Erfworld 119, page 107

[fendrin]

Erm, Failure to do your duty is failure. Betrayal is a choice. Once must choose to not do their duty, or choose to do something that is against their duty.

I suppose I make a distinction to fail to do your duty and failure to successfully achieve the goals of your duty.

To me, 'failing to do one's duty' means choosing not to do it.
Choosing to do what duty demands and failing is different.

Or if there's another predictamancer working for another side. Then there can be logical contradictions.

Example:
Side A is defending five cities, and has enough troops to defend two of them from side B's force.

Side A uses a predictamancer to determine which city/ies side B will attack the heavies. They then put most of their troops in those cities.

Side B uses a predictamancer to determine which city A will defend the least, and attack that city.

The simplest way I can see to reconcile that is if you assume that predictamancy can only function based on the current parameters.

To use your example, Side A's predictamancer predicts what Side B would do if Side A did not move their forces. Thus if Side A consulted the predictamancer, moved their forces accordingly, then consulted the predictamancer again, they would get a different answer.

This would make predictamancy much less useful, even if it is 100% accurate, because that accuracy is predicated on not acting on the information in such a way that causes the relevant parameters to change. This of course makes predictamancy only particularly relevant in the short term... make long-term predictions such as Banhammer needing Jillian unusual.

All the more reason to suspect that the predictamancer wanted Jillian popped for reasons of their own...

There are some interesting paradoxes about predictamancy.

See above.

Assuming that one can only make predictions at the start of one's turn...

I would say that a predictamancer has a limit based on fatigue, like Maggie does.