Methods · Metacognition & self-regulation
E1
Error cataloguing
Build a record of your own mistakes that tells you what to study, in what order, and that gets more valuable every week instead of less.
When to use it
- Your study block is ending and you want tomorrow's next move written down, not left as a vague feeling.
- You got something wrong today and do not want the belief behind the error to vanish before you name it.
- You keep losing marks in the same vague area and need a record that can show a pattern across weeks.
What to ask AI
Type something like: “Here are this week's error-log entries. Don't summarise them. Tell me the single misunderstanding that explains the most of them, which entries it does not cover, and three problems that would test whether I've fixed it.”
When the middle field is hard, paste the problem and your working and ask: what rule was I applying that made this seem right?
Keep the hard part on your side: you write what you believed, in your own words, before you ask for clustering.
How it helps
Almost every other method here is something you do in a session. Error cataloguing is something you build across a term.
Three weeks before an exam, “what should I study?” is usually answered by feeling. “I'm bad at organic chemistry.” That is too vague to act on, and often wrong.
What you are bad at is usually two specific wrong rules inside the topic. Each instance looks like carelessness. Fifteen of them, written down, look like a diagnosis.
Errors are not random. They cluster around a small number of rules you applied correctly, except the rule is wrong. Naming that rule is what stops it.
AI helps with the weekly clustering. You supply the instances. Working memory does not span three weeks; the log does.
How to run it
- Use three fields only: what I did (specific), what I believed (the rule that made it seem right), what's true (the correct rule in your own words).
- Log at the moment of correction, not at the end of the session, or you will only remember being wrong, not what you believed.
- Log confident-and-wrong mistakes plus anything you have now got wrong twice. Skip one-off arithmetic slips unless they repeat.
- Keep entries under a minute. Aim for five to fifteen a week in one place.
- Once a week, paste the week's entries and ask for the underlying pattern, the entries it does not cover, and three re-test problems.
- Act on the pattern: do the re-test problems, re-derive the true rule without looking, and schedule a check a week later.
Example exchange
I've got three error-log entries from this week on derivatives. Can you help me find if there's one wrong rule behind them?
Paste each entry with what you did, what you believed, and what's true. I'll look for one candidate cause, you decide if it fits.
First: differentiated sin(3x) as cos(3x). I believed the derivative of sin is just cos. Second: same thing with cos(x²). Third: missed the product rule on xe^x. I somehow treated e^x like a constant.
The first two look like ignoring an inner function. The third may be a different rule. Don't force one story onto all three. Which two feel like the same mistake to you?
Yeah, the first two. I keep forgetting the chain rule when there's something inside.
Good. Leave entry three for a separate day. Attempt two chain-rule problems closed-book, then come back and tell me whether the old rule still shows up.
Copyable prompt
Here are this week's error log entries. Don't summarise. Tell me the single
misunderstanding that explains the most of them, which entries it doesn't cover,
and three problems that would test whether I've fixed it.
ENTRIES:
[paste WHAT I DID / WHAT I BELIEVED / WHAT'S TRUE for each]
The Tell
If your log is three weeks of neat entries and you still cannot name one misconception you have eliminated, the method has flipped.
You are keeping a diary of being wrong instead of a tool that stops the wrong rule.
Weekly pattern analysis plus a re-test is the work; logging alone is paperwork.
Principle evidence
Strength of the underlying learning idea, not a claim about AI products.
Strong as a learning idea. Recording errors by cause supports deliberate practice on weaknesses and metacognitive monitoring, learning from errors, not just counting them. The value is in clustering causes over time, not in having a tidy notebook.
AI delivery evidence
Whether an AI tutor delivers this method well is a separate question.
AI can help cluster entries and propose root causes, but that clustering quality is not established by tutoring trials here. Treat pattern claims as hypotheses you test with re-attempts, especially the middle field, which you should write yourself.