Adaptive Practice Set¶

An adaptive practice tool — instantiates Formative Feedback Loop

Reads each attempt and automatically picks the next item, difficulty, or scaffold, so the practice set continuously reshapes itself around exactly where the learner is still weak.

An Adaptive Practice Set is a bank of practice items wired to a selection policy that reads each response and decides what to present next — an easier item, a harder one, a hint, or a fresh attempt at the skill just missed. Its defining move is that the sequence is not fixed in advance: the set spends the learner's time on whatever the evidence says is still weak and stops spending it on skills already shown. Where a human deciding the next drill after reviewing attempts is Practice Review, the adaptive set is the automated closed loop — signal in, next item out, with no human in the moment. That is both its strength and its hazard: it can respond instantly and individually at scale, but only to whatever it can measure.

Example¶

A seventh-grader working through fraction addition in a practice app misses two items where the denominators differ. The set doesn't march on to the next topic; the selection policy infers the weak skill is finding a common denominator, drops difficulty, and serves a short rung of common-denominator items with a worked hint on the first. She clears three in a row, so the policy raises difficulty back to mixed-denominator addition and interleaves one earlier-mastered item to confirm it stuck. On the mixed items she succeeds — and a few items later the set re-serves the common-denominator skill once more, cold and with no hint, to check the gain held rather than being hint-propped. Fifteen minutes of practice have been spent almost entirely on the one sub-skill that was actually broken.

How it works¶

Items are tagged by skill and difficulty, and each response updates a running estimate of the learner's mastery per skill.
A selection policy — not a fixed order — chooses the next item to target the weakest or most informative skill, adjusting difficulty to keep success in a productive band.
After adapted practice, it re-serves the targeted skill cold to verify the gap actually closed, rather than accepting a hint-assisted success.
When one skill resists repeated adaptation, it stops iterating and hands off instead of grinding forever.

Tuning parameters¶

Selection objective — target the weakest skill, hold a fixed success rate, or maximize information per item; aggressive weakness-targeting improves fastest but can feel relentless.
Difficulty step size — how far difficulty moves per response; large steps adapt quickly but overshoot, small steps are smooth but slow.
Mastery threshold — how much evidence counts as "learned" before a skill is retired; set high and practice drags, set low and gaps slip through.
Escalation trigger — how many failed adaptation cycles on one skill before handing off to a human or a reteach; guards against the algorithm looping forever on a gap it can't close.
Recheck spacing — how long after a success the skill is re-tested; wider spacing tests durability, tighter spacing tests immediate uptake.

When it helps, and when it misleads¶

Its strength is something a fixed worksheet cannot give: near-one-to-one responsiveness at scale, keeping each learner in the band where practice is neither trivial nor hopeless,^[1] and closing its own loop fast so gaps get worked the moment they appear.

Its failure mode is that it can only adapt to what it measures, and it will faithfully optimize the proxy — item score — even where that diverges from durable capability. Push weakness-targeting too hard and it drills to the test; a shallow skill model mislabels the gap and adapts to the wrong thing; and because the loop is opaque, a learner or teacher may not see why it decided what it did. The classic misuse is trusting rising in-set scores as proof of learning without ever checking transfer to un-adapted, real tasks. The discipline that guards against it is the cold recheck plus a periodic human look at whether the gains show up outside the practice set.

How it implements the components¶

progress_signal — every item attempt is the raw evidence the set reads to drive selection.
adjustment_rule — the selection policy is the adjustment rule: it maps the current mastery estimate to the next item, difficulty, or scaffold.
recheck_cycle — the cold re-serve of a targeted skill closes the loop and confirms the gain held.
escalation_threshold — the failed-cycle limit that stops the algorithm iterating and routes the learner to a human or reteach.

It does not name the gap in human-meaningful terms (gap_diagnosis — that's Misconception Probe and Practice Review), write actionable prose feedback (actionable_feedback — Coaching Check-In, Formative Rubric), or ground itself in an explicit standard (criteria_reference — Formative Rubric).

Instantiates: Formative Feedback Loop — it is the automated, in-the-moment version of the loop, generating its own signal and closing on it without a human.
Sibling mechanisms: Practice Review · Rapid Feedback Cycle · Coaching Check-In · Draft Feedback Cycle · Exit Ticket · Formative Rubric · Low-Stakes Quiz · Misconception Probe · Peer Review Protocol · Progress Dashboard

References¶

[1] Vygotsky's zone of proximal development — the band between what a learner can do unaided and what they cannot yet do even with help. Instruction is most effective when it targets that band; an adaptive set's difficulty-tuning is an attempt to keep practice inside it. ↩