Worries about 'Top down' or 'Full non-indexical conditioning' The essential worry is (I think) straightforward, but for precision let me define it carefully in a simple case. Consider two theories A and B, both of which predict many 'universes' (i.e. a given observer can only ever observe one universe) across which two properties, say 'color' (C) and 'spin' (S), can each take two values (C=blue, red), (S=left, right). Suppose we can describe a suitable measure over these universes to define probabilities P_A(C), P_A(S), P_B(C), and P_B(S), where, e.g., P_A(C) is the probability that a randomly chosen universe in theory A takes the given value of C. Now suppose: P_A(C,S)=P_A(C)P_A(S) and also P_B(C,S)=P_B(C)P_B(S) [just for simplicity] P_A(blue)=0.9999999999=1-P_A(red) P_B(blue)=0.5=P_B(red) P_A(right)=0.99999 P_B(right)=0.99999 Finally consider two cosmologist, Bob and Alice, who are evaluating the theories A and B. As of monday, when Bob starts to work on the problem, neither the color nor spin of the universe has been measured, in the precise sense that no data in his experience (and thus conditioned on using FNC) is correlated with C or S. Then Bob asserts that theory A predicts the universe is blue with very high confidence, whereas theory B does not say much either way. (Note that I don't see any difference here in considering 'Bob' or the set of all observers with the same knowledge set as Bob, i.e. non-indexical conditioning). On tuesday, a skilled group of Canadian cosmologists measures that C=red. Bob does not read his email that day, and knows nothing of this. But that afternoon, Alice (who has heard of the measurement result) starts her analysis. C=red is part of what must be conditioned on, so she cannot predict C. But she can predict S in both theories, concluding that theory A predicts S=right with strong confidence, as does theory B. On wednesday the two physicists converse, as there is news of an upcoming measurement of S. Alice tells bob about the measurement of C=red. "Terrific", says Bob, "theory A is ruled out. Let's now go and test whether theory B holds up, since it did OK with the C measurement." But Alice obsjects "hold on. Theory A is fine, we just have to condition on C=red. But it can still predict S." Bob and Alice are at an impasse, but agree to disagree. They can agree, however, that both theories predicts S=right with strong confidence. Finally S=left is measured. Now Bob and Alice find that they are unable to think what to do. They are inclined to rule out both theories, but of course before the measurement, Alice, at least, had already accepted theory A, which Bob had ruled out at even higher confidence using C=red. Now suppose Fred returns from vacation. He has not heard of either C=red or S=left. He accepts both as part of the conditioning in both theories A and B, and makes predictions for the 'Odor' (O) of the universe... etc. As should be clear, I am very perplexed because: a) Whether a theory is ruled our or not appears to depend upon what day it is evaluated, b) even after all measurements are in place, scientists are unable to agree on whether or not a given theory is ruled out. c) in fact it appears that no such theory (making probabalistic predictions) can ever be ruled out, since I can always find a suitably ignorant scientists (or set of them) who can include all of the relevant measurements -- that would seem to rule out the theory -- in their 'full conditioning'. Of course this could all be phrased in different more complex ways (if, say, P_A(C,S) did not factorize); but I think this simple example gets to the core of it. I don't think this is some nit-picky or contrived worry: this sort of conditioning is *exactly* what, for example, Hawking and Hertog, use when they take some highly improbable situation (e.g. that the inflaton field started high up on the potential) as conditionalization, then try to predict other things such as the amplitude of initial density perturbations -- if they get it wrong, what is to stop someone next week from taking the 'wrong' value as a condition in their approach (as long as its probability is not strictly zero)?