seeing the successful prediction of A when we detected the result of B, since we don’t know for sure if X is real, one might even argue that even this successful prediction doesn’t have much epistemic worth either. The same argument would apply to C, D, E, and any further detectors that see X - their epistemic worth would all collapse like a pile of dominos. This would seem to prohibit the possibility of science and inductive reasoning altogether.
I hope that the above example suffices to demonstrate the untenability of this position. The proper epistemic attitude is that after seeing both A and B detect X and agreeing with each other, that increases our confidence in both method A and B of detecting X. When C also detects X, that increases our confidence in C as well. Whether A, B, and C’s agreement has anything to do with say, fudging their data has to be judged on their own merits regardless of the temporal order of the detections.