So, it's a visual support for easy 2D random walk. Cool.
Though with single step only it's just a less-messy form of navigable graph.
For example, in the below random outdoor terrain generator I made, Plains are favoured (bottom of the Hex Flower) over other terrain (e.g. Woods and Hills) and Mountains are quite disfavoured (top of the Hex Fowler). Also, the way this set up, you can’t go from plain to mountains without first passing through hills. Therefore, while it is randomness, there is also a sense of continuity.
View attachment 102795
As mentioned in the comments above, you could have a weather Hex Flower that progress from mild weather to severe weather, passing through intermediate states. Also, due to probability structure, and the continuity limitations, you can ensure that severe weather is rarer than mild weather.
The one change I'd make in general would be to add a 7th option: the status quo, where you stay in the same hex. This could be done using his 2d6 model by saying that if you roll doubles other than double 1 or double 6 you stay put.
Yes.
So the obvious modifications:
* A simple way is to just with some probability "not walk", indeed.
** It's good for cases when drift back toward some equilibrium value is desirable, as all tweaks can be combined in the 2nd roll: not just "walk 1 cell/don't walk", but "walk 1 cell/move 1 step toward (X,Y),
then walk 1 cell/move 1 step toward (X,Y)/don't move".
* Another is to walk in a random direction using
more than 1 roll per step, to have a distribution pulling toward the current node.
** It will naturally happen if we walk in (1d6) direction e.g. 2 times on each step (with a result map that doesn't mind moving that many nodes per step, of course), because it forms a bell curve profile where "back and forth" is the most probable outcome:
On the example map, walking twice gives R=0 (S): 1x
6/36, R=1 (H F F/P P A/P A): 6x
2/36, R=1.5 (H F P P A H): 6x
2/36, R=2 (M F/H F/P P A/P A/H): 6x
1/36
Since dice values are constant vectors, it's just 2d6, order doesn't matter.
Simpler "always walk a step" methods ostensibly allow solitary features (results that should not be lumped too closely, like "oasis" or "rain of frogs"). But then the same result would have high probability of appearing again with 1 result in between (by walking 1 node away, then 1 node back).
Thus it would be better to generate rare and very rare results on the second pass, in context of already set results of the first pass (so that "oasis" appears only in the table for desert, and "rain of frogs" only on the table for irregular precipitations), perhaps via rolling separately on the tables for each.
The weather one would be interesting if we changed the direction of the trend by season.
Using this sort of a random walked map? Shape it as a ring (big grid with blocked center) and mark a continuous loop of cells with sequential dates. Then use one of the "walk + drift" methods, but move the mark to which the value "gravitates" into the next cell when its date is reached.