My overall CAT mock score went up between the first half of my log and the second, and my Quant accuracy went down 10.9 points while it happened. It dropped from 78.3% to 67.4%. VARC accuracy stayed at about 50%. DILR accuracy rose from 65.1% to 71.7%. The total hid all three movements because VARC and DILR gains outweighed the Quant slip. A single overall number would never have shown me this, and a percentile would have shown me even less, as the previous post on this log explains.
The data is my own Karma Yogi log: 39 scored full mocks from 10 May to 22 October 2026. I split it into the first 19 (10 May to 3 August) and the last 20 (8 August to 22 October). Accuracy here means pooled correct answers divided by pooled attempted questions, not an average of per-paper percentages.
The section table
| Section | Period | Attempted | Correct | Wrong | Accuracy | Mean score |
|---|---|---|---|---|---|---|
| VARC | First 19 | 17.8 | 8.9 | 8.9 | 50.1% | 20.1 |
| VARC | Last 20 | 20.3 | 10.3 | 9.9 | 51.0% | 23.9 |
| DILR | First 19 | 9.9 | 6.5 | 3.5 | 65.1% | 17.6 |
| DILR | Last 20 | 11.3 | 8.1 | 3.2 | 71.7% | 23.1 |
| Quant | First 19 | 12.1 | 9.5 | 2.6 | 78.3% | 27.2 |
| Quant | Last 20 | 13.2 | 8.9 | 4.3 | 67.4% | 24.6 |
Values are per-paper means. "Wrong" is attempted minus correct, so it includes any answer that carried no penalty.
What each row says
- VARC: more attempts, same hit rate. I attempted 2.5 more questions per paper and got 1.4 more right, so accuracy stayed near 50% and I added about one extra wrong answer per paper (8.9 to 9.9). The score rose by 3.8 because the extra correct answers outnumbered the extra wrong ones. It also got steadier: the standard deviation of VARC score fell from 10.4 to 5.4. Half of everything I attempt in VARC is still wrong, though, and that is the number to attack next.
- DILR: both volume and accuracy improved. Attempts rose from 9.9 to 11.3 and accuracy from 65.1% to 71.7%. The mean score gained 5.5. This is the only section where doing more and doing it better moved together.
- Quant: attempts went up, accuracy went the wrong way. I attempted 13.2 questions against 12.1 and got fewer correct, 8.9 against 9.5. Wrong answers rose from 2.6 to 4.3 per paper. Mean Quant score fell 2.6 points. In the first half, Quant was my safest section. In the second, it became the one leaking marks.
Why a rising total hid this
The overall mean score rose from 64.9 to 71.5 between the two halves. VARC added 3.8 and DILR added 5.5, which is 9.3 marks of gain, against a 2.6 loss in Quant. Net the three and you get the 6.6-point headline, give or take rounding. The loss looks small because the others covered it. If I had tracked only the total, I would have seen steady progress and no reason to look at Quant.
There is a confound I should name. The last 20 contain 10 of my 12 real past CAT papers (the other two fall in the first 19), so the second half is a different mix of papers as well as a later date. I cannot separate a genuine Quant accuracy slide from "past-paper Quant sections were harder for me" using this data alone. That is why the fix below is a tracking habit, not a conclusion.
Which section moves the total most
Across all 39 mocks, the correlation between the overall score and each section score was 0.71 for VARC, 0.73 for DILR and 0.44 for Quant. In plain terms, on days my total was high, VARC and DILR were usually high too; Quant tracked it far less. Two cautions apply. The overall score is the sum of the three sections, so some correlation is built in. And 39 points is a small sample. I read it as "my good and bad days are mostly decided in VARC and DILR, and Quant is the steadier but now slipping section", not as a law about CAT.
What I log after every mock now
- Attempted and correct per section, not only score. Score hides whether a bad paper was too many guesses or too few attempts.
- Wrong answers per section, written next to the section. A rising wrong count with a flat or falling score is the Quant pattern above, and it shows up two mocks before the score does.
- Accuracy as pooled correct over pooled attempted across a window of five mocks. One paper's accuracy is noise. A five-mock window is the shortest one I trust.
This is the same logic as a per-mock error log, one level up: the error log tells you why a question went wrong, and the section table tells you which section to open it for first. For the attempts side of the same trade-off, see CAT negative marking strategy.
Where this is weak
- One candidate and one season. This is a log, not a study. A different person will have a different shape.
- The halves are not matched. The second half is later and has a different paper mix, as noted above.
- Per-section marks follow CAT's scheme of positive marking for correct answers and a deduction for wrong multiple-choice answers. Wrong answers on typed-in answers carry no deduction, and I did not log MCQ versus typed-in separately, so I cannot say how many of my wrong answers actually cost marks.
- Correlations on 39 points are suggestive only.
The per-section attempts, correct count, score and percentile are fields on every mock in Karma Yogi, which is what made splitting the log into halves a few lines of arithmetic rather than a weekend project.
End of essay
- Anish Guruvelli