Flashcards: SM2 vs SM5

Logseq implements SM5 as Anki did at first. Anki later changed its implementation to SM2 because of problems they found in SM5. There was a rebuttal from SM5 author, claiming that SM5 is indeed superior but that Anki’s implementation was based on a rough sketch of the algorithm:

For the sake of clarity, dozens of minor procedures were not published. Those procedures would require a lot of tinkering to ensure good convergence, stability, and accuracy. This type of tinkering requires months of learning combined with analysis.

Now, I’m afraid that Logseq is in the same boat that Anki was, because it’s based on the same rough sketch of SM5, not including dozens of minor unpublished procedures and months of tinkering.

So, given Anki authoritative opinion on these matters, what makes you believe that SM5 is a good choice for Logseq (over SM2)?


Note: I believe that by setting srs/learning-fraction to ~0 something like SM2 will be recovered asymptotically, since the OF matrix won’t adapt in that case. Notice that srs/learning-fraction can’t be set to exactly 0, because 0 (and 1) will be replaced by the default 0.5. Perhaps this constraint is neither necessary nor desirable, I don’t know.