Use caseThink
Learn a new field on the canvas, and find out what you got wrong
A map of the field with your own answers written into it: the foundations you were missing, an order to take the topics in, and a list of the specific things you believed and got wrong.
- You start with
- A field you need to get functional in and have not studied. A paper, a course PDF or a few chapters in the folder help, but are not required.
- Time
- About an hour per pass, usually over several sessions — one cluster at a time.
- Your part
- Nearly all of it: answering, being wrong, reading the explanation. That ratio is the point.
What you'll end up with
- The prerequisites you actually lack, from the bottom layer up — and the ones you can safely pick up as you go.
- A study order: what to ask, in what sequence, each step a card you expand when you reach it.
- A written record of where you were wrong, grouped by what was being tested — including the confident mistakes.
It is a file in your folder, so the second evening starts where the first one stopped.
How to learn a new field with self-testing, step by step
This is the one workflow where the AI doing more would make the outcome worse — a field you had summarised for you is a field you cannot use next week.
Put the field in the middle, and pull your material in.
YouClick ＋ Topic and type the subject, or drag your PDFs and notes onto the canvas. Turn the knowledge base on so questions come from the real text, not the topic name.
Find out what is missing underneath first.
You ask · MindMax answersApply the prerequisites audit and rate each item yourself. The last card matters most: things widely treated as prerequisites that you can pick up as you go — the inflated entry price most people bounce off.
Turn the "not sure" items into questions.
You ask · MindMax answersAnything you rated as unsure can be quizzed straight from that card. Self-assessment is unreliable in both directions; this separates the gaps you have from the ones you only think you have.
Get an order, then study in that order.
You ask · MindMax answersApply the study-path skill: a sequence of what to ask, not a reading list. On the cards you work through, use the Feynman move — four empty slots and no answers, because the value is whether you can fill them.
Test yourself, and mark your confidence before you answer.
You ask · MindMax answersApply the self-test skill to a cluster you think you have finished. Multiple choice is marked on click; written answers are reviewed together. A wrong answer marked as sure is flagged as a blind spot.
Read the errors sideways, not one card at a time.
YouOn the card above a cluster, review the records across all its cards at once. The same misunderstanding on three cards is the thing worth fixing. The summary groups by what was tested and deliberately gives no score.
Fix the cluster, then move up.
You ask · MindMax answersRun the common-mistakes skill where your errors clustered: the wrong version, why it tempts, how to catch yourself. When a cluster has no gaps left, go up a level, and save what survived as a note.
Why it works here
It tests you instead of summarising
Prerequisites audit, study path, self-test and common mistakes are AI Skills you apply to a card, with the answers left for you to write.
Thinking models on the canvas →Confident mistakes are caught separately
Marking confidence before you answer turns a wrong-but-sure answer into a flagged blind spot, and the cluster review shows the same mistake across cards.
How answering and review work →Questions come from your material
With the knowledge base on, questions are drawn from the papers and chapters in your folder, not from a model's general impression of the subject.
Ask questions across a folder →Also used here: Your own folder, not a database · Guide: thinking on the canvas
Where it gets hard
- A topic with no material attached gets generic questions.
Without documents and the knowledge base on, you are testing yourself against a model's general impression of the subject. Attaching the source changes the output more than any other setting.
- Written answers are graded by a model, and grading can be wrong.
The correct answer is shown next to yours — read both and make the call. Occasionally the review fails outright; re-submit the card and your answers stay where they are.
- Confidence marked after the fact is worthless.
It is tempting to tick "I was sure" on the ones you got right. The blind-spot report is only as honest as that one habit.
- It will not test what your material never covered.
Questions come from the content in front of it. A gap missing from your sources is one it cannot show you — which is why step 2 exists, and why a second source is often worth adding.
- There is no review scheduler.
This is not a spaced-repetition system; nothing reminds you in three days. What you get is a written record of where you were wrong, in a file you can reopen.
- The wall grows faster than you can study it.
Expanding every branch produces a canvas nobody works through. One cluster per session, finished, beats a hundred cards nobody answers.