How we build the questions

Where the facts come from

Every card must be traceable to an authoritative, governing source for its subject. No card is authored from memory, from general websites, or from AI paraphrase of other sources. The primary sources by subject type:

Secondary sources — recommended textbooks, examiners' reports, Ofqual-accredited revision guides — are used to clarify wording and confirm what the mark scheme actually rewards. Examiners' reports are particularly valuable: they reveal which phrasings examiners accept and which common student answers lose marks, and that knowledge directly informs how card answers are worded.

Mapping to the curriculum

Card topics and subtopics map directly to the structure of the relevant specification — not to how a textbook organises content, and not to how a teacher might plan lessons. Topic names are the spec's main content areas; subtopic names are the spec's sub-sections or learning outcomes. No topic names are invented.

This matters because the exam is written against the spec, not against any particular teaching sequence. A student who has practised every spec subtopic has covered everything that can be assessed — regardless of how their teacher chose to order the content across the year.

Board coverage. For multi-board subjects (History GCSE covers AQA, Edexcel, OCR, Eduqas, and WJEC; Geography GCSE covers AQA, Edexcel, OCR, Eduqas, and WJEC), each card carries a boards array. A card tagged ["aqa", "edexcel", "ocr"] only appears for students on those boards. A card tagged with all boards covers a topic that every board assesses. A school sets its class to a specific board, and the practice picker filters automatically — students never see cards that fall outside their own specification.

KS3 subjects carry an empty boards array [] by design: no exam board applies to KS3 content.

How many cards each topic gets

Not all spec subtopics deserve the same number of cards. A simple rule like “eight cards per subtopic” produces thin coverage on the long-answer questions that decide A and A* grades, and wastes authoring effort on minor definitional points tested at one or two marks.

Educator uses a four-class depth framework. Each subtopic is classified when a subject is built; the class determines the card target:

Classifying a subtopic requires reading the published mark scheme, not just the specification. The same subtopic heading can sit at different depths across different boards — one board might dedicate a 12-mark question to it, another might only test it as part of a broader question.

Card types and why they exist

Different knowledge points are best tested differently. Naming a vitamin is not the same cognitive act as identifying whether a statement about it is true or false. Using multiple card types across the same subtopic means students encounter the same knowledge from different retrieval angles — which research shows strengthens retention more than repeating the same prompt format.

Recall — short answer

The student types a word or short phrase. Used for names, dates, terms, and single-fact identifications where there is one canonical answer. Answers are capped at six words. This is a strict rule — if the correct answer cannot be expressed in six words, the question is testing understanding rather than recall, and belongs in a different format.

MCQ — multiple choice

One correct answer plus three wrong-but-plausible distractors. The distractors are not silly alternatives — they are the answers a student who half-understands the material might actually give. MCQ is used when the knowledge point has obvious near-synonyms (making a recall answer ambiguous) or when a closed answer space is needed. For History GCSE, MCQ answers are frequently paragraph-length explanations, matching the way examiners assess contextual reasoning on the real paper.

Cloze — fill in the blank

A sentence with one blank; the student supplies the missing word. Used when the knowledge point is a term embedded in a natural process description — physical geography processes, food science mechanisms, biological systems. The single-blank rule is absolute: two blanks in one card produce ambiguous marking and split attention between two separate knowledge points.

True / False

A factual claim the student marks as true or false. The claim must be definitively one or the other — no nuanced edge cases. False cards are written to represent a specific common misconception, not just a negation of the true fact. Within any subtopic, the split between true and false cards is kept roughly equal to prevent students from pattern-matching.

Match — matching pairs

Four pairs of related terms presented simultaneously — the student taps a left-side item then its correct right-side match. Used when four similar concepts are easily confused with each other and need distinguishing together (plate boundary types and their features, countries and development categories, landform and formation process). Each left item must have exactly one unambiguous right-side match.

Year — date identification

The student types a four-digit year. Used in History for events where the specific date is assessed (“In what year was Hitler appointed Chancellor?”). Answers are validated as a number; there are no distractors.

Fill the gap — paragraph cloze

A model paragraph is shown with one word or phrase blanked out. The student selects the correct option from MCQ-style choices — the same interaction as a standard MCQ, but with a longer prose context rather than a standalone question. Used in Religious Studies and History where the knowledge point sits most naturally inside a sentence of explanation rather than as an isolated fact.

Mark it — mark-scheme literacy

A model answer is shown on screen. The student selects which mark band it earns — for example, “3–4 marks” or “7–8 marks” — from a list of the band descriptors. The meta-skill being built is mark-scheme literacy: understanding what distinguishes an adequate answer from an excellent one before the real exam. These cards appear in the dedicated Mark It practice mode.

Most students use Educator on a mobile phone. Recall and cloze cards require the virtual keyboard, which shifts the layout and interrupts the tap-based practice flow. Because of this, recall cards are capped at roughly 15% of any subject's corpus — a knowledge point that can be expressed as a cloze or MCQ without losing pedagogical value is converted, not kept as a recall card by default.

Quality rules every card must pass

A card that teaches a wrong fact is worse than a missing card. A card with an ambiguous answer trains students to guess. The following rules are non-negotiable and are checked automatically before any card set is promoted:

• One unambiguous answer. Only one correct answer may exist. If a teacher could reasonably argue that an alternative phrasing is also correct, the question is reworked to be specific enough to exclude it.
• Answerable in under three seconds by a confident student. If a student who has studied the topic well would need to stop and think, the card is likely testing understanding rather than recall. It is either reworked or flagged for a future “explain” card type.
• No open-ended prompts. Prompts like “Give an example of X” or “Name one Y” have multiple correct answers that no accepted-answers list can cover exhaustively. A student who gives a perfectly valid but unlisted answer gets marked wrong. These are always converted to MCQ, where the answer space is closed.
• MCQ distractors must be plausibly wrong, not obviously wrong. The standard test: if a student who half-knows the material could not pick the distractor, it is too weak and must be replaced. Conversely, if a student who does not know the answer could identify the correct option because it is noticeably longer than the others, the distractors must be expanded to match.
• No first-letter hints. Prompts that quote a letter matching the first letter of the answer (“What ‘H’ does binary search do at each step?”) are forbidden. This pattern emerges when an authoring batch has exhausted the legitimate facts on a subtopic and starts manufacturing filler — the card no longer tests the underlying knowledge, it tests letter-matching.
• At most two cards per knowledge point. One recall and one MCQ on the same fact is acceptable — they test from different angles. Three or more cards on the same point are duplicates and the weakest are removed.
• Numerical values verified against current sources. Exam boards update mark schemes. A figure that was correct in a previous specification may have changed. Any card whose answer is a specific number is reviewed against the current authoritative source before promotion, and flagged for annual re-verification.

Foundation, Higher, and Both

Every GCSE card carries a tier label. This is not a difficulty rating — it reflects which tier the exam board actually assesses the knowledge at:

Foundation

Facts and definitions that appear in Foundation-tier mark schemes. Basic, lower cognitive-demand knowledge. Students sitting the Foundation paper only see Foundation and Both cards.

Higher

Knowledge only assessed at Higher tier, or requiring higher-order understanding — causation, significance, mechanism, application. Students sitting Higher see all three tiers.

Both

Knowledge assessed across both tiers. The majority of cards in any subject sit here. This is correct — most core factual recall appears on both papers.

A school sets its class to a tier; the practice picker applies the filter automatically. Individual learners pick their own tier during onboarding and can change it in their profile at any time.

Expert review and the promotion gate

Cards are authored in waves — a batch covering one or more spec subtopics — and each wave goes through a multi-stage review before it is promoted to the live card bank.

Automated checks

Two scripts run against every wave before any human review:

• Card linter — checks structural rules: answer length, distractor count, board array format, no empty required fields, tier values, no orphaned subtopics. A wave cannot be merged if any linter error fires.
• Smell test— flags pedagogical quality patterns that the linter cannot detect: first-letter hints, self-referential prompts (the answer appears verbatim in the question), template repetition (five or more cards in the same subtopic sharing the same opening four words), and MCQ distractor length imbalance. Flags at the “high” severity level block promotion; “warn” severity requires documented review.

Subject specialist sign-off

A qualified subject teacher reviews every wave. Their role is not copy-editing — it is checking that each card tests what the mark scheme rewards, that the wording matches how the subject actually uses the term, and that no card teaches a subtly wrong framing that would mislead students under exam conditions.

For each new subject, a subject-specialist teacher reviews the full corpus before the status changes from Coming soon to Live. A subject sitting at Coming soon on the homepage has a complete card bank and working practice engine; it is waiting on that pedagogical sign-off.

Ongoing refinement

The card bank is treated as a living document, not a one-time publication. Several mechanisms keep it accurate over time:

• Card reports. Teachers and students can flag any card as incorrect, ambiguous, or poorly worded. Flagged cards are reviewed and corrected or removed. During the pilot phase this is restricted to staff by default — at pilot volume, crowdsourced signal is noise, not evidence.
• Annual numerical review. Cards whose answers are specific numbers — nutritional reference values, case study statistics, physical constants — are tagged for annual re-verification against current authoritative sources. Exam boards update mark schemes; a correct answer from a previous specification may no longer be accepted.
• Spec change tracking. When an exam board publishes a revised specification, the affected subjects are audited against the new version. Cards covering removed content are retired; gaps in new content are added.
• Coverage audits. Periodic automated audits check that every named section of every board specification has met its depth target. Thin subtopics are surfaced as authoring backlog items and addressed in the next content sprint.

What Educator does not cover

Spaced repetition is the right tool for discrete factual recall, and Mark it cards build mark-scheme literacy on top of that. Some parts of GCSE assessment remain out of scope:

• Full essay writing. Educator builds the factual foundation and mark-scheme awareness that strong extended answers draw from — but the act of constructing, structuring, and refining a multi-paragraph response under timed conditions requires teacher feedback and practice papers. No card type replaces that.
• Hands-on practical work. Laboratory work, workshop practice, and similar activities require physical presence and supervised instruction. Educator covers the underpinning knowledge those practicals assess, not the practicals themselves.
• AI-generated content. No card in Educator is generated at runtime by AI. Every card is pre-authored, linted, and reviewed. The curriculum is fixed; what the adaptive algorithm personalises is the order and frequency of review, not the content itself.

Further reading

• Subjects — all subjects, board coverage, card counts, and GCSE vs KS3 in practice
• Spaced repetition & SM-2 — the algorithm that determines when each card is shown: intervals, ease factors, mastery levels
• Practice modes — how Standard, Speed Round, Marathon, and Drill Weak Cards each use the card bank
• HoD guide — making the case to SLT, reading the class analytics, and running a pilot