ProctorSys → Research
Original researchCollege papers are written. Coaching papers are multiple choice.
Before building the way students answer, we wanted to know how exams are actually written — not how exam software assumes they are. So we read 101 real papers by hand. The answer split cleanly in two, and not in the direction either half would predict: of 61 university engineering papers, only 5 could be answered by choosing an option. Of 40 coaching-institute mock papers, all 40 could.
We have since counted the official CBSE Class XII sample papers too, and they are the awkward case the other two let a vendor avoid: half the questions can be auto-scored, but only 23% of the marks. For the entrance and recruitment exams — JEE, NEET, NDA, CDS, SSC — we counted nothing and cite their published schemes instead, because those are not in dispute.
What the 56 written papers actually asked for
They were not simply "long answer" questions. They were the ordinary working of an engineering course: derive an expression, calculate a value from given conditions, draw and label an arrangement. One asked the candidate to express quantitatively what changes when a plate is moved by 2.1 × 10⁻⁴ cm.
None of that can be typed comfortably, and none of it can be scored by comparing a selected option against a key. It has to be written, and it has to be read by a person.
The coaching papers were the exact opposite, and for a reason that is obvious once stated: a JEE or NEET mock exists to rehearse an entrance exam that is itself multiple choice. Four options, one key, negative marking. Nobody sets a derivation in a paper meant to simulate a test that never contains one.
Why this decides which software an institution can use
Most online examination platforms are built around a question that can be auto-scored: one stem, four options, one key. Everything downstream assumes it — the import format, the marking, the instant result.
For a coaching institute that assumption is correct, and such a platform works. For a university department, where ~92% of the papers we counted did not look like that, it can run a small minority of the exams. The rest either get rewritten into multiple choice — changing what is being assessed — or stay on paper. In practice they stay on paper, and the software gets used for practice tests rather than for the exams that matter.
This is why we treat the two as one product rather than two. An institution rarely sits on one side of the line forever: a college runs objective internals as well as written semesters, and a coaching centre that adds board-exam batches starts setting written papers. Software that only does half of it has to be replaced the moment that happens.
The practical test for any exam platform: take one of your own past papers, unmodified, and ask whether the system can run it as written.
If the honest answer is "after we convert the questions", the conversion is the real cost — and it recurs every term.
Then we looked at school boards, and found the hard case
The split above is clean because both halves are extremes. A school board paper is neither, and it turns out to be the most demanding paper of the lot — not because it is difficult to read, but because it is both formats on one page.
We counted the official CBSE Class XII sample papers for 2025–26, published by the board itself. These are CBSE's own statement of how the exam is built. Fourteen of them state both how many objective questions they contain and their totals, so those are the fourteen counted.
Half the questions, but 23% of the marks. Objective questions on these papers are worth exactly one mark each; the marks live in the two, three, four, five and six-mark answers that follow them. The consistency is more striking than the average — the objective share of marks lands between 15% and 26% on every single paper, from Political Science at the bottom to Physical Education and History at the top.
So a school cannot run its board-pattern exams on multiple-choice software: three-quarters of the marks are written. It cannot ignore multiple choice either: half the questions are. It needs both, on the same paper, marked in the same sitting — which is the case that the two extremes above let a vendor avoid.
Papers from cbseacademic.nic.in. Every figure is on the first page of the paper it came from, so any of it can be checked in a couple of minutes. Fifteen of the 29 papers we downloaded are not counted, and the reasons are in the limits below — including three in languages none of us reads well enough to count a paper in. The full working is available on request.
The entrance and recruitment exams need no counting
For JEE, NEET, NDA, CDS and SSC we have counted nothing, and we are not going to pretend otherwise: these exams publish their own format, and it is not in dispute. Counting papers to rediscover what the examining body states in its own notice would add a number without adding a fact.
- SSC — "Tier-I will consist of Objective Type, Multiple choice questions", in an examination that "consists of two tiers of Computer Based Examination". Scheme of Examination, ssc.gov.in
- NDA and CDS — "The papers in all the subjects will consist of objective type questions only. There are four alternatives for the answer to every question." UPSC examination notices
- NEET (UG) — 180 multiple-choice questions for 720 marks, four marks for a correct answer and minus one for a wrong one. NTA
- JEE (Main) — auto-scorable throughout, but worth a distinction the others do not need: Section A is multiple choice and Section B is answered by entering a numerical value. Scorable without a human, not chooseable from a list. NTA information bulletin
Which puts the four categories in one line. Entrance and recruitment exams are wholly objective by design. Coaching institutes set practice for them, so their papers match. University departments set derivations and diagrams. School boards set both at once. The format is decided by who is setting the paper — and only one of those four groups can be served by software that does a single format.
The limits of this study, stated plainly
Correction, 2 September 2026. An earlier version of this page said the Hindi and Sanskrit papers were excluded because CBSE publishes them "with non-Unicode fonts that no parser, including ours, can extract". That was wrong. Both Hindi papers extract cleanly — correct, readable Devanagari. The error was ours: our legibility check measured the proportion of ASCII characters, which marks every non-English document unreadable by construction. We were measuring our own test and reporting it as a fact about CBSE's files.
The Sanskrit paper is genuinely damaged — its font carries an incomplete character map, so 38% of its words come out with Latin letters spliced into them. The other two are simply in a language none of us reads well enough to count a paper in, which is the honest reason and the one now given above. The count of 14 is unaffected: these three were never in it.
This is a small, non-random sample and it should be read as an indication, not a national statistic. We would rather say so than have somebody discover it later:
- 101 papers counted, from two sources. The 61 university papers came from one engineering student's own folder; the 40 coaching papers from mock tests run by institutes. Papers too faint, cropped or incomplete to classify with confidence were set aside rather than guessed at. Not sampled across institutions, states or boards.
- Engineering and entrance-exam coaching only. A commerce, law or humanities paper may distribute differently, and we have not counted any. The split we found should not be assumed to hold for streams we did not read.
- "Readable" was a filter. Papers too damaged or poorly scanned to read were excluded, and their format is unknown.
- Counted by hand, by us. We had an interest in the answer, which is exactly why the method and the raw figures are stated rather than summarised.
- The CBSE figures are sample papers, not papers actually sat. They are the board's own model for the exam, which is what we set out to measure, but a real sitting could differ. One board, one year, one class — nothing here covers ICSE, any state board, or classes below XII.
- The CBSE count is a count of declarations. Where a paper states "Section A: 16 MCQs of 1 mark each", we took it at its word. We checked that against the questions themselves on one paper, not on all fourteen. An automatic count of the options printed on the page was tried first and abandoned: it disagreed with the papers' own declarations by up to 16 questions, and we would rather say that than publish the number it produced.
- 15 of the 29 CBSE papers we downloaded are not in the count. Three are in Hindi and Sanskrit, which nobody here reads well enough to count a paper in, and twelve do not state a total we could turn into a percentage. Their formats are not represented here.
- For JEE, NEET, NDA, CDS and SSC we counted nothing. Those rows are quotations from the examining bodies' own published schemes, linked in place, and should be read as citations rather than as findings of ours.
What it is enough to establish is narrow and, we think, sufficient: the format of an Indian exam paper is set by who wrote it, not by the subject. Written answers are not an edge case in university engineering exams, and multiple choice is not an edge case in coaching. Software that assumes either one universally is solving a different problem from the one half its market has.
What we built because of it
This finding is the reason ProctorSys does two things most exam platforms do not.
- It reads the paper you already printed. Upload the PDF or Word file as your department wrote it, and the questions, their numbering and their marks are recovered from it, then shown to you for correction. No retyping into a template.
- It accepts handwriting. The student writes on paper as always, photographs each sheet with their phone during a short supervised window after submitting, and the examiner marks those pages onscreen beside the marks box.
Neither is a feature we would have prioritised from a feature list. Both came from counting.
See it run on one of your own papers
The fastest way to judge any of this is to hand us a past paper and watch what the software makes of it. Twenty minutes, nothing to install.