Lindley, if a molecule can't self-reproduce, it cannot 'spread' its 'advantages'. Other molecules disappearing from the environment also doesn't help it, if it can't self-reproduce. AKA NO Darwinian evolution.
Feel free to post a scenario for Darwinian selection where the actors are not self-replicating - if you have one.
I'm not referring to Darwinian selection per se (that requires reproduction). More that if certain environments favor a particular molecular intermediate heavily, then that form may dominate in that environment; and the probability of any other outcome *prior* to that point in the chain becomes of little importance. Hence, the decision tree could be heavily pruned in places.
This works on the theory that many reactions are reversible, so even if a molecule undergoes the "wrong" reaction at some point it may eventually end up in the "right" state due to conditions. Also, there can be more than one way to synthesize a given molecule, so that needs to be considered as well.
Our inability to design a self-replicating molecule is highly relevant, showing us we're talking about a highly complex construct - that is not easily achievable by putting some environments one after the other. You need some very, very specific environments - so subtle, we didn't figure out which despite a lot of searching (not blindly, but guided by science).
You're the one arguing the number of possibilities is immense. I'm not convinced "science" gives us enough of an advantage to claim we could find in a few years when nature took billions of years to figure out. Heuristics only improve state space search if you pick the right one.
I'm not sure how it affects the math, but I'd also point out that even though the number of possible molecules increases incredibly fast, the number of possible *reactions* at each step does not. There are a (comparatively) restricted number of ways functional groups can interact. This *might* indicate your assumption of 100! possible outcomes is overly pessimistic.