The receipts

How we know.

Every claim on this site that sounds like a number was measured, not guessed. This page is where we show the work — in full, on this page, not behind an email you have to wait for.

We catalogued every sub-part of all 26 maths-optional papers from 2013 to 2025 — 804 in total — solving each one and verifying the mathematics with code. Every solution was classified against a map of the syllabus's testable areas.

Then we ran a blind test. We built our coverage map from 2013–2024 alone, held back the 2025 paper entirely, and asked: could a student prepared only on the older papers attempt the unseen one? The answer was yes — a strategic candidate could reach the full attemptable marks on both 2025 papers, choosing optimally among the optional questions. In thirteen years, only a single genuinely new topic ever appeared, and it sat in a question you could skip.

That is the whole idea behind our method: the maths-optional syllabus is finite, and here is the map that proves it.

Maths Optional

The 13-year coverage map.

Every one of 804 sub-parts, sorted into the 13 syllabus areas UPSC itself uses — the same 13 you'd plan a study calendar around. Bar length is how often an area was tested, not a promise about how it will be marked next year.

26 papers · 13 years (2013–2025) · 804 sub-parts classified.

Paper I 431 · 53.6%

Ordinary Differential Equations 88 · 10.9%
Linear Algebra +1 genuinely new topic in 13 years 81 · 10.1%
Calculus 77 · 9.6%
Analytic Geometry 64 · 8.0%
Dynamics & Statics 64 · 8.0%
Vector Analysis 57 · 7.1%

Paper II 373 · 46.4%

Numerical Analysis 73 · 9.1%
Mechanics & Fluid Dynamics 62 · 7.7%
Partial Differential Equations 56 · 7.0%
Algebra 50 · 6.2%
Real Analysis 48 · 6.0%
Complex Analysis 44 · 5.5%
Linear Programming 40 · 5.0%

The one exception: 2025-P1-Q4c-ii, the quotient space V/W (dim P₄/P₂) — matched no atom in the 2013–2024 map. 8 marks, in a skippable question.

13 years, 13 syllabus areas — 804 sub-parts, mapped.

Sub-part count = how often this area was tested, 2013–2025 (not a marks-weighted score share, not a prediction of future papers).

CSAT · Quantitative

Three families, most of the paper.

Ten years of CSAT quant (2016–2025), 444 questions, counted by hand by a math-and-statistics house — not estimated from a syllabus document. Every family, every count, visible; nothing merged away to make the chart tidier.

73%
  • Arithmetic & numeracy 128 · 28.8%
  • Number properties 100 · 22.5%
  • Logical-quant reasoning 96 · 21.6%
  • Counting & combinatorics 45 · 10.1%
  • Data sufficiency 35 · 7.9%
  • Statement validity 35 · 7.9%
  • Data interpretation 5 · 1.1%

Data interpretation: 5 of 444 — about 1%

Ten years of CSAT. Five DI questions. The pie-and-bar-chart drilling most aspirants over-invest in is the rarest thing on the paper.

Three question-families carry 73% of every CSAT quant mark.

10 years of CSAT (2016–2025), 444 quant questions. Uniform marking (2.5 marks each, no partial credit) means share of questions = share of marks, exactly.

CSAT · Comprehension & reasoning

Seven types. Not a vocabulary test.

The same discipline, applied to the half of CSAT most aspirants fear the most — because it feels unmeasurable. It isn't. Every one of 284 verbal-reasoning questions solved so far sorts into one of seven recurring types, and every trap we found was built by twisting the passage's own logic, never by hiding a hard word in it.

  • Which inference the passage forces 89 · 31%
  • Which unstated premise the argument needs 81 · 29%
  • The passage's load-bearing thesis 51 · 18%
  • What the passage literally says 21 · 7%
  • The stance the text commits the author to 19 · 7%
  • What the passage asserts (vs. merely raises) 9 · 3%
  • Free-standing deduction — no passageno passage 14 · 5%

Classification (what type of question this is) is settled for all 284. Official-answer verification is complete for 2021–2025 (145 items); for 2016–2020 (139 items) it rests on our own blind-derived reading or a recalled key, pending primary-source confirmation — those years are excluded from anything we score for money.

Seven question-types. Zero vocabulary tests. Ten years.

270 of these are genuine reading-comprehension questions — a passage, then a question that turns on reading it precisely. The other 14 are free-standing logic puzzles with no passage at all (statement-and-conclusion, does-it-follow). Both families are held to the same standard: the key must be provable — from the passage’s own lines, or from the stated premises — never from outside knowledge, and never from a word’s dictionary meaning.

Zero of 284 questions ask you to know a word’s meaning. Every trap is built by twisting the passage’s logic — not by hiding a hard word in it.

Want the coverage map in your inbox, to study from later? We'll send a copy.