Game randomizers (such as those for older Zelda and Metroid games) address this sort of problem with logic to ensure that a seed is solvable. If you need item A to enter dungeon B, item A can't be in that dungeon, for example. This can either be applied as part of the random generation, or as a post-generation verification step which discards the seed if it isn't solvable.
I think this sort of thing would be valuable for randomized run-based games like Slay the Spire, not only to ensure consistency, but also to open up greater flexibility for design. Instead of using fuzzy heuristics so that nearly all seeds are solvable, you can construct a greater diversity of scenarios without risking player frustration.
I feel like it is pretty indistinguishable in STS if you lost because you made an incorrect decision somewhere or the run is unwinnable. Something like Zelda or Metroid you can reach a point and the game has some kind of obvious solution or key you need. STS has too many solutions to the problems of what you face and no way to go back and recheck things. The fact that it is almost impossible to find an unsolvable STS seed seems to indicate that their approach is pretty good. Player frustration in STS except for the tiny number of runs indicated in this post is entirely skill based.
That's easy because the only check for solvability is item availability, so by setting a dependency chain you can ensure it. We can't do this in StS, since the main obstacle is being able to do enough damage while staying alive, not items per se. Solving that would require simulating battles which is way too expensive due to number of choices.
I think this sort of thing would be valuable for randomized run-based games like Slay the Spire, not only to ensure consistency, but also to open up greater flexibility for design. Instead of using fuzzy heuristics so that nearly all seeds are solvable, you can construct a greater diversity of scenarios without risking player frustration.