An error log is one row per mistake, classified by the reason it happened - and the reason is almost never "I did not know this". That distinction is the entire value of the thing. A list of question numbers you got wrong is a list; it tells you nothing you can act on. A log that separates a concept gap from a misread stem from a set you should never have attempted tells you what to do next week.
Two numbers from my own 20-mock log make the case better than an argument would. Across three months my DILR score moved +0.16 marks per 30 days - an r-squared of 0.000 against time, no trend at all - while my overall score climbed from 41 to 91. And on the mock where I scored 2 marks in DILR, my own note from that week reads that 30 marks were available on that paper. The knowledge was there. What was missing was a record of why the marks kept escaping, and a question-number list could never have held it.
What an error log actually is
One row per question that went wrong. "Wrong" includes four things, and most people log only the first:
- You marked the wrong option.
- You skipped it and should not have.
- You spent far longer on it than it was worth, and got it right.
- You got it right without being able to justify why. This is the highest-value row in the log and almost nobody writes it, because the mock report shows a green tick and moves on.
That last one is worth pausing on. A lucky elimination and a solid solve look identical in every score report you will ever see. The only place the difference can be recorded is a log you write yourself, and it is precisely the question that will go wrong in the actual exam.
The eight categories
The taxonomy is the working part. Fewer than about six categories and everything collapses into "silly mistake", which is not a diagnosis. More than about ten and you will stop classifying at all.
| Category | What it means | What fixes it |
|---|---|---|
| Knowledge gap | You did not know the concept or formula | Study the topic. The only category that study time fixes |
| Concept misapplied | You knew it and used it in the wrong place | Practice on mixed sets, not on a chapter at a time |
| Calculation slip | Method right, arithmetic wrong | Mental-math drills and a slower final step |
| Misread the stem | You answered a question the paper did not ask | Re-read the question after solving, before marking |
| Option distortion | The option overstated or narrowed what the passage said | Verify each option against a specific line, not a memory |
| Selection error | You attempted a set or question you should have left | Timed triage practice. Not more solving |
| Time error | Right answer, far too many minutes | A hard per-question cap, enforced in practice |
| Right for the wrong reason | Correct answer you cannot justify | Treat exactly like a wrong answer. Re-solve it properly |
Two real rows from my own log, so this is not abstract.
28 June, VARC, score 5 out of 66. The passage discussed only Descartes; the option said "Descartes and other philosophers", and I took it. Category: option distortion. Not a knowledge gap, not carelessness in general - a specific, nameable failure to check the scope of an option against the text. My VARC across those five mocks ran 3, 36, 5, 39, 9. That is not a learning curve; it is a section where the same class of error kept recurring and averaged into invisibility.
30 July, DILR, score 2. Note: 30 marks were available. Category: selection error, twice over - committed to the wrong set early, and had no time left when I realised. This is the row that explains why more DILR practice did nothing for three months. I was fixing knowledge gaps I did not have.
The columns
Keep it to seven. An error log that takes fifteen minutes per mock survives; one that takes an hour does not.
| Column | Why it earns its place |
|---|---|
| Mock and question number | So you can find it again on pass three |
| Section | Section-level error mix is the pattern you are hunting |
| Topic or set type | Turns twenty scattered rows into one named weakness |
| Category (one of the eight) | The whole point. One category only - forcing a choice is what makes it useful |
| What I actually did | One line. Written in the first person, not as a rule |
| What the correct move was | One line. The thing you will re-read on pass two |
| Status: open / closed | Closed only after you re-solved it cold and got it right |
Resist adding a "difficulty" column and resist writing full solutions into it. The solution is in the mock report. What is not anywhere else is the sentence describing what you did instead.
The error log is what a revision pass is made of
Here is the part that most mock-analysis advice gets wrong. "Revise the mock" is usually taken to mean re-open the paper, and by pass three that means re-solving forty questions you got right the first time, with the answers already in memory. It feels productive and it is nearly worthless.
The unit of revision is the open rows in your error log, not the paper.
The schedule I run has eight rounds, R0 through R7, and each interval counts from when you finished the previous round rather than from the mock date - so slipping once pushes everything back by the same amount instead of dumping four overdue rounds on you at once.
| Round | Gap from previous round | What the pass consists of |
|---|---|---|
| R0 | - | Sitting the paper. Nothing else |
| R1 | 3 days | Build the log. Classify every error into one of the eight. Do not re-solve yet |
| R2 | 7 days | Re-solve the open rows cold, from the question only. Close the ones you get |
| R3 | 14 days | Re-solve whatever is still open. Anything failing twice is a topic, not a question |
| R4 | 21 days | Read the category mix, not the questions. Which category is growing? |
| R5-R7 | 30, 45, 60 days | Cross-mock only. Rows sharing a topic across several papers |
Why R1 is three days later and not the same evening. Straight after a mock you are still arguing with the score. A first pass done in that state classifies everything as "silly mistake" and finds the errors that flatter you. Three days is enough for the result to stop being personal and not so long that you have forgotten what you were thinking when you picked the option.
Why R2 re-solves rather than re-reads. Reading your own note and nodding is recognition, not recall. Cover the note, do the question from the stem, and see whether the fix survived a week. Rows that survive R2 are genuinely closed. Rows that do not are the ones worth your next month.
What the evidence supports, and where it stops
The case for putting days between passes rather than doing them all in one sitting is strong. Cepeda, Pashler, Vul, Wixted and Rohrer's 2006 meta-analysis in Psychological Bulletin pooled 14,811 participants across 254 studies and found spaced study produced 47.3% correct on the final test against 36.7% for massed study. That is a large, extremely well-evidenced effect and it is the reason the passes are days apart at all.
Now the limits, because they are real and they are usually hidden:
- The specific interval range you care about is the thinnest part of the evidence. In that paper's own table, the 2-to-7-day retention bin came out at p = .190 - not statistically significant, on nine studies and 435 people.
- The expanding shape of the schedule is not evidence-backed. The same meta-analysis compared expanding against fixed intervals and found 62.0% against 58.6%, p = .61. No meaningful difference. I run an expanding schedule for practical reasons - it spends less attention on a mock already squeezed dry - not because it is proven better.
- None of it involves CAT, or mocks, or error logs. It is verbal recall in laboratory and classroom settings. The transfer is an assumption I am making, and you should know I am making it.
I went through that paper's tables in detail in a separate post on spaced repetition, including the parts that argue against the product I built.
After ten mocks, the log becomes your syllabus
The compounding value is not per mock. It arrives when you can sort every row in the log by category and by topic across ten or twenty papers at once, and read what comes out.
If a third of your open rows are selection errors, no amount of topic revision will help you, and every hour spent on it is misallocated. That is exactly the mistake my own log documents: my QUANT rose steadily from 26.8 to 35.6 because QUANT responds to topic revision, so I kept doing topic revision, while DILR sat flat at an r-squared of 0.000 for three months because its errors were never knowledge errors in the first place.
Concretely, at R4 and beyond, stop looking at questions and look at counts:
- Knowledge gaps falling, other categories flat. Studying is working; execution is now the constraint.
- Selection errors steady across mocks. Triage is a distinct skill and you are not practising it. Sit sets under a hard clock and grade yourself on what you skipped, not on what you solved.
- Option distortions clustered in one question type. That is a reading habit, not a vocabulary problem. Verify options against a line of text before marking.
- Right-for-the-wrong-reason rows rising. You are getting faster and less certain. This is the one that precedes a crash.
Where this is weak
I cannot show you that the error log raised my score. That is the honest bottom line and it belongs here rather than in a footnote. I kept one, and the section I kept it hardest for is the section that did not move for three months. What the log did was tell me why it was not moving. Acting on that is a separate thing that I did too late.
The eight categories are mine, not a validated instrument. They came from classifying my own errors and noticing which distinctions changed what I did next. Somebody else's mix of mistakes may want a different cut - a JEE aspirant probably needs a separate row for sign and unit errors that CAT does not.
Classification is subjective and it drifts. The line between "concept misapplied" and "misread the stem" is genuinely blurry, and my own labels almost certainly got looser as the log grew. Counts across twenty mocks should be read as a rough shape, not a measurement.
The intervals are a design choice. As above: spacing is well evidenced, this particular sequence is not. If a fixed seven-day cadence is easier for you to sustain, the evidence does not say you are worse off, and a schedule you keep beats an optimal one you abandon.
One aspirant, 20 mocks, 90 days. Everything numeric here describes my own log and nothing wider.
Start smaller than you think
Do not build a fifteen-column log for the next mock. Take your last one, write the eight categories down the side of a page, and put a tally mark for every question that went wrong. That takes about ten minutes and it will already tell you something - if a quarter of your marks are going to one category, you now know what next week is for. The full row-per-question version is worth it from about mock five, when the cross-mock patterns start existing.
The related discipline, if you want the surrounding process, is how to analyse a CAT mock properly - the error log is the artefact that pass produces.
Karma Yogi stores each analysis session against the mock it belongs to, tagged with its revision round, so R2 knows what R1 left open and the next round's due date chains off the day you actually finished the last one rather than off the mock date. Free, and it works the same way for JEE and NEET papers.
End of essay
- Anish Guruvelli