Our process
Built from the real papers.
Verified three independent ways.
A wrong answer in a mock does more harm than no mock at all. It teaches you the wrong thing, and you pay for it. A mock pitched at the wrong level is not much better, because it gives you a score you cannot use. So every question we write sits where the official syllabus and the real past papers overlap. Its answer is settled independently three times. And its difficulty is checked against a real question until the two cannot be told apart. Below is the whole process, and how it applies to each exam we publish.
The pipeline
Nine steps, and a mock fails if any one of them fails.
- Step 1
We measure the real paper — we do not guess it.
The mock is built against measurements of real past papers, not an opinion about the exam.
Read the detailHide the detail
Every past paper we could obtain has been digitised question by question and then measured: how many marks each topic actually carries, how long a question typically runs, how the correct answers spread across the options, and how hard the questions really are. Those measurements are what the mock is built against — and every new paper the exam board releases is measured and folded in before the next mock is written, so the target keeps moving with the exam.
- Step 2
A blueprint is written before a single question exists.
Every slot in the mock is fixed in advance — topic, question type and marks.
Read the detailHide the detail
The blueprint fixes each slot before anyone starts writing: which topic it comes from, which question type it is, how many marks it carries, and how the correct answers should be spread across the options so that no one letter is the answer far more often than it should be. We match the spread that real papers actually show, rather than making it perfectly even — a perfectly even spread is itself a sign of a machine-made mock.
- Step 3
Every question is written fresh — from the overlap of two things.
Where the official syllabus and the real past papers meet. Nothing is copied from either.
Read the detailHide the detail
Each question fills one specific blueprint slot, and it is written from the intersection of the official syllabus and what the real papers have actually tested. That overlap is the whole point. The syllabus tells us what the examiner is allowed to ask; the past papers tell us what the examiner actually asks, and how. A question drawn from the syllabus alone can be perfectly valid and still be nothing like this exam; a question drawn from the past papers alone is just a reworded PYQ. Both together is what makes a mock feel like the real one. The answer is worked out and recorded at the moment the question is written. Past-year papers themselves are published separately in our PYQ material, where they belong — a mock test is not a place to recycle them.
- Step 4
A nothing-copied check, in both directions.
Checked against the whole past-paper bank, and against the other questions in the same mock.
Read the detailHide the detail
Each new question is compared by machine against every past-paper question we hold, and against the other questions in the same mock — which is the half most people miss. Two questions written separately can end up testing the identical trick, and then you have paid for the same question twice. Anything too close to a past question, or to its own siblings, is rewritten before it goes any further.
- Step 5
Two independent AI solvers, from two different companies, solve the mock blind.
Neither solver is shown the intended answer, and they come from two different companies on purpose.
Read the detailHide the detail
Each solver receives only the question, exactly as you will see it, and its access is restricted so it cannot reach our answer key. The two solvers are deliberately built by different AI companies, because two copies of the same system make the same mistakes: they would agree with each other while both were wrong. An answer is accepted only when the author’s answer and both blind solvers agree separately. Where they disagree, the question is held back and settled by hand — and those disagreements are the most useful signal we get, because they usually mean the question itself was ambiguous, which is a defect rather than a hard question.
An answer is accepted only when three independent routes reach it separately — and two of the three never see the intended answer. - Step 6
Numerical answers are re-computed by machine.
Recomputed by a computer algebra program, using a different method from the one used to write the question.
Read the detailHide the detail
For every question where you type a number instead of choosing an option, the answer is recomputed from scratch by a computer algebra program, deliberately by a different route than the author used. Reasoning can talk itself into a wrong number and then defend it. Two independent computations landing on the same value cannot.
- Step 7
A human reads the whole mock in one sitting.
An IIT Roorkee mathematics post-graduate signs the mock off question by question.
Read the detailHide the detail
Every question is read exactly as it will be printed for you, and four things are checked by hand: that there is exactly one defensible answer, that the wording cannot be honestly misread, that no two questions test the same idea twice, and that no option accidentally gives away another question’s answer. Sign-off is recorded one question at a time. A question that is not signed off does not ship.
- Step 8
Difficulty is measured against the real paper, not asserted.
Judged blind against a real past-paper question of the same type and the same marks — and released only when the two cannot be told apart.
Read the detailHide the detail
Anyone can say their mock feels like the real exam. We check it. Each of our questions is paired with a real past-paper question of the same type and the same marks, and an independent judge — which is not told which of the two is ours — is asked which one would take a well-prepared student longer to answer correctly. What we are looking for is a draw. If our question is picked far more often it is too hard; if it is picked far less often it is too easy; either way it goes back to be rewritten. We release when the judge cannot reliably separate the two, because that is what "like the real exam" has to mean if it is going to mean anything. And it is what makes your score worth reading: a mock exists to tell you where you actually stand, and a mock pitched above or below the real one gives you a number you cannot use. Pairs are always matched on question type, because a multiple-choice question can be attacked by elimination and a typed numerical answer cannot, so comparing across types would prove nothing.
Our question and a real past-paper question, judged blind. We release when the judge cannot tell which one is ours. - Step 9
Automated release checks, and any single failure blocks release.
Formatting, marking, tolerances and a test-render of every formula. All must pass.
Read the detailHide the detail
The finished mock runs through a fixed set of automated checks before it can go live: mark totals and negative marking, answer formats, the accepted tolerance on every numerical answer, figure integrity, and a test-render of every single mathematical formula on a real maths engine so that nothing reaches your screen as broken symbols. All of them must pass — not all but one. The set is different for each exam, because the exams are different, and it grows every time we find a new way a mock can go wrong.
Why it matters
What all of this buys you.
The answer key is right.
Three independent agreements and a machine recheck stand between an author’s slip and your score.
Every question is one this exam could actually ask.
Written where the official syllabus and the real past papers overlap — never from the syllabus alone.
It sits exactly where the real exam sits.
Measured question by question against real past-paper questions, and released only when a blind judge cannot tell them apart.
Our mock is shaped like the real paper.
Section weights, question counts, marking and duration all come from measurements of the actual papers.
You have not seen these questions before.
Every question is original and checked against the full past-paper bank.
Nothing renders broken.
Every formula is test-rendered before release, so no question arrives as a row of question marks.
Being straight with you
What AI does not do here.
We use AI for the slow work. It drafts questions, solves them, compares thousands of questions with each other, and rechecks the arithmetic.
But it never decides anything. An answer goes into a mock only when two AI solvers from different companies reach it on their own. Wherever a number is involved, a program rechecks it as well.
Then a person reads every question in full and signs it off. And if even one automated check is failing, the mock is not released.
AI does the work. A person decides, and puts their name to it.
Past-year papers
Real papers, typed exactly — and solved before we look at the key.
A past-year paper is the real examination, so there is nothing to invent and everything to get exactly right. It is the opposite job from building a mock: the questions belong to the exam board, and our work is to reproduce them without a single slip and then prove that the solution beside each one is correct. This is how every past paper reaches the site.
The official paper is typed in once, and then locked.
Every question, option, table and formula goes in exactly once, in the order the paper prints it.
Read the detailHide the detail
The paper is transcribed question by question, with its options in the printed order and every figure cropped out as an image. That single transcription is then frozen: it becomes read-only, and every later stage works from it rather than from the original. Anything that has to change afterwards is a logged amendment that sends that question back through the checks — never a quiet edit.
The typed paper is checked against the original, picture beside picture.
Re-reading would only repeat our own misreading, so we compare images instead.
Read the detailHide the detail
We render the typed paper and set each question beside the original page, then ask whether the two say the same thing. Comparing two pictures is a different kind of check from reading, and it catches the slips reading misses: a dropped minus sign, an exponent that moved, one-half where the paper printed one-quarter, options quietly reordered. Every figure crop is looked at before release, and the paper is scanned for duplicated or contaminated questions. If any of it fails, solving does not start.
Every question is solved before the official key is opened.
The solvers see only the question, never an answer — so a solution cannot be bent to fit a result we already knew.
Read the detailHide the detail
Solvers work from the locked paper, which provably contains no answers, and the official key stays out of reach until they have finished. Questions with several correct options get the most careful treatment, because those are the easiest to get subtly wrong. Any question whose answer is a typed number is also recomputed by a computer algebra program, because a number has no list of options to give a wrong answer away.
Then the official key is opened — and every disagreement is investigated.
A published key is evidence, not authority.
Read the detailHide the detail
Our blind answers are laid against the official key. Where they differ, fresh solvers who are not told which way the disagreement points solve the question again, and a stubborn case is settled by computation, never by a vote. Some disagreements are our own slip, and are fixed. Others turn out to be the official key, or the printed question itself, being wrong. Either way we find out why before anything is published — an answer that lands on no printed option is escalated, never smoothed over.
The worked solution is written only for an answer that is settled.
The idea being tested, the standard method, and a shorter route where one honestly exists.
Read the detailHide the detail
Only now is the full solution written, so nobody polishes an explanation for an answer that later changes. Then two things are checked by machine: every answer line must equal the settled answer, and every question and option must be identical, character for character, to the locked transcription — so writing a solution can never alter the question it sits under.
Automated delivery checks, and any single failure blocks release.
Answer formats, marks, and a test-render of every formula.
Read the detailHide the detail
The finished files run through a fixed set of checks before they can go anywhere: the file the site imports is validated row by row, mark totals and answer formats are confirmed, and every mathematical formula is test-rendered on a real maths engine so nothing reaches your screen as broken symbols. A check we have never seen fail is not trusted — each one is first given a deliberately broken paper, to prove it can catch the fault it exists for.
A person reads it and decides when it goes live.
Nothing is published by the pipeline on its own.
Read the detailHide the detail
The finished paper is handed to a person, who goes through it and gives the go-ahead. And the checking does not stop at release: when we find a slip in something already live, we correct the paper and replace it.
Each past paper reaches you two ways.
A timed online test
Attempt the paper under real exam conditions on screen, and get your score and analytics the moment you submit.
A fully solved PDF
Every question worked through step by step, with the concept behind it and a shorter method where there is one.
Exam by exam
How this applies to each exam.
The nine steps above are the same for every exam. What changes is the paper being imitated — its sections, its question types, and where it places figures — because a mock is only worth attempting if it matches the paper you will actually sit.
IIT JEE
The pipeline is being adapted to JEE’s Physics and Chemistry sections. Until a JEE mock can clear the same checks as the mocks above, we are not publishing one. We would rather be late than sell you a mock we cannot stand behind.
IIT JAM — Mathematics
- The real paper
- One paper in three question types — multiple choice, multiple select where more than one option can be correct, and questions where you type a numerical answer with no options at all.
- What we measured
- Every past paper we hold, question by question — including how often each question type appears, and the shape of the answers themselves.
- Figures
- None. Across every real question we measured, not one carried a figure, so ours carry none either.
IIT JAM Mathematical Statistics: past-year papers are published; our mocks are in build.
Browse these papersGATE — Mathematics
- The real paper
- General Aptitude first, then the Mathematics section — mixing multiple choice, multiple select and typed numerical answers across one-mark and two-mark bands.
- What we measured
- Every past paper we hold, question by question — including which question types appear in which mark band, and in what order.
- Figures
- General Aptitude only, matching where the real paper places them.
CSIR NET — Mathematical Sciences
- The real paper
- Three parts — a general aptitude Part A, then Part B and Part C in Mathematical Sciences. Each part carries its own marks per question, its own negative marking, and its own limit on how many questions you may attempt.
- What we measured
- Every past paper we hold, question by question — topic weighting, question length, how the answers spread across the options, and difficulty.
- Figures
- Part A only. The real paper puts figure-based questions there and nowhere else, so ours do the same.