If you are testing A vs notA, and you exclude some scenarios in which A is true (and dont exclude anything else), that is (by definition) evidence for notA and against A. (at least by a bayesian definition)

Now, might be that the priors for A were very large, and A is still the most likely hypothesis. But the evidence just received reduced those priors

-----------------

(I know that the case in point does not fit the rather strict requirements of the first paragraph. But I think the affirmation "the hypothesis that reducing the workload improves the life of the worker, while still very likely, is now a bit less likely" is true in this case.)

(The phrase in " " sounds odd to me. If I knew numbers, it would be much better to say P(A) was 95% and now is 90%)

"A mild change in X did nothing, so you should assume a major change in X will do nothing?"

Wrong direction. We are not entitled to take a result that a minor change had no effect on the target variable and treat that as evidence that a major change must do the thing we expect it to do. We must accept this as evidence that in fact the major version of our change will, at the very least, do something other than what we expected; our theory made predictions and our theory predicted wrong. This is not a thing to be glossed over lightly! Forcing further reductions in hours may very well have some other net-negative benefit, for instance.

You are all, frankly, making exactly the mistake I'm talking about, and doubling down on it.