Brugle,
I was not particularly suggesting that randomness would directly affect that particular property of a walker, merely emphasising that in my tests any randomness there may be has been removed.
I am picturing in my mind my recent encounters with diagonal 'reversers' and I cannot remember one that contradicts what you say. I have a few in my current 'Valentia' project and they all 'reverse' to NE or SE (with one bizarre exception - see below). If your observation is correct (and I don't doubt for a moment that it is) that would suggest that NE is the preferred direction for any walker's first tile of travel whatever random seed is used (I don't see why a walker undergoing a mode conversion on a diagonal path would be a special case). I expect that to be established one way or another after tonight's scheduled experiments (which this time, hopefully, will not take me until 5.30 am).
My reference to StephAmon's comments on 'false start' walkers and their behaviour "even with randomization" further supports the idea that a walker's first tile of travel is simply fixed to a NE preference (and clockwise 'next bests') without modification. Indeed, the standard 'default' walk will begin towards NE if possible.
As for that exception - I mentioned the point about diagonal walkers on a 'mesh' passing through the target tile and continuing diagonally without changing direction until they reach the edge of the grid. I have seen examples of walkers grounding to a tile on the NE edge of the grid (travelling east) where you would expect them to turn SE, but they turn south! One would not have expected that choice to be available to them.
I don't fully understand, but I've said a few times that the game doesn't seem to treat a dense mesh of intersections as separate junctions requiring choice. However, these are rare in normal gameplay. I merely point out, since we discussed some tests of yours elsewhere, that (in Caesar 3 at least) mesh-like networks do not produce 'normal' behaviour and tests on them are likely to be misleading.
It is very satisfying, after all this time, to sit and watch walkers twist and turn through perhaps a dozen intersections and to be able to accurately predict every step from spawning to return. Of course, I can only do that because I've removed an element of variation in the algorithm. It's not of much practical use except, as Trurl says, in the case of walkers making multiple passes through the same intersection and, as he notes, the common feature in Caesar 3 which would cause fewer problems with the application of knowledge is the granary. If Pharoah works the same way we might ponder just how long should those stub roads be where walkers can turn into booths, bandstands and pavilions.
Note - attention is drawn to a rewrite of one paragraph of utter rubbish in my previous post which I screwed up first time
Edit: Since I am not yet qualified to post back-to-back and I have more information I am breaking my own rule and starting a new thread. I suppose this one is over-long now anyway - it takes some time to load over my connection.[This message has been edited by Trium3 (edited 06-24-2008 @ 10:19 PM).]