A NEET-PG score by itself does not tell you much — what matters is the rank it turns into, and that depends on how the whole cohort performed that year. Kinase's NEET-PG Rank Predictor takes your score out of 800 and estimates your All India Rank using real published NBEMS data, not guesswork. Here is exactly how the model works, why it is built the way it is, and how accurate it actually is.
Key Takeaways
- ✓Built on real NBEMS score-vs-rank data from 2020–2023, tested against the actual 2025 result it never saw while training
- ✓Accurate to within ±165 ranks in the top 1,000, and ±301 ranks in the top 5,000
- ✓Uses isotonic regression on percentile, not raw rank averaging — a higher score can never produce a worse rank
- ✓One-time ₹49, unlocks unlimited re-checks for a full year — restore access with just your email
1Why Rank Is Not Just “Your Score, Looked Up on Last Year’s Table”
The obvious approach — find what rank your score got last year and use that — breaks down for one simple reason: the NEET-PG cohort keeps growing. About 1.61 lakh candidates appeared in 2020; by 2025 that was 2.30 lakh — roughly 7.8% growth a year. Rank 10,000 in a cohort of 1.6 lakh and rank 10,000 in a cohort of 2.3 lakh are not the same achievement; the second one beat a lot more people to get there.
So the predictor does not map score directly to rank. It first estimates your percentile — the fraction of candidates you finish ahead of, which stays comparable across years — and only then converts that percentile into a rank using the projected size of the target year's cohort (about 2.48 lakh projected for 2026).
2The Model: Isotonic Regression, Not a Simple Curve Fit
Under the hood, the score-to-percentile relationship is fitted with isotonic regression — a technique that produces a monotone (never-decreasing) step function by construction. In plain terms: it mathematically guarantees that a higher score can never map to a worse (higher-numbered) rank. That is not a rule checked after the fact; it is baked into how the curve is fitted, and it is covered by automated tests at every 1-mark and quarter-mark step across the entire score range.
Kinase Tip: a lot of “rank predictor” tools are really just last year's cut-off table with your score plugged in. Ask what happens to a tool's output for a cohort that grows 8% a year — if the answer is “nothing changes,” the tool is silently drifting stale.
3Two Data Problems Most Predictors Get Wrong
2020 was a different exam
NEET-PG 2020 was 300 questions worth 1200 marks. From 2021 onward it has been 200 questions worth 800 marks. A raw 2020 score of 600 is not comparable to a 600 today — the model rescales every 2020 score onto the current 800-mark axis before it ever touches the fit.
2024 has no published marks
NBEMS released percentile and rank for 2024 but never released raw scores. Without a real score to anchor to, that year cannot inform a score-to-rank model — using it anyway would mean inventing marks that were never published. It is excluded from training and kept only as separate reference data.
4How Accurate Is It, Really? (Tested on 2025)
The model trained only on 2020–2023 data, then was tested against the actual 2025 result — a year it never saw while fitting. That is a genuine out-of-sample test, not a score against its own training data.
| Actual rank band | Average error |
|---|---|
| Top 1,000 | ± 165 ranks |
| Top 5,000 | ± 301 ranks |
| Top 10,000 | ± 1,588 ranks |
| Overall (full range) | ± 6,181 ranks |
Accuracy is deliberately strongest at the top of the merit list, where a few hundred ranks change which branch and college are realistic. Lower down, absolute error grows but stays small relative to the rank itself.
5What You Actually Get
- Your estimated All India Rank plus a realistic likely-range, not a single false-precise number
- A full score → rank curve with your exact position marked
- What your score actually scored in each past NBEMS session, side by side
- Marks needed to reach headline targets — top 1k, 5k, 10k, 25k, 50k
- A live, fully anonymous community comparison — no name, email or IP is ever stored against a score check
Kinase Tip: re-run the predictor after every mock test. Since access lasts a full year for one ₹49 payment, tracking how your estimated rank moves as your score improves is the whole point — not a one-time lookup.
Quick FAQ
Does it account for category or state quota?
No. It predicts your All India Rank in the common merit list only. Category rank, state quota and seat allotment depend on reservation, domicile and counselling rounds, none of which are part of NBEMS's published score-and-rank data.
Is any of my personal data stored or shown to other users?
No individual candidate records are used anywhere in the model — it is built entirely from aggregated, publicly published NBEMS results. The optional community comparison logs only a score, an exam year and an optional state; it never stores or shows a name, email, phone or IP.
Know where you stand before results day
Enter your score, see your estimated AIR and likely range in seconds.
Try the NEET-PG Rank Predictor →