The Revision Loop That Survives Three Weeks

Why what you revised in March is gone by May

Two things are working against you, and they pull in the same direction.

The first is ordinary forgetting. Material studied once and never retrieved decays fast, and it decays fastest in the first few days. That is not a failure of attention or intelligence. It is what memory does with anything it is not asked for again.

The second is worse, because it hides the first. When an explanation is clear and well made, following it feels like knowing it. Psychologists call this the fluency illusion: your brain reads the ease of understanding as evidence of mastery. So the better the video, the more confident you are, and the less likely you are to test yourself. Good teaching makes you overrate what you have learned.

The gap shows up at a predictable moment. You could follow every step at the time. Under a question, with nothing on screen, you cannot produce it. Recognition and recall are different abilities, and only one of them is on the table in an interview.

The loop, in four stages

What follows is not a study timetable. It is four things that have to happen to one piece of material, in this order, with gaps between them. Each stage takes minutes. The gaps do most of the work.

Stage 1: learn it once, properly

Before you open anything, write down the five or six concepts the topic is actually made of, in the order they build on each other. For a DCF: what it is solving, forecasting cash flow, the discount rate, terminal value, the bridge to equity value, and the sanity checks.

This takes two minutes and changes what watching does. Knowing the parts before you meet the mechanism frees up the attention you would otherwise spend working out where you are. It also turns a search into a lookup: you are no longer browsing for a topic, you are finding an explanation of concept four.

Then watch one explanation per concept rather than one long video per topic. A single teacher is rarely best at all six, and a ninety-minute lecture spends most of its length on things you did not come for.

Stage 2: test before you feel ready

The instinct is to revise until you feel confident, then test. Do it the other way round. Being tested on material produces substantially better retention than restudying it, and the gap grows the longer you wait before the real thing. In the review that rated ten common study techniques, practice testing and spacing were the only two rated high utility. Rereading and highlighting, which are what most people actually do, came out low.

The output of the check is not the score. It is the list of things you got wrong, which is the only part of the session worth keeping.

Stage 3: explain it back, then mark yourself

This is the stage most people skip, and it is the one that maps onto the interview.

Multiple choice tests recognition: four options are on screen and one of them is right. An interview tests production: nothing is on screen and you have to build the answer out loud, in order, without prompts. Passing the first tells you very little about the second.

So take the real question, answer it from memory with nothing open, and say it aloud if you can. Only then look at the points a strong answer contains, and mark your own against each one honestly. Half-said does not count.

The marking is not admin. Re-reading your own answer against a named list is a second pass of retrieval, and it converts a vague sense that it went alright into a specific list of what was missing. That list is what you carry forward.

Stage 4: the return trip

Two days later, answer the ones you got wrong. Cold, with no reminder of what the point was before you attempt it. Get it right and leave it for a week. Get it right again and leave it for three. Get it wrong at any stage and it goes back to two days.

Three properly spaced retrievals is usually enough for interview season. The intervals matter more than the total time: the same forty minutes split across three sessions retains far better than forty minutes in one, and the massed version feels more productive while you are doing it. That feeling is the thing to distrust.

Coming back before it is due is not a bonus. If it is still fresh, retrieving it is easy, and easy retrieval strengthens almost nothing. The gap is the mechanism.

Once you have three topics, mix them

Everything above practises one topic at a time. That is blocked practice, and it quietly does half the work for you: on the DCF page, the answer is about DCFs. You never have to decide what kind of question you are looking at.

An interview is the opposite. Nobody announces the topic. Working out which framework applies is a separate skill from executing it, and blocked practice never exercises it.

The fix is a mixed set: questions drawn across every topic you have studied, shuffled, with the topic hidden until you have answered. In a randomised trial across 54 classes and 787 students, students given interleaved practice scored 61% against 38% on a delayed test. That is a large effect for a change that costs nothing but the shuffle.

One rule when you build your own: never put two questions from the same topic next to each other. A shuffle that happens to cluster three DCF questions has quietly put you back in a block.

The prompts

Six prompts, one per job. They are the actual instructions, not sketches of them, and the marking one is the same instruction set behind the rubric marking in Primer. Square brackets are yours to fill in.

A word on the marking prompt in particular. The rules are ordered, and the order is doing real work: an assistant asked to mark an answer will drift into teaching it, crediting points you nearly made, and softening the verdict. Each numbered rule closes one of those doors. Delete them and you get a friendly conversation instead of a mark.

You are helping me build a syllabus for a technical interview topic, not teach it to me yet.

TOPIC: [e.g. DCF valuation / system design / how LLMs work]
ROLE I AM INTERVIEWING FOR: [e.g. investment banking summer analyst]

List the 5 to 7 concepts this topic is actually made of, in the order they build on each other. For each one give:
- a short title (under 8 words)
- one sentence saying what it is
- one sentence saying why an interviewer asks about it

Rules:
- Order matters. Concept 3 should need concepts 1 and 2.
- No concept that never comes up in an interview for this role, however interesting.
- Do not explain the concepts. I want the map, not the lesson.
- If two of your concepts overlap, merge them.

Write me 6 multiple-choice questions, one per concept, on the list below.

CONCEPTS:
[paste your concept list]

Rules for each question:
- Test whether I understand the mechanism, not whether I remember a definition.
- Four options. The three wrong ones must be things a half-prepared candidate would actually believe, not obvious filler.
- Give the answer and a two-sentence explanation of WHY it is right and why the most tempting wrong option is wrong.
- No question answerable by elimination alone.
- No "all of the above".

Give me the questions first, all six, with the answers in a separate block underneath so I can attempt them without seeing them.

I am preparing for an interview question. Write the marking rubric, not the answer.

QUESTION: [e.g. "Walk me through a DCF"]
ROLE: [e.g. private equity analyst]

Give me the 5 to 7 points a strong answer contains. Each point must be markable: I have to be able to read it and say yes or no about my own answer, without judgement calls.

Good: "Said you discount at WACC, and said why: the cash flow belongs to all capital providers."
Bad: "Showed good understanding of the discount rate."

Rules:
- Content only. Not delivery, not structure, not confidence.
- Include at least one point that separates a good answer from a complete one: the thing a strong candidate volunteers unprompted.
- Do not write the model answer. The rubric is the deliverable.

Mark my answer against the fixed list of points below. Do not teach me.

QUESTION: [paste it]

POINTS A STRONG ANSWER CONTAINS:
[paste your rubric]

MY ANSWER:
[paste what you wrote or said, typos and all]

Rules, in order of importance:
1. The list of points is complete and final. Give exactly one verdict per point, in the order given. Never add a point, never merge two, never drop one.
2. Judge only what my answer actually says. Do not credit a point because I obviously know it or because it follows from something else. Half-said is not covered.
3. Do not explain the concepts. If a point is missing, say what is missing in one sentence and stop. I will go back to the source for the explanation.
4. Quote my own words as the evidence for every verdict, so I can check your marking.
5. Be honest rather than encouraging. A hard 4 out of 6 with reasons beats a generous 6.
6. This is a spoken answer typed out. Ignore punctuation, filler and ordering. Mark the content.

Then give me: one sentence on the strongest thing in the answer, and the single change that would most improve it.

I got these points wrong two days ago. Ask me about them again, cold.

MISSED POINTS:
[paste the points you did not cover]

Rules:
- Ask them one at a time, as short open questions. Wait for my answer before the next one.
- Do not remind me what the point was before I answer. That is the whole exercise.
- After each answer, say covered or not covered, quoting me, and move on. No teaching.
- At the end, list which ones I still have not got. Those come back in another five days.

Draw me a 10-question set across all of the topics below, shuffled.

TOPICS:
[list them, e.g. DCF, LBO, three-statement modelling, accretion/dilution]

Rules:
- Do NOT tell me which topic a question belongs to until after I have answered it. Working that out is the exercise.
- Never put two questions from the same topic next to each other.
- Ask one at a time and wait for my answer.
- After each answer: right or wrong, which topic it was, and one sentence on why.
- At the end, tell me which topic I was weakest on and nothing else. I will go back to that one myself.

A worked week

Two topics, started three days apart, as it actually runs.

When the same thing keeps failing

If a concept comes back wrong twice, stop answering it. Twice usually means the explanation never landed, not that you forgot, and a third attempt at the same question just measures how well you remember getting it wrong.

Go back to the source for that one concept, watch it again properly, and then reset it to the start of the ladder. If it fails a third time after that, the problem is upstream: something the concept depends on is missing. A DCF terminal value that will not stick is often a growth-and-discount-rate problem, not a terminal value problem.

What this does not fix

This loop makes content stick. It has nothing to say about whether your answer is followable, whether you buried the recommendation, or whether you hedged three times before committing to a number.

A rubric can tell you the terminal growth assumption was missing. It cannot hear that your walkthrough was technically complete and impossible to follow, or that you sounded unsure of something you actually know. That gap closes with a person on the other side of the table, and it is worth being clear that no amount of self-marking substitutes for it.