To fix before documenting

[bmad4ever]

To fix before documenting

[bmad4ever]

3. continuation: since entropy is no longer multiplied w/ final cost, at the very least, zero entropy nodes should have their cost adjusted to zero. At first glance this seems important, as not prioritizing collapsed cells will remove analysis of impossible states in the immediate vicinity of the search space; and, since the algorithm is greedy, it is important to validate them soon rather than later, since backtracking is disincentivised via the costs to speed up the generation.

Alternatively, or complementary, entropy should work similarly to temperature, lowering the random weights and the freq adjustments when low.

[bmad4ever]

addressed in commit 02a9112.

keeping old costs w/ high random weights at high temperatures is not ideal, but changing this behavior requires a custom search implementation or hijacking the py_search priority queue.
I am not keen on changing the behavior; especially without comparing against normal wfc implementations first.
It may be something to explore in the future, but, for now, this issue is closed.