michael_nielsen

I think I got it from http://SuperMemo.com .

It's pretty likely I got it from SuperMemo. Ironically, I don't remember. But then, I didn't put the source in Anki...

Incidentally, a few months ago I checked the rule of thumb, based on my average card review time, frequency of correctness statistics, and Anki's spacing algorithm. It came out to around 2-3 mins for an average card, based on reasonable multi-year extrapolations.

Good that you checked. I figured it was right from eyeballing my own Mnemosyne time stats as well but didn't calculate it out or anything.

I needed to do some extrapolation of long-term failure rates (since I don't have data, for the obvious reason), so don't take the number too seriously. It was a check to see that the number was in the right ballpark.

Basically, the problem with long-term failure rates is that it seems likely to me that Ebbinghaus/Leitner/etc type decay models get pretty wonky in the tail. If I review after 2 years, get a Q right, does that mean I shouldn't see the Q again for 5 years? Etc.

Probably shouldn't. The Mnemosyne dataset goes back over a decade at this point, and multi-year followups have been done, so I figure any serious misfit would be noticed by now. And if you aren't naturally exposed to any recall over decades, how important really could it be?

You have vastly more confidence than I in what questions are carefully studied & answered. As to the natural exposure question, part of the point (for me) is to use the ideas as creative stimulus, so having access to infrequently used ideas is a +.