The evidence

Flora is built on two findings from cognitive psychology. Before writing a word of the marketing page we reviewed the literature behind them, sceptically, and the review is what the page is allowed to say. Where the two disagree, the review wins.

What’s well supported

Spacing your practice out beats doing it all at once. Cepeda and colleagues synthesised 839 assessments from 317 experiments in 2006; Donovan and Radosevich found a mean effect of d = 0.46 across 63 studies. It goes back to Ebbinghaus in 1885 and it has survived psychology’s replication crisis better than almost anything else in the field.

Testing yourself beats reviewing. Roediger and Karpicke had students learn science passages in 2006. A week later, the students who had tested themselves recalled 61% of what they’d read; the ones who had reread it recalled 40%. Five minutes after studying, the rereaders had looked better - which is roughly why most people reread.

These are the two best-evidenced study techniques there are. When Dunlosky and colleagues rated ten common techniques for Psychological Science in the Public Interest in 2013, distributed practice and practice testing were the only two rated high utility. Highlighting and rereading came out bottom.

Longer gaps mean remembering for longer. Cepeda’s 2008 study taught facts with review gaps of up to three and a half months and tested up to a year later. The best gap is roughly 10-20% of how long you want to remember something for, which is why intervals expand.

What we won’t claim

That our scheduler makes you learn more than any other. No controlled trial establishes that a sophisticated algorithm beats simple consistent spacing on learning outcomes. Karpicke and Roediger found in 2007 that equal intervals beat expanding ones at a two-day delay; what mattered more was that the first test was delayed enough to be effortful. Flora uses a plain ladder of intervals for that reason, and we’d rather spend the effort on card quality.

That this builds understanding. The spacing benefit is large for discrete facts and attenuates or disappears for complex subject matter - Donovan and Radosevich found no effect at all for genuinely complex tasks. Recalling a definition is not the same as recognising the thing in the wild.

Any outcome figure of our own. We haven’t measured Flora against anything, so we don’t have a number, and the ones in this category usually come from surveys of people who stuck with an app. Roughly half the people who try Duolingo stop within a week; a figure measured on the ones who stayed says very little about what happens when you download something.

What that leaves

Adherence, mostly. Across the literature, the difference between people who end up remembering and people who don’t is overwhelmingly not the schedule they used - it’s whether they kept going. That’s why Flora is built to make the days short and to remove the card-writing rather than to make the scheduling clever.

Where the rest of it is

The full review runs to a few thousand words with the citations attached, and it sits with the code rather than here, because its job is to constrain what we build. If you’d like a copy, ask.

What types of information Flora works best on is the practical version of this page.