The dice model models only if there is a NS memory facility in the organism to effect selection because with the dice there is something/someone to save the sixes for success.
Like some kind of 'diceome' where child rolls inherit their parent's diceomes.
If only there were some kind analogy in biology that 'saves' successful genetic mutations and some kind of way that insures that the mixes that are better at reproducing are the ones whose 'saved genes' get passed on to their children.
The thing with analogies is that if you aren't careful, you'll miss the point. We could create an analogy that was closer to the real thing, but the analogy would get more and more complicated until we might as well do away with it. Once everyone agrees that cumulative selection can do interesting things, we could just discard the dice analogy altogether. Or...
...maybe we could have a population of piles of dice, and each pile of dice replicates but those piles of dice with more sixes in them get to create more children than those with fewer and each replication introduces a random roll to one of the die in each pile and...