Forecast from 19 papers · 2,249 questions

NIMCET 2027 Forecast

Every NIMCET paper from 2008 to 2026 was parsed, classified and back-tested to answer one question: what is most likely to be asked in June 2027, and how confident can we be about it?

19 papers parsed
2,249 questions extracted
300 recycled question pairs found
±1.4 questions — measured back-test error

The paper you will sit

The section split is not a guess. NIMCET changed it in 2023 — not 2026 — and has held it for four consecutive papers.

Mathematics
50
600 marks · unchanged since 2008
Analytical Ability & Logical Reasoning
40
480 marks · unchanged since 2008
Computer Awareness
20
240 marks · was 10 until 2022
doubled in 2023
General English
10
120 marks · was 20 until 2022
halved in 2023

Three findings that change what you study

6 → 1
Vectors is gone

Vector & 3D questions ran at 5–8 per paper for 18 years, then collapsed to a single question in 2026 when the revised syllabus dropped vector algebra. Those slots moved into Algebra (+3.6) and Sets & Functions (+1.8).

1 in 10
Questions are recycled

300 cross-year near-duplicates were detected, 44 of them near-verbatim. The gap peaks at 2 years and 4 years — which points 2027 straight at the 2025 and 2023 papers.

1.43 vs 1.81
Don't just copy last year

A slow exponential average of the whole history predicts each topic to ±1.43 questions. Predicting "same as last year" scores 1.81 — the worst of eleven methods tested.

Mathematics, topic by topic

Blue bar is the 2027 prediction, pale band is the 80% interval from back-test residuals, red tick is what 2026 actually asked.

2027 prediction
80% interval
2026 actual

Reasoning · Computer · English

Nineteen years, one grid

Questions per topic per paper. The last column is the 2027 prediction. 2012 is partly missing (59 of its questions exist only as images in the source PDF) and 2015 stops at question 90 — both are down-weighted, never scored against.

0 → 12+ questions

The recycling evidence

Detected by cosine similarity over question stems with every numeral masked, so a question re-asked with different numbers still matches. Whole multi-question puzzle sets get lifted intact.

QuestionAppeared inGap
Clock: when do the needles coincide between 3 and 4 o'clock2018 Q80/Q85 → 2022 Q874 yr
"Five houses lettered A–E in a row" puzzle set2009 Q61–65 → 2012 Q86–883 yr
"Family of six, two married couples" puzzle set2015 Q12–15 → 2017 Q92–942 yr
"Nine individuals, three committees" puzzle set2018 Q113 → 2020 Q21–222 yr
"A–G travelling in three vehicles" puzzle set2024 Q61–63 → 2025 Q1041 yr
Two-plant factory, defective item, Bayes2019 Q13 → 2023 Q1034 yr
Tangent to x = a·cos2t, y = 2√2·a·sin t2015 Q45 → 2023 Q1128 yr
"EXAMINATION coded as 56149512965"2015 Q35 → 2016 Q371 yr
Synonym: DEBACLE2018 Q24 → 2022 Q644 yr
Caterpillar climbing a 75-inch pole2017 Q105 → 2018 Q891 yr
Circle touching the X-axis and a circle at (0,3)2015 Q70 → 2023 Q578 yr
Syllogism: "All mangoes are golden in colour"2018 Q96 → 2022 Q714 yr
City literacy ratio (40% adults illiterate)2018 Q76 → 2020 Q102 yr
Bootstrap loader: first instruction location2008 Q110 → 2017 Q309 yr

Where repeated questions come from

Read this as a study order: a 2-year gap points at 2025, a 4-year gap at 2023. Those two papers should be the ones you can recognise on sight.

Does the method actually work?

Each method predicted year Y using only papers up to Y−1, scored against the real paper, for every target from 2016 to 2026. Error is in questions per topic.

MethodMean abs. errorWhat it does
EWMA α = 0.20  selected1.431Exponentially decaying weights over all 19 years
EWMA α = 0.351.436Faster decay, more weight on recent papers
Blend 45/30/251.464Last year + 3-year mean + all-time mean
Median of last 51.469Robust to a single odd year
Mean of last 51.493Flat recent window
Linear regression1.591Extrapolates the trend line
Last year only1.703Assume 2027 = 2026

Only two topics show a statistically significant trend across 2008–2025: Statistics is rising (Mann-Kendall τ = +0.57, p = 0.002) and Logical deduction puzzles are falling (τ = −0.44, p = 0.012). Everything else is stationary noise — which is exactly why mean-reverting methods beat trend-following ones here.

What to study, in order

Ranked by marks at stake against effort, at 12 marks per question.

PriorityAreaQMarksWhy it ranks here
Tier SNumber systems & Boolean logic10120Biggest single block in the paper; mechanical and fully learnable
Tier SLogical puzzles & conditional grouping10120Largest Reasoning block and the most-recycled question type
Tier SCalculus896Largest Math topic, stable for 19 years, zero syllabus risk
Tier SAlgebra & progressions896Absorbed the vector slots — 11 questions in 2026
Tier STrigonometry784Height & distance appears almost every single year
Tier AArchitecture, OS & memory hierarchy784Grew with the doubled Computer section
Tier ACoordinate & conic geometry672Circle first — it is the most repeated sub-form
Tier AArithmetic word problems672Sits in the Reasoning section, not Mathematics
Tier ASets, relations & functions560Rising; absorbed vector slots
Tier AProbability560Present in all 19 papers
Tier BSeries, coding-decoding, syllogism12144Low effort, high volume — drill patterns, not theory
Tier BBlood relations, seating, clocks, direction9108Fully mechanical once you have a notation
Tier BStatistics336The one significantly rising Math topic
Tier BGrammar & vocabulary560English is only 10 questions now — cap your time
SkipVector algebra, para jumbles, verbal analogy, linear programming00Removed from syllabus or absent for 8+ years

What this forecast cannot do

Read this before you trust any number above.

  • No individual question is predicted. The generated 2027 paper reproduces concept, structure and difficulty — not content. Expect concept-level hits, never question-level ones.
  • Small topics are noise. Permutation & combination, number theory and complex numbers sit at 0–3 questions, so their intervals run to zero. Do not over-read them.
  • Two papers are damaged. 2012 lost 59 questions to image-only pages and 2015 stops at Q90, so pre-2016 trends carry more uncertainty than the tables suggest.
  • A new syllabus revision would invalidate this. If NIMCET publishes a revised syllabus for 2027, the section split and the vectors call must both be re-checked first.
  • The recycling rate is a floor, not a ceiling. The detector only reads cleanly-extracted text stems — 1,400 of 2,249 questions — so the true repeat rate is higher than 1 in 10.