The Dose. What the spacing research licenses.

One study then mapped the whole gap-by-delay grid. Over 1,300 people learned trivia facts, reviewed them after gaps from minutes to 3.5 months, then sat a test up to a year later. For every retention goal there was a sweet spot: review too soon and the second pass adds little, wait too long and the memory is gone before you return. Practicing at the measured optimum improved final recall by 64% over no gap at all (d = 1.1).

The sweet spot is a ratio, not a constant, and the ratio itself shrinks: roughly 20-40% of a one-week goal, down to 5-10% of a one-year goal. Want to remember something for five weeks? The interpolated optimum was about an 8-day gap. For a year, about 27 days. Kang's 2016 practice guide compressed this into the working rule most tools use - space reviews at about 10-20% of the retention interval.

It licenses spreading practice out and letting the gaps grow with the goal. It does not name 30 seconds, it does not name daily, and it was measured on trivia facts in a lab. Those are the first two places our design goes-beyond-the-data, and the assembly section below marks them.

The same meta-analysis demolishes a popular refinement. Expanding schedules - short gaps first, then progressively longer ones, the signature feature of several flashcard algorithms - measured g = 0.034 against plain even spacing. Nothing. The authors' own words: the results do not support the wide belief that intervals should be progressively increased. Spacing is the active ingredient; the fancy curve is branding.

The most practical version of this is called successive relearning: retrieve until correct, then come back days later and do it again. After 533 participants and more than 100,000 scored responses, Rawson and Dunlosky printed a prescription. Recall a concept correctly three times up front, then relearn it three times at widely spaced intervals. In a real upper-level course, students who did this every few days beat their own baseline material by a letter grade or more in both experiments (d = 0.54 and d = 1.10, about 10-16 points). Small classes, one course, but this is the closest published relative of a spaced daily rep.

The efficiency numbers matter for dosing. In the relearning experiments, once students had learned a concept, each spaced relearning pass took about a minute or two per concept. Durable memory does not come from hours. It comes from returns.

Martinengo's 2024 synthesis covers 23 studies of spaced digital education for health professionals, 19 of them randomized, 3,371 participants. Spaced beat massed online education on post-intervention knowledge by about a third of a standard deviation (SMD 0.32, CI 0.13 to 0.51), graded moderate certainty, and held a similar edge on retention. The authors attach their own warning - most included studies carry unclear or high risk of bias, and heterogeneity is substantial. An earlier Phillips review of 17 clinician studies found the same direction: knowledge moves, behavior sometimes, patient outcomes almost never measured.

The nulls from the retrieval practice page still stand, and one new one joins them. The 537-resident email trial moved its own quizzes but not the specialty exam. The 634-resident cluster trial was null on certification scores. And in 2025, the nearest published relative of our exact product - one multiple-choice question texted to 293 pediatric residents every weekday morning - found no exam effect for the plainest possible reason: residents stopped answering. Median 34 questions answered in eleven months; a fifth never answered one.

Read that last null carefully, because we did. It is not evidence that a daily question cannot work. Adherence collapsed before anyone could test the cognition. It is evidence that an opt-in side channel with no social context dies. That is the strongest argument in this literature for delivering the question where the team already is - and it is an argument about plumbing, not proof that our plumbing works.

The brevity choice has weaker direct evidence, and we say so. No meta-analysis carries a quiz-length or session-duration moderator - we looked. What exists is circumstantial. In Kim's 10,514-employee workplace dataset, real training sessions already averaged about three minutes and two to three questions, and the spacing effect showed up anyway. Zheng's lab study of retrieval under working-memory load found the benefit only when the task left spare capacity, which argues for small bites (30 people, one study).

The word for tiny lessons is microlearning. The honest read of that literature: it is mostly marketing. The two serious reviews are scoping reviews of about 17 studies each, with one randomized trial between them and zero studies measuring the outcomes that matter most. We do not cite microlearning as evidence. The evidence is the spacing and retrieval literature above; micro is just the size our dose happens to be.