Methodology & editorial policy

1. A cited fact corpus, not scraped text

We maintain a structured corpus of testable facts — dates, names, numbers, institutions — paraphrased in our own words from the Life in the UK handbook (3rd edition) and checked against gov.uk sources. We never reproduce handbook text: facts aren't copyrightable, their expression is (Crown copyright). Every fact carries its chapter/section reference. See Sources for the full posture.

2. Original questions, generated then judged

Questions are drafted against corpus facts in the three real formats (single answer, "select TWO", true/false), with plausible same-category distractors. Every draft then passes an independent automated judge — a separate model that never sees the writer's reasoning — which verifies the answer against the cited fact, checks for ambiguity, leaked hints and format violations. Fails are discarded, not fixed-up.

3. The corpus is versioned for the next edition

The handbook will eventually be re-issued. Our corpus is versioned per edition, so when a new edition lands we load it alongside the old one, regenerate and re-judge content against it, and switch — with the old edition retired, not silently mixed in.

4. Item analytics catch what review misses

Live questions accumulate statistics: how often they're answered correctly, how well they discriminate between strong and weak candidates, how long they take. Questions that are too easy, too hard or behave strangely get flagged for editorial review and retired if needed. Retired questions are replaced, never edited in place — so a question you saw yesterday can't quietly mean something else today.

5. Every card has a report button

Users can report any question in two taps. Reports land in a moderation queue reviewed by an editor; confirmed problems retire the question immediately. This is the fastest error channel we have, and we treat it as a gift.

6. Translations are drafts until reviewed

Explanations, lessons and glossary entries are machine-translated first and flagged as such internally; native-speaker review upgrades their status. Questions themselves are never translated — the real test is in English.

What we don't do

• No "real past-paper questions" — they aren't published; anyone claiming them is lying.
• No handbook text dumps.
• No invented pass-rate claims — we publish rates only when our own cohort data supports them.