A prompt or review is most effective right before forgetting occurs, but not after

What it is

Timing a review or prompt isn't "earlier is always better" or "later is always better" — there's a specific window: reviewing after a memory has started to become harder to retrieve, but before it has failed completely, produces better results than reviewing while the memory is still fresh, and better results than waiting until it's already been entirely lost. This window is a specific application of the desirable difficulty framework: landing the review right in the difficulty band where retrieval takes effort but still succeeds.

Why it happens

If a memory is still fresh at review time, retrieval takes almost no effort, and this effortless recitation contributes little to consolidation — the effort level is too low for the pathway to get meaningfully reinforced. If review is pushed too late and retrieval fails outright, the "review" degrades into a fresh learning episode rather than a reinforcement of the existing trace, and the extra consolidation payoff is lost the same way. The real payoff comes from the middle ground where retrieval is effortful but ultimately successful — that perceptible effort is precisely the condition that triggers the memory being re-consolidated more strongly, not the act of reviewing by itself.

Studying it

Studying this window means manipulating the gap between review/test and the last exposure (short, medium, long), recording whether each retrieval attempt succeeds or fails, and then checking how much a successful-but-effortful retrieval at different gaps boosts performance on a later final test. The typical finding is that a moderate gap — where retrieval becomes effortful but still succeeds — produces a markedly larger boost to subsequent retention than a very short gap where retrieval is effortless. Effort level itself can be used as a predictor of how much benefit a given review will produce, rather than treating review as a simple yes/no variable.

Where it stops holding

This holds only when the effortful retrieval attempt ultimately succeeds; if review timing is pushed so late that most attempts fail outright with no feedback given, the benefit of this window disappears entirely. Predicting exactly "when forgetting is about to happen" is itself difficult — forgetting rates differ across people and content, so applying one fixed, uniform threshold to decide "now is the moment to remind" across all scenarios inevitably carries prediction error, and drifting either too early or too late moves away from this optimal window.

Applying it

• Design reminder/review triggers to land as close as possible to, but not past, the predicted forgetting threshold — rather than assuming "the earlier the reminder, the safer." If historical data shows roughly when a given type of content or a given user typically starts forgetting, schedule the next touchpoint near that predicted point rather than well in advance.
• When reliable prediction data is unavailable, err slightly early rather than slightly late — a late reminder lands after retrieval has already failed and forfeits the consolidation bonus entirely, which is usually a worse outcome than a reminder that arrives a bit early and costs the user only an effortless recall.
• Validation: compare a "triggered near the predicted forgetting point" scheme against a "triggered well in advance on a fixed schedule" scheme, measuring unprompted recall weeks later — if the former achieves comparable or better long-term recall with fewer or equal reminders, the timing choice itself is doing the work.

Related

• Same group: A6.15.1 The optimal spacing interval should expand as memory strength grows, not stay fixed · A6.15.3 Reviewing too early has low marginal benefit because forgetting hasn't yet begun · A6.15.4 Reminder and re-education timing can directly borrow the spaced-repetition scheduling curve
• Nearby: A6.14 Retrieval from long-term memory · A6.06 Forgetting curve
• Search terms: desirable difficulty · region of proximal learning · review timing