Adaptive Review Scheduler¶
Software or tool — instantiates Retrieval-Spaced Reinforcement
A scheduling tool that calculates next retrieval intervals from recall performance, priority, or decay assumptions.
An Adaptive Review Scheduler is the timing brain of a reinforcement loop: it decides when each item should next come up for retrieval, and nothing else. It is deliberately blind to the content of the item and to how the prompt is shown — it takes in a recall outcome, an item's risk tier, and an assumed forgetting rate, and it emits a next-review date. Its defining move is that the interval is computed, not scheduled by hand: a good recall pushes the next attempt further out, a fumble pulls it back in, and a high-stakes item is held on a tighter leash than its recall alone would justify. Where a flashcard app or a quiz asks the question, the scheduler only answers "when should we ask again?" — which is why one scheduler can drive many different prompt formats at once.
Example¶
A company runs security-awareness training and wants employees to actually retain how to spot a phishing lure, not just pass the annual module. Each micro-topic — "hover before you click," "finance never emails wire changes," "report, don't delete" — is registered with the scheduler and tagged with a risk tier: wire-fraud topics are Tier 1, poster-level hygiene tips are Tier 3. When an employee gets a spaced micro-check right on the first try, the scheduler multiplies their interval for that topic (say from 14 days toward 30, then 60); when they misfire, it collapses the interval back to a few days. Tier 1 topics carry a cap: even a perfect streak never lets the interval stretch past 45 days, because the cost of forgetting is too high.
The result is that after a quarter, the finance team is being re-prompted on wire-fraud cues roughly monthly while a reliable employee's interval on low-risk hygiene tips has quietly stretched to a season. Nobody built that calendar by hand — the scheduler derived it, one recall outcome at a time, from an assumed decay curve and each topic's tier.
How it works¶
The scheduler holds, per learner-item pair, a small state: current interval, last outcome, and tier. On each retrieval outcome it runs three coupled decisions — how far to move the interval (the adjustment rule), what forgetting curve to assume for this item class (the decay model), and what floor/ceiling the item's tier imposes. A common backbone is an expanding-interval rule seeded by an SM-2-style formula[n1], where each success scales the interval by an "ease" factor and each lapse resets it. The scheduler then writes back the next due-date and hands off; it never renders the card or grades the answer. Because timing is decoupled from content, the same engine can pace flashcards, scenario drills, and huddle prompts from one queue.
Tuning parameters¶
- Expansion factor — how aggressively intervals grow after success. Larger values cut review load fast but risk letting fragile items slip past their forgetting point unseen.
- Lapse penalty — how hard a miss shortens the interval. A steep penalty recovers fragile items quickly but can trap a merely-distracted learner in over-review.
- Tier caps and floors — per-priority ceilings on how long an interval may grow. Tight caps protect high-stakes items but raise total workload.
- Assumed decay steepness — how fast the model presumes memory fades between attempts. Steeper assumptions front-load reviews; flatter ones save effort but bet on durability.
- Load smoothing — whether the engine spreads due-items to avoid review pile-ups. Smoothing keeps daily load humane but nudges some items off their ideal date.
When it helps, and when it misleads¶
A scheduler earns its keep when there are far more items than anyone can review evenly, because it spends attention where forgetting is actually happening instead of on a flat calendar. It also makes review load predictable and lets a whole program share one timing policy.
Its failure mode is trusting a model over reality. The forgetting curve is an assumption, and if it is wrong — the item is harder than its class, or the learner's context changed — the scheduler will confidently stretch an interval right past the point of collapse, and no one notices until an incident. A classic misuse is tuning the expansion factor purely to minimize review count, which optimizes for a light workload rather than durable recall and quietly hides forgetting behind a tidy queue. The guarding discipline is to treat the decay assumption as falsifiable: watch realized recall against what the model predicted, and steepen the curve (or tighten tier caps) whenever items are failing on their first spaced attempt more often than the model expects.
How it implements the components¶
spacing_interval— its core output: the concrete next-due gap for each learner-item pair, expanded, held, or contracted per attempt.interval_adjustment_rule— the success/lapse formula that moves the interval is the scheduler's central logic.decay_assumption— an explicit per-class forgetting model is the prior the engine schedules against and revises.item_priority_tier— risk tiers set the floors and caps that override raw recall when stakes are high.
It does not present the item or grade the attempt: knowledge_item, retrieval_prompt, and recall_performance_signal are owned by its nearest twin, Spaced Flashcard System, whose captured outcome the scheduler merely consumes as an input.
Related¶
- Instantiates: Retrieval-Spaced Reinforcement — supplies the timing engine the whole loop hangs on.
- Consumes: Spaced Flashcard System and Retrieval Quiz — their recall outcomes are the scheduler's input.
- Sibling mechanisms: Spaced Flashcard System · Knowledge Retention Dashboard · Post-Training Recall Check · Retrieval Quiz · Recurring Practice Prompt · Refresher Training Protocol · Scenario Recall Drill · Skill Maintenance Drill
Editorial Notes¶
Form Classification¶
Form family: Control, Automation & Runtime
Rationale: The mechanism is a scheduling tool that calculates next retrieval intervals from recall performance, priority, or decay assumptions, so its operative form is state-dependent runtime control or automated actuation.
Independent corroboration: The frozen evidence defines Adaptive Review Scheduler as 'A scheduling tool that calculates next retrieval intervals from recall performance, priority, or decay assumptions', so its operative form is Control, Automation & Runtime.
Nearest alternative: Rule, Policy & Commitment — It schedules reviews state-dependently during operation rather than conducting the review itself.
Review outcome: Independent reviewer agreement; medium confidence.
Origin Attribution¶
Primary origin: Education & Pedagogy
Origin pattern: Cross-disciplinary synthesis
Present-day reach: Specialized
Rationale: Spaced-repetition practice computes expanding review intervals from recall outcomes so instructional time returns to material near its expected forgetting point.
Related originating lineages:
- Cognitive Science — Memory-strength and decay models contribute the inferred state used to place the next review.
- Computer Science & Software Engineering — SuperMemo's SM-2 and successor algorithms operationalized item-specific ease factors, risk tiers, and automatic next-review dates.
- Psychology — Experimental memory research supplies forgetting curves, retrieval practice, spacing effects, and desirable difficulty.
Review resolution: Scheduling review from recall performance and spacing is an education and pedagogy mechanism. Cognitive and psychological memory models are formative, while computing makes the schedule adaptive; the page generalizes these established lineages without claiming an encyclopedia invention.
Attribution caveat: The named scheduler is computational, but it is a direct educational implementation of memory research.
Review outcome: Reconciled after independent review; high confidence.
Notes¶
The scheduler is a policy, not a UI. Keeping it separate from the prompt tool means you can change when items recur — steepen decay, tighten a tier — without touching what is asked, and you can point one timing policy at several different prompt mechanisms.
[n1] The SM-2 algorithm, introduced in Piotr Woźniak's SuperMemo, sets each item's next interval by scaling the prior interval by an "ease factor" that rises with easy recalls and falls with hard ones — the canonical expanding-interval adjustment rule that most adaptive schedulers descend from. ↩