"Adaptive" has become the least informative word in question-bank marketing, covering everything from a topic filter to a longitudinal learner model, and the buyer's problem is that the word costs nothing while the machinery varies enormously. This page is the commercial companion to our terminology hub, /blog/adaptive-learning-medical-education-rise-dynamic-sequencing: the ten implementations that hide behind the one word, ordered by sophistication, with what each can and cannot do for a candidate, what can be verified from outside a product, and the minimum threshold we think the label should earn.
The ten implementations, weakest to strongest
One, user-selected filters: you choose topics; the system obeys; adaptive in no sense, and frequently marketed as personalisation. Two, incorrect-question mode: replay of your errors; useful review, no schedule, no model. Three, weak-topic analytics: dashboards showing where you are weak; performance-informed display, with action left to you. Four, performance-based recommendations: the system suggests what to practise next from your results; the first rung of genuine responsiveness. Five, rules-based reappearance: wrong answers return after fixed rules, three days, then seven; scheduling without a learner model. Six, spaced scheduling: intervals adjusted by your retrieval success per item or concept, a real spaced-repetition system. Seven, difficulty adjustment: items served easier or harder against estimated ability, the computer-adaptive-testing mechanism, an assessment function distinct from weakness revision. Eight, mastery estimation: a modelled estimate of your state per knowledge component, beyond raw percentage. Nine, knowledge tracing: the estimate updates across your whole interaction history, accounting for difficulty, opportunity and forgetting. Ten, study orchestration: the system decides across activities, questions, review, mocks, revisit, using performance, time, coverage and deadlines. A product may honestly implement several rungs and none of the others, and the commonest marketing move is describing rung three or four in rung nine's vocabulary.
What can be verified from outside, and what cannot
Externally observable, by any candidate on a trial: whether your next questions change after deliberate errors in one topic; whether wrong items reappear, and roughly when; whether difficulty visibly shifts with performance; whether recommendations persist across sessions, which distinguishes a model from a session heuristic; and whether you can override the algorithm, learner control being its own quality signal. Requiring vendor disclosure: what the system observes, correctness alone, or timing, confidence, hint use; whether estimates account for difficulty and forgetting; and how mastery figures are computed. Requiring outcome validation, and possessed by almost no one yet: evidence that the adaptive mechanism improves delayed, unseen-item learning against non-adaptive use of the same content, the experiment our own observatory programme exists to run on our own product rather than assert. Those three tiers, observed, disclosed, validated, are how our comparisons label every adaptivity claim, ours included.
The minimum threshold, and the buyer's five-minute test
Our proposed floor for the label: "adaptive" should mean at least rung four, the system changes what you get next in response to how you perform, with rungs one to three described honestly as filters, review modes and analytics. The buyer's test costs one trial session: answer a run of questions deliberately wrong in one topic and right elsewhere, then watch the next twenty items; a genuinely adaptive system's selection visibly bends toward the sabotaged topic, a spaced system additionally books the errors for return, and a system that serves the same mix regardless has told you its rung, whatever its landing page says. Which mechanism you actually need depends on your problem, gaps, forgetting, readiness, and the mapping is the terminology hub's job; this page's job is simpler, making the word answer for the machinery.
Frequently asked questions
Is rung ten always the best product?
No: orchestration adds value in proportion to preparation complexity, and a focused candidate with known gaps may get everything they need from rungs four and six; the ladder prices claims, it does not rank needs.
Where does iatroX sit on the ladder?
Adaptive selection with weak-area targeting, spaced scheduling, mastery estimates and study planning, rungs four, six, eight and elements of ten, with tutoring alongside; and our adaptivity's learning benefit is exactly the claim we have committed to testing rather than asserting.
What is the single biggest red flag in adaptivity marketing?
Timing-precision language, "exactly when you are about to forget", offered without any published model or validation; record it as a vendor claim and run the five-minute test.
Do any providers publish their rung honestly?
Some describe mechanisms with real precision and some describe rung three in rung nine's costume; the encouraging trend is that disclosure is becoming a competitive move, and buyers running the five-minute test accelerate it more than any editorial can.
Does adaptivity matter for short preparation windows?
More, not less: with weeks rather than months, wasted repetition on strong topics is the unaffordable cost, and rung-four targeting is the mechanism that prevents it; orchestration and long-horizon spacing matter comparatively less at that range.