Yes; OTOH, if you can already guess in which direction your observations will be moved by more rigorous study, you should move them already.

Not quite true. Say I have a d100 and I have two hypotheses---it is either fair or it is biased to roll '87' twice as often as any other number. I can already guess that my observation from rolling the die once will move me in the direction of believing the die is fair (ie. I can reliably guess that I will roll any number except 87). However, if I do happen to roll an 87 then I will update towards 'biased' to a greater degree.

It isn't the direction of the expected evidence that is conserved. It's the expectation (the directions of all the possibilities multiplied by their respective degrees).