Practise with a purpose.

A good training session tells you what to work on and whether the next attempt improved.

1. Find a useful focus

Your training centre groups server-recorded answers by topic and question type. It shows unassisted accuracy, answer count, timeouts and recurring high or low estimates. Hinted answers appear separately. Imported device scores do not contribute. The diagnosis uses up to the latest 200 saved flights with answers, so it reflects recent play rather than a lifetime label. Untimed practice can suggest a focus, but cannot by itself establish a strong or established assessment. Timed evidence is tracked separately.

A small number of unassisted answers is an early signal. We do not call that mastery or a proven weakness. “Developing” begins at five unassisted answers. “Established” needs at least 12 unassisted assessed answers across two flights; assessed answers come from daily play or 30-second baselines, retests and reviews. A “Strong” label additionally needs at least 80% assessed accuracy with no assessment timeouts. You can accept the suggested focus or choose any supported topic and format yourself. Year topics use numerical estimates.

2. Establish a baseline

Answer seven questions with 30 seconds each and no hints. Leaving or refreshing does not pause or reset the server clock. Your first accepted answer is final. The baseline includes the same task patterns you will meet in the retest.

3. Build reference points

Coached practice gives you seven untimed questions and three optional hints. Each hint reduces that round’s points by 20%. Feedback explains whether your estimate was high or low, whether the clock intervened, or how relative ordering works. Numerical transfer questions ask for totals, differences, twice a value or an average; date questions can ask for a gap in years. Reveals show the source values and link to the topic’s source.

4. Retest fairly

The retest uses fresh questions, matching task patterns, the same topic and format, and the original 30-second clock without hints. The report compares mean accuracy in percentage points, timeouts and answer time. Average answer time includes independent answers that were neither skipped nor timed out; timeouts are counted separately. A 50% baseline and 70% retest is a gain of 20 percentage points. Fresh questions do not necessarily mean unfamiliar entities: a small topic can revisit a mountain, country or book. A retest gain may include remembering earlier values. One cycle is evidence about that session, and later reviews check whether it lasts.

5. Return when a review is due

Your first retention review becomes due one day after the retest. A completed review with at least 80% mean accuracy, seven unassisted answers and no timeouts moves the interval through 3, 7 and 14 days. A lower score or timeout resets it to one day. Reviews use fresh questions and the same assessment clock. You can work on another focus while a plan rests.

Your notebook and saved plans

You can keep up to 20 active plans. The mistake notebook keeps up to 30 recent answers that were below 80%, timed out or used a hint. It shows the correct references, source and a suggested approach. The Train button starts a plan for that skill using fresh questions. Plans and phase attempts are saved on the server. Archive a plan to clear your active list; restore it to continue. Archiving preserves its attempts and review history.

Training adds completed flights and earned XP to your passport, but it does not affect daily streaks, daily missions or ranked daily results. Guest access depends on your browser cookie; clearing it can disconnect your saved passport.

Read the core scoring rules · Explore the topic sources · Test yourself with friends