Skip to content
Heidi

Exercises

A short set of questions, a few minutes. Built from the rules Heidi applies herself, from the lines people actually say — and from the words you kept.

What you keep stays in your browser. The questions from the vocabulary need no account.

Just: Commands and requests: the polite form ends in -edPractise everything

Practise or test?

A test only asks questions that can be marked outright. Writing and cards are marked by you, which you can practise but nobody can measure.

There are no questions for this yet. That does not mean the topic is unimportant — only that the pack has no examples for it so far.

Why the exercises are built this way

Every decision here can be checked. Where the research gives a direction and not a number, the number is marked as our judgement — not as a finding.

  • Being asked beats looking again.

    So no exercise shows the answer first. Across 222 studies the advantage of quizzing over restudying is g ≈ 0.50.

    Yang, Luo, Vadillo, Yu & Shanks 2021

  • Later beats sooner — and the advantage grows with time.

    So a kept word comes back after 1, 3, 7, 16 and 35 days rather than daily. In a second-language meta-analysis: g ≈ 0.76 on immediate tests, g ≈ 1.15 on delayed ones.

    Kim & Webb 2022

  • Right once is not enough; right twice, spaced, is the point.

    So a missed question comes back before the sitting ends — three questions later, not immediately. The gap of three is our estimate: the study gives the direction, not the number.

    Rawson & Dunlosky 2011

  • Choosing without feedback can fix the wrong form in place.

    So every multiple-choice question shows the right answer at once and asks again later. Feedback increases what such tests are worth and reduces exactly that harm.

    Butler & Roediger 2008

  • Producing it yourself sticks better than reading it.

    So there are gapped sentences and open questions, not only choices. The effect is robust but smaller than its reputation suggests.

    Bertsch, Pesta, Wiscott & McDaniel 2007

  • Mixing helps — but not always, and we say so.

    So no two questions of the same kind follow each other. The meta-analysis is clear: the benefit depends on how similar the material is, and on very similar material mixing can hurt.

    Brunmair & Richter 2019

  • No streaks, no points, no percentage.

    A streak measures how much Heidi you consumed while looking like a measure of learning. What is here are counts of what you did.

    Yang, Luo, Vadillo, Yu & Shanks 2021

The whole method