Spaced Learning Practice
Spacing is when. Retrieval is what the learner does
Distributed practice, spaced practice and the spacing effect describe closely related ideas: divide study across sessions instead of one massed block. Cepeda and colleagues’ foundational synthesis covered 839 assessments, 317 experiments, and 184 articles; the best interstudy interval increases as the final retention interval increases.
There is no universal sequence such as “1, 3, 7, 14, 30 days.” Starting cadences are pilot hypotheses, not scientific constants.
Terms that matter
| Term | Precise role | What it is not | Design implication |
|---|---|---|---|
| Distributed or spaced practice | Scheduling learning or practice across time | Simply shortening a long course | Define the required retention horizon and multiple encounters |
| Spacing effect | Observed retention advantage of spaced over massed encounters | A specific product or fixed cadence | Test whether the effect appears in the client’s context |
| Retrieval practice | Recalling, choosing, explaining, or performing without first seeing the answer | Passive rereading | Use questions, scenarios, simulations, and demonstrations |
| Feedback | Correcting errors and explaining why an answer or action succeeds | A score alone | Give immediate corrective feedback, then revisit the misconception |
| Interleaving | Mixing related problem types so learners must select the right method | Randomly mixing unrelated content | Use after foundations exist, especially for discrimination and judgment |
| Microlearning | A short delivery format | Proof that learning is spaced or retrieved | A three-minute video is not spaced learning unless it participates in a sequence |
The first consulting artifact records capability, target population, consequence of forgetting, desired retention horizon, observable behaviour, and acceptable error rate. Scheduling follows from that map.
| Required retention | Transparent starting hypothesis | Adaptation rule |
|---|---|---|
| 30 days | Day 0, 2, 7, 14, 30 | Bring forward an item after error or low confidence |
| 90 days | Day 0, 3, 10, 30, 60, 90 | Extend gaps after correct, confident application |
| 12 months | Initial learning plus reviews around weeks 1, 4, 12, 26, and 52 | Add event-triggered refreshers before critical work |
In a temporal study of more than 1,350 people, the optimal gap fell from about 20%–40% of a one-week horizon to about 5%–10% of a one-year horizon.
- Diagnostic — retention map, baseline, pilot hypothesis
- Design — experience architecture, retrieval plan, interval policy
- Convert — scenario items, rationales, SME approval, accessibility
- Integrate — channels, SSO, LRS, dashboards
- Operate — campaigns, burden caps, remediation
- Evaluate — delayed tests, transfer proxies, go/no-go
Headless engine overview
The scheduler stays headless so the practice complements rather than replaces a client’s LMS or academy. Use 1EdTech LTI where a secure tool connection is needed; use xAPI-style statements and a learning-record store for cross-channel activity.
Experience → competency graph
LMS/LXP, mobile web, email, Teams, Slack, CRM, simulations, manager coaching. Competency graph: capability → objective → misconception → scenario → evidence → business behaviour.
Scheduler → LRS
Transparent interval rules first; later, bounded mastery estimates. Event stream to learning-record store: exposure, response, confidence, latency, feedback, mastery estimate, work outcome.
Experimentation
Randomisation or staggered rollout, holdouts, power assumptions, stopping rules, reproducible analyses.
Governance
Model and prompt registry, evaluation sets, approval history, audit logs, incident management, retention policy. NIST and UNESCO guide human escalation for GenAI.
Market demand is validated by an established enterprise interval-reinforcement vendor, a frontline daily-reinforcement enablement platform, and a multidimensional adaptive-learning vendor. Whitespace for this practice is independent evaluation, white-label integration, and human-assured AI item operations — not inventing another named competitor product.
A large consumer language-learning app’s operational scheduling research used 12.9M learning traces, reduced recall-prediction error by more than 45% against several baselines, and improved daily engagement by 12%. Useful for adaptive scheduling methods — language vocabulary is not equivalent to audit judgment or physical skill. Evolve from transparent rules to adaptive scheduling only after sufficient client data exists.