Could a national exam survive being leaked?
Every year more than twenty lakh students sit NEET-UG on one question paper, so the whole exam rests on a single sheet staying secret. Keeping any one artifact safe at that scale is genuinely hard, and nearly impossible without a proper system. I treated this as a design problem and Re:Neet proposes four concrete changes, reasoned toward one step at a time and open to argument.
One sheet carries too much
When a single identical form decides tens of lakhs of ranks, the whole exam has to travel a long physical chain of custody, and it stands or falls with that one sheet. In 2024 a leak was found to be localised and the Supreme Court saw no systemic breach, yet the 2026 cycle was still cancelled and re-run, because once trust in the single form slips, there is little to do but sit the whole exam again. That is less a story of blame than a sign that the system is asking a lot of one piece of paper.
Most of the loudest response has been about consequences. The education minister resigned in July 2026, and the anti-leak bill now before Parliament would raise sentences, have states designate fast-track courts, and put investigations on a two-month clock. All of it is worth having, and all of it starts after the paper is already out. The single sheet is still the single sheet.
So, one question. Could the exam keep measuring exactly what it measures today and still lose that single point of failure? The four changes below are the answer this proposal arrives at.
Computer based test
Deliver the exam as CBT across shifts with per-shift percentile normalization, the same machinery JEE already runs at scale, with question and answer order shuffled per candidate. The government has committed to CBT from 2027 and the machinery is borrowed and well understood, so the software is the easy part.
What nobody has published is how many invigilated computer seats India can field in a compressed window at this scale, and until someone counts them, that number sets the calendar. The other half of that bill is familiarisation: a candidate who has never sat at a computer against a clock needs practice on the real interface long before exam day.
Computers mean shifts, and that turns out to be okay
India cannot seat twenty lakh candidates at once, so the exam runs in shifts. At an illustrative three lakh concurrent seats, over twenty lakh candidates fit into about eight sessions across three or four days. The seats are a procurement problem. Comparing scores across those shifts is the design problem.
Different papers can still be fair
It is a fair worry: once two candidates don't share the question paper, it becomes hard to compare their scores. By using a sampling algorithm, we can assemble every question paper to one blueprint, and raw scores can become percentiles within each shift, so a rank reflects how ready you are rather than which paper you happened to draw.
The whole country designs the exam
Even a CBT is not immune to leaks if it has only 180 questions. A natural instinct is to grow the bank: 1,000 questions, or 10,000, or even 10 lakh. The design this project landed on is community-submitted questions, vetted and rewritten by experts, and then published on purpose. One month in advance, the raw question bank (not the final one) goes public for practice.
In today's age, if the bank leaks then it spreads like wildfire. So, we add an expert rewrite step after the submission step. We can also start with a small bank that only top institutes can submit questions to. Just like JEE Advanced. As time passes, we can allowlist more people and even leverage AI to help with the rewriting process.
Harder question formats
A paper made only of single-correct MCQs leaves every blind guess a one-in-four shot, and a penalty deep enough to stop that also punishes partial knowledge. So the paper itself changes, gradually over about five years: 160 questions, 40 each across Physics, Chemistry, Botany and Zoology, every subject carrying 25 single-correct MCQs, 10 multiple-correct MCQs, and 5 integer-answer numericals in Physics and Chemistry or diagram-based questions in Botany and Zoology. The formats are borrowed from JEE Advanced because they change the arithmetic: a blind shot at full marks falls from 1 in 4 to 1 in 15 on multiple-correct and to almost nothing in an integer field, so the exam rewards understanding structurally instead of taxing luck after the fact.
Optimal marking scheme
The bank size and the marking scheme are not arbitrary; the metrics to optimize are clear. Minimize the reward for memorizing, maximize the reward for understanding, so that students are conditioned to learn rather than to memorize and regurgitate.
So rather than defend those numbers on a slide, the design measures them on the live bank. Sweeping every marking scheme from +1/0 to +5/−4 on the same simulated cohort shows behaviour turns on a single line, a blind shot pays whenever the correct mark is more than three times the penalty, and today’s +4/−1 sits just on the side of that line where the one-in-four gamble is still worth taking. And if you picture a student’s recall as the slice of a public pool they can cram in a month, something sharper falls out: make the pool large enough and memorization stops ensuring a seat, while the rank list stays nearly as accurate as before, because understanding takes over the job. That is more convincing to watch than to be told, so please poke at the knobs yourself.
45% → 1%
Simulation result at a 2.5% seat rate: winning recall falls as the public pool grows from 10,000 to 250,000 questions.
50% → 65%
Simulation result at a 2.5% seat rate: winning understanding rises as the public pool grows from 10,000 to 250,000 questions.
2.5%
Simulation setup: a 2.5% seat rate, or 50,000 of 20 lakh candidates, as the public pool grows from 10,000 to 250,000 questions.
Follow along as the bank grows
Occasional notes on new questions, presets, and open practice releases. No spam.