Across my 39 scored CAT mocks, the mean overall score rose from 64.9 in the first 19 to 71.5 in the last 20, a gain of 6.6 marks. The median moved from 64 to 72.5. That is real progress, and it is easy to oversell, so here is exactly what moved and what did not. Scores below 50 fell from five to one. Scores of 70 or more doubled, from six to twelve. My best score did not improve at all: it was a 96 in July, and the best of the second half is a 91. The bad days got less bad. The good days stayed where they were.
This is the third post from the same log. The section accuracy post shows which sections drove the change, and the percentile post explains why I trust raw scores more than percentiles here.
The two halves, side by side
| Measure | First 19 (10 May - 3 Aug) | Last 20 (8 Aug - 22 Oct) |
|---|---|---|
| Mean overall score | 64.9 | 71.5 |
| Median overall score | 64.0 | 72.5 |
| Spread (standard deviation) | 17.2 | 12.2 |
| Lowest / highest score | 40 / 96 | 39 / 91 |
| Mocks under 50 | 5 | 1 |
| Mocks of 70 or more | 6 | 12 |
| Mocks of 80 or more | 5 | 7 |
| Mean VARC / DILR / Quant | 20.1 / 17.6 / 27.2 | 23.9 / 23.1 / 24.6 |
| VARC spread | 10.4 | 5.4 |
The lowest score barely moved (40 against 39) and so did the highest (96 against 91). What changed is the middle: the spread narrowed from 17.2 to 12.2, and the count of mocks at 70 or better went from 6 of 19 to 12 of 20. That is a reliability gain, not a ceiling gain. In an exam where one bad section can cost a shortlist, a higher floor is worth having, but it is a different kind of progress from "I can now score 100".
It was a step, not a slope
Block means tell a different story from the halves. Splitting the log into blocks of eight mocks:
| Block | Dates | Mocks | Mean score | Spread |
|---|---|---|---|---|
| 1 | 10 May - 20 Jun | 8 | 63.5 | 14.9 |
| 2 | 27 Jun - 29 Jul | 8 | 63.4 | 15.7 |
| 3 | 30 Jul - 30 Aug | 8 | 75.5 | 16.4 |
| 4 | 7 Sep - 8 Oct | 8 | 69.5 | 7.5 |
| 5 | 11 - 22 Oct | 7 | 69.9 | 16.3 |
Blocks 1 and 2 are the same, 63.5 and 63.4: two months with no gain at all. Block 3 jumps by 12.1 marks. Then blocks 4 and 5 settle around 70, which is below the August peak. So the honest description is: flat, one step up, then flat again at a level a few marks under the peak. A trend line through all 39 points would look like a gentle slope and would describe none of the blocks. If you only look at a rolling average on a dashboard, you can miss that you have been on a plateau since September. This is the same pattern as in why mock scores plateau, and it is a better reason to log sections than to chase the total.
What I checked before believing it
- Is it just the past papers? The last 20 contain 10 real CAT papers and 10 coaching mocks. Their mean scores are 71.3 and 71.8, so the paper type is not what lifted the second half.
- Is it one outlier? The median rose by 8.5 marks (64 to 72.5), so a single big score is not driving the mean.
- Is the percentile telling the same story? It looks bigger: mocks at the 95th percentile or above went from 3 to 16. I distrust that number for the reasons in the percentile post, so I report it and do not build on it.
What I cannot tell you
- Why the step happened in August. It coincides with starting real past papers, but the second-half comparison above says paper type is not the lever. I have not isolated a cause, and I will not invent one.
- Whether the plateau is a plateau. Two blocks at about 70 is not enough to call it. It could be noise, or it could be where my current preparation tops out.
- What this means for the real CAT. These are practice scores. Nothing here predicts an official percentile.
What it changes about the next four weeks
- Stop reading the total. The total says "up". The sections say VARC and DILR up, Quant down. The decision is in the sections.
- Aim at the ceiling, not the floor. Fewer disasters is done. No mock has beaten 91 since a 94 on 1 August, and the only section where I am leaking marks I used to own is Quant.
- Keep the spread as a number I watch. A narrowing spread is the most dependable thing in this log, and a widening one would be the first sign of a bad week.
Where this is weak
- One candidate, one season, 39 attempts. This is a log, not a study.
- The halves are split at a convenient count, not at a principled boundary. A different split would shift the numbers a little. The block table is there so you can see the shape without trusting the split.
- Different papers have different difficulty, and I have not adjusted for it.
- Mean, spread and counts are descriptive. Nothing in this post is tested for statistical significance.
Every figure here is a query over the mock list in Karma Yogi: score, section scores and percentile per mock, ordered by date. If you log those four things after each paper, you can run the same split on your own data in a few lines.
End of essay
- Anish Guruvelli