Every question from twelve CDS sittings, broken down to sub-topic level and weighted towards what the recent papers are actually doing rather than what the syllabus says. Read the cut lines, then build the timetable.
Question counts are as they stand in the papers themselves, sitting after sitting. IMA / INA / AFA candidates write all three papers; OTA candidates write English and General Knowledge only.
Frequency tables average away the most useful thing here: question formats that were killed off, and ones that arrived from nowhere. Practising a dead format is the most expensive mistake available to you.
Word-meaning items — a pair of words and one option describing both, a word with two candidate meanings, List I / List II word-to-meaning matching — did not appear at all through 2023-II. Since 2024-I they run 20, 15, 15, 15, 29.
At 29 of 120 in 2026-I this is the largest single block in the English paper, and it is not the underlined-word synonym you have been drilling. It asks for a definition cold, with no sentence to lean on.
Cloze passages run 10, 0, 10, 10, 10, 0, 0, 10, 10, 0 across the ten sittings. When the format appears it is always a full ten-blank passage — never two or three questions.
Roughly every other paper carries one, and 2026-I did not. Ten marks is too much to walk into cold: keep one passage a week in rotation rather than preparing it as a block.
Sentence completion ran at 10 a paper for five straight sittings and has not appeared since 2023-II. Single-sentence fill in the blanks ran at 10–20 a paper and stopped dead after 2024-II. Ordering of words within a sentence was 10 in every paper until 2026-I, when it was zero for the first time in ten sittings. Comprehension passages went to zero in both 2025 papers and came back at only 5 in 2026-I.
That is up to 30 marks a paper reallocated. If your practice book predates 2024, it is drilling you on the wrong things.
Ordering of sentences (S1–S6 with four jumbled middles) is exactly 10 in all ten sittings. Parts of speech has never dipped below 10, and went to 15 in 2025-II.
Twenty marks, ten papers, no drift. Nothing else in this dataset is that dependable — start here and finish it early.
Trigonometry is the steadiest thing in the paper: 143 questions, 12–13 a sitting, never below 8, plus 2 more a paper in heights and distances. No other block in Mathematics holds that tightly.
It is the single highest-yield block in Mathematics and the one that rewards formula drill most directly.
Number system has gone from 13 a paper before 2024-II to 17.8 across the last four, driven by divisibility, remainders and primes. Mensuration has moved the other way, from 16 to 12.
Data interpretation has quietly gone: zero in eight of twelve sittings, including both of the last two. It is no longer worth a study block.
Geography, Current Affairs and History sit at roughly 22, 21 and 21 questions a paper and have barely moved in ten sittings — about 54% of the paper before you touch science.
Physics, Chemistry and Biology hold near-exactly 9–10 each: 29 marks of school-level science that is far more predictable than current affairs.
Polity ran at 17–19 questions in 2021-II and 2022-I. Across the last four sittings it averages 9. Economics has slid the same way, from 16 in 2024-I to an average of 8.5 since.
Moving in the other direction: Computer Science was absent until 2023-II and has delivered 5, 5 and 6 in the last three papers, from a very small and very learnable base.
Tick a row when you have finished it — the bar tracks the share of expected marks you have covered, and it is remembered on this device. Every sub-topic is ranked by expected questions in your next paper — a blend weighted 60% to the last four sittings and 40% to the full record, so recent shifts move the ranking without one odd paper dominating it. The brass rule is the cut line: everything above it accounts for 80% of the marks.
Change in questions per paper, last four sittings against everything before them. Positive means the recent papers want more of it than the older papers did.
Topic weights sitting by sitting. Darker means more questions. Read across a row to see whether a topic is dependable or bursty — a row that alternates between 0 and 10 is telling you not to bet a week of study on it.
Built from the ladder above, in the order that converts hours into marks fastest. Mathematics leads because it pays 1 mark per question against 0.83 elsewhere, and because its sub-topics are finite and finishable.Built from the ladder above, in the order that converts hours into marks fastest. With only two papers to sit, English carries the most movement and the most recoverable marks, so it leads.
Every wrong answer costs a third of a mark, so an attempt only pays if you are right more than 25% of the time. Set your realistic attempt count and strike rate for each paper — they are rarely the same — and see what the sitting is actually worth.
Every sub-topic in the ladder links through to real questions from these papers on SquadPrep, where you can drill them topic by topic with answers and explanations.
Source. The CDS question papers themselves — every question from twelve sittings, 2020-II to 2026-I. Mathematics is present for all twelve; English and General Knowledge for the ten from 2021-II onwards, so per-paper averages are computed against each subject's own sitting count, not a common denominator.
Expected questions per paper = 0.6 × the last-four-sitting average + 0.4 × the all-sitting average. The recency weight is set to let a genuine format change surface within two sittings while stopping a single unusual paper from rewriting the ranking.
Trend = last-four-sitting average minus the average of everything before it.
Reliability = the share of sittings in which the sub-topic appeared at all. High expected value with low reliability means a bursty topic — know it, but do not build a week around it.