Most people practise until they feel tired, without tracking what went wrong. So tomorrow's session is the same as today's, and the same mistakes come back.
A ledger fixes that. Each miss gets one row: the topic, the cause, a question, and an answer. The cause matters most. 'Failed two sum' tells you nothing, while 'didn't test empty input' describes a habit you can change.
AI helps twice here. It can read your mock and name the cause, and it can turn a miss into a clean flashcard. Spaced review then brings the card back just as you're about to forget.
- Cause: The habit behind the miss, in a few words: off by one, skipped clarifying, no test before moving on.
- Spaced repetition: Reviewing at growing gaps, such as 1, 3 and 7 days. The revise page on this site does it for you.
- Half-life: How long a miss takes to lose half its weight. Here, 7 days: old misses matter less unless they come back.
Steps:
- Log straight after each session. Two minutes, while you still remember.
- Name the cause, not the problem. Ask AI if you can't tell: 'What habit caused this?'
- Write one question per miss. One fact per card. Long cards don't get remembered.
- Start each day from the ledger. Practise your weakest topic first.
- Review the causes weekly. A repeated cause gets a rule, such as 'run it before I accept it'.
Spreadsheet first: A spreadsheet with four columns is enough. The code below shows what's worth calculating once you have rows: recent misses count more, and repeated causes stand out.
Prompt, Turn a miss into a card:
I got this wrong in practice:
[problem and what I did]
Reply with: 1) the habit that caused it, in under 6 words, 2) one flashcard question that tests the idea, not this exact problem, 3) an answer of at most 2 sentences. If you're unsure of a fact, say so.
Code, What the ledger tells you. Two views of the same rows: which topic is weak now, and which habit keeps coming back.
/** Recent misses count more: a miss loses half its weight every 7 days. */
export function weakestTopics(misses: Miss[], now: number, halfLifeDays = 7): [string, number][] {
const weight = new Map<string, number>();
for (const m of misses) {
const ageDays = Math.max(0, now - m.at) / DAY;
weight.set(m.topic, (weight.get(m.topic) ?? 0) + 0.5 ** (ageDays / halfLifeDays));
}
return [...weight].map(([t, w]): [string, number] => [t, Math.round(w * 100) / 100]).sort((a, b) => b[1] - a[1] || a[0].localeCompare(b[0]));
}
/** If the same cause keeps coming back, that habit is worth more than any single topic. */
export function repeatedCauses(misses: Miss[], atLeast = 2): [string, number][] {
const counts = new Map<string, number>();
for (const m of misses) counts.set(m.cause, (counts.get(m.cause) ?? 0) + 1);
return [...counts].filter(([, n]) => n >= atLeast).sort((a, b) => b[1] - a[1] || a[0].localeCompare(b[0]));
}
0.5 ** (ageDays / halfLifeDays): A miss loses half its weight every 7 days. A miss from today counts 1, and one from two weeks ago counts 0.25.repeatedCauses: Counts causes without any decay, since a habit that keeps showing up matters however old the first case is.
Code, Exporting to flashcards. Anki imports tab-separated text, one card per line.
/** Tab-separated lines that Anki can import. Tabs and line breaks inside a field would break a row. */
export function toAnki(misses: Miss[]): string {
const clean = (s: string) => s.replace(/[\t\r\n]+/g, " ").trim();
return misses.map((m) => `${clean(m.question)}\t${clean(m.answer)}\t${clean(m.topic)}`).join("\n");
}
replace(/[\t\r\n]+/g, " "): A stray tab or line break inside a field would split the row and scramble the card.Failure, Saving cards you never checked: Models make confident errors. Check each card before it goes in your deck.
Failure, Too many cards: A 300-card backlog gets abandoned. Add only what you actually missed.
- Why log the cause separately from the topic?
- Topics change every day, while habits repeat across topics. Fixing one habit improves every round.
- Why make old misses count less?
- If you stopped repeating a mistake, it's probably fixed. Recent misses show your weakness today.
Sources:
- How I used AI to practice for my Meta interview (Individual): https://dev.to/karuha/how-i-used-ai-to-practice-for-my-meta-interview-and-got-the-offer-543n