How the NEET PG prediction works

Historical ranks come from the counselling allotment PDFs published by the Medical Counselling Committee. Those are lists of who was allotted what: rank, quota, college, course, seat category and candidate category, for each round. We parse them in full — no sampling, no manual transcription.

Currently loaded: 1,62,377 allotments across 2,169 colleges and 100 courses.

• 2025 Round 126,869 allotments
• 2025 Round 220,785 allotments
• 2025 Round 318,669 allotments
• 2025 Stray Vacancy2,782 allotments
• 2024 Round 124,649 allotments
• 2024 Round 216,181 allotments
• 2024 Round 313,991 allotments
• 2024 Stray Vacancy2,314 allotments
• 2023 Round 215,184 allotments
• 2022 Round 120,953 allotments

For each combination of college, course, quota and seat category we take the deepest rank that was actually allotted that seat. Two versions matter and we keep both: the round-one closing rank, which is what it took to get the seat immediately, and the final closing rank across every round including stray vacancy, which is what it took to get the seat at all. They are far apart — later rounds routinely reach several times deeper — so a predictor that quotes only one of them is answering a different question than the one you asked.

The 2026 college predictor uses recency-weighted historical cutoffs within the exact college, course, quota, reservation, PwD status and round. Each historical observation has weight 1 divided by its age in years: 2025 receives twice the weight of 2024. The 2025 chronological holdout improved both error metrics and interval coverage over carrying forward 2024 alone. This is an observed demand proxy, not candidate preference-list data. Other directory and state-reach summaries retain their earlier projection model.

On 21,549 exact matched combinations, the 2025 baseline had a mean absolute error of 5,762 ranks and median absolute error of 2,312 ranks. Weighted history reduced mean error to 5,669 ranks.

Rounds 1 and 2 have pre-2025 same-round training history. Their historical ranges covered71.8% of tested 2025 outcomes. Round 3 and stray lack earlier training pairs. This is below nominal 80% coverage; the ranges are descriptive, not a guarantee.

For 2026 we refit the range using the 80th percentile of absolute log errors on available consecutive-year history, rounded outward to hundreds. Unsupported rounds use the widest supported round range. Chance is an approximate logistic interpolation of this range. Medium or Low confidence reflects matching history and sample size; it is not a calibrated percentage of certainty.

The 2024 and 2025 MCC AIQ and DNB matrices record initial counselling availability. The 2026 NMC snapshot records sanctioned capacity across all quotas, and excludes pending increases and new-course approvals. They are different quantities. 2026 counselling-seat growth is unknown, so no seat adjustment is used. A seat-aware model could not be evaluated on the available historical matrices. See dataset coverage and official downloads.

Unresolved institution/course identities and duplicate matrix entries are withheld from capacity joins and reported. Missing data is never turned into a zero or used to declare a course new or removed. Detail sheets identify each available source and its page.

• This is MCC counselling only — All India Quota, deemed and central universities, and DNB. State counselling runs separately with its own cutoffs and is not covered.
• The seat matrix changes. New colleges and new seats appear each year, and a seat that did not exist last year cannot have a closing rank.
• Where a score is entered instead of a rank, the score-to-rank step is an approximation from published rank analyses — NBEMS publishes no official table. Enter your actual rank if you have it.
• A cutoff is the outcome of how thousands of other candidates filled their preferences. It can move for reasons no model can see.