Personally, I'd rather tweak the AI so that it's not so sensitive to the turn order in the first place, especially since I have no easy way of knowing in advance what will be the turn order (that is, without running a part of AI algorithm). That, and the units order would shift depending on the board state. Besides, most of the time the order shouldn't be relevant - the fact that it is in some cases I consider more of an accident at work.
What I see as a flaw here - compared to the intended vision of the game - is that a shadow is a doofus and decides it cannot move before the guard makes way. In other words, an AI oddity that some levels hinge on.
If units coordination was improved - for example, a unit would revise its plan after other units planned their move - then the internal order of enemies processing wouldn't matter as much. That's because either:
a) the guard is processed first, the shadow follows the guard and plans their move
b) the shadow is processed first - not finding any opportunity to advance, then the guard is processed leaving some room for the shadow, then the shadow revises the move and decides to advance after the guard
Both situations would result in both guard and shadow advancing.
As I see it, improving the AI this way (and similar):
- would make it act in a more common-sense way, which is easier to predict than enemies getting blocked by decision order, and thus puzzles can't rely on this oddity
- would open up some attack opportunities that the current AI would miss (e.g. an esper stepping away, making path for gunner to the attack position); this would be tricky to predict, but I think it's preferable to situation when player sees such an attack opportunity and gives up on trying out some move, only to later find out that move was quite legitimate
- would render quite a few level unwinnable (right now more likely levels to be broken by new AI would be 4 and 8)
Overall, it would move focus away from careful planning (is it how Twinfold puzzles work?) towards trial and error and educated guesses. You could rely on AI to use an immediate attack opportunity and could expect it to advance towards the spots to attack the player from, but trying to predict which specific formations it would advance to would be quite futile and inefficient compared to simply trying out a some move.
I think for this specific game - with increasingly complex interactions between enemies - emphasizing heuristics works better over focusing on strict planning. With this kind of gameplay, undo feature is probably much more essential than drowning the player with detailed information, because it facilitates the trial and error. Of course, improvements to displaying basic and easy to process information are still desired (e.g. which enemies are at the attacking position or one move-turn away from it). Nonetheless, with AI having more flexible move order the information about units order becomes purely cosmetic if not misleading.
--------------------
About level 5:
I don't really think there's so much ambiguity at play here? The move order shouldn't be so relevant here, as long as you expect AI to move closer guard first, leaving room for the further guard. Same goes for level 2 - because closer guards advance earlier, they leave room for others and thus move in a mostly coordinated manner.
There is some ambiguity about specific fields enemies might choose (if multiple fields are at the same turn distance from the player) and it's pretty easy to get attacked by a gunner because of their broad attack range, but otherwise enemies consistently stick to the basic "move towards the nearest attack spot, if reached the attack spot attack the player" principle.
There is so much room to move around that collective AI behaviour is barely affected by exact units order, if at all. So even if I did provide the move order information, I don't think this information would be so much more useful compared to basic player intuition (in level 4 it's different because the room is narrow and there are slower and faster units).