Who can win a government MBBS seat?
Who can win a government MBBS seat?
A model of NEET marks for JNV students, judged against the in-state government MBBS bar for each student's own state and category, with a forecast for this year's CoE and Nodal NEET cohorts
About this paper. This is an internal analysis by Avanti Fellows, shared for transparency. It has not been peer-reviewed. It was prepared with substantial help from AI tools (Anthropic’s Claude), which wrote much of the code and drafted much of the text under the direction of Avanti staff. The numbers are produced by code from the underlying data, and the sources and methods are documented so they can be checked, but errors are possible. Read the findings as preliminary, and contact Avanti Fellows before citing them.
The answer first
Section titled “The answer first”About 93 of this year’s 363 Class-12 JNV CoE and Nodal NEET students (26%) are expected to win a government MBBS seat in NEET 2027, and about 39 of 147 Class-11 students (26%) in NEET 2028, each scored at their own centre’s level of coaching. “Win” means scoring at or above the marks the last student admitted to a government MBBS college in their own state needed, for their category.
What the model does differently from earlier attempts:
- It predicts NEET marks, not a yes/no. The B1 bar (a government MBBS seat) runs from 177 marks in some North-East states to over 530 in Uttar Pradesh. A yes/no model learns the average bar of the year it was trained on and applies it everywhere. That is how an earlier version claimed 94 NEET CoEs.
- Each student is judged against their own state’s government bar, from the College DB (government colleges only; private, deemed, management and NRI seats are excluded at source). In-state seats are the door our students actually chase.
- Where seats did not fill, the seats bind. In a few North-East cells the bar is the qualifying gate. There the model caps the count at the government seats open to that state’s students: in-state colleges, plus the Central Pool and the regional institutes.
This model also feeds the Navodaya-wide potential in jnv-potential, which uses it for every NEET number.
1. The model
Section titled “1. The model”Model B (jee-main-prediction) predicts a JEE percentile. This model uses the same machinery on NEET raw marks, out of 720:
- the marks are modelled on a bounded (logit) scale, so a prediction cannot leave 0-720;
- a school effect lets small centres borrow strength from the rest;
- the spread between years is carried into every forecast as uncertainty;
- the CoE lift may vary with the board mark.
It is trained on 788 past CoE and Nodal NEET sitters (2024-26, first attempts, with a 10th board mark).
“Sat” means NEET marks are recorded, the corrected definition in the golden record (#430). The marks come from
production’s total_marks, which is what the B1 flag itself is computed from (data-assistant#420). The percentile
column total_score is never used.
| Factor | Effect (logit of marks / 720) | What it means |
|---|---|---|
| Board average, per mark (around 85) | +0.05 | Steady climb with marks |
| Studying at a CoE (vs Nodal) | +0.50 (s.e. 0.09) | The largest single effect |
| Girl (vs boy, same marks, category, centre) | −0.03 (s.e. 0.05) | No clear difference on NEET marks |
| Category, vs EWS: General / OBC / SC / ST | −0.11 / −0.16 / −0.25 / −0.17 | Marks; each category’s own bar is applied afterwards |
| Cohort 2025 / 2026, vs 2024 | −0.26 / +0.20 | Year-to-year paper difficulty, which feeds the uncertainty band |
| School (random effect) | spread 0.18 | Centre-to-centre differences |
| What none of these explain | spread 0.58 | Two alike students still land far apart |
What stands out: the CoE effect (+0.50) is the largest term, and unlike JEE there is no clear girls’ gap on NEET marks (−0.03). Students below a board average of 61, the training floor, are scored as if at it, the same rule as Model B.
2. Each student’s bar
Section titled “2. Each student’s bar”A student’s bar is the in-state government MBBS floor for their domicile and category: the College DB’s median
of the five loosest government colleges in that state, projected to 2026 (neet_dim_marks_matrix_2026). The
All-India-quota floor is used only where a state publishes none.
- MBBS, not BDS. In small states the government BDS floor collapses to the qualifying gate while MBBS is a real competition (Chandigarh: MBBS 532, BDS 213).
- Domicile is NCST’s
domicile_statewhere it names a state cleanly (it is free text), otherwise the JNV’s own state. The two agree for 98.9% of students where both can be read. - PwD sub-pools are ignored; the student’s base category is used.
The bars (marks out of 720). Italic cells sit at the qualifying gate, where government seats did not fill. A † marks a cell that is not VERIFIED in the College DB: INDICATIVE (official but thin or OCR-derived), or a state with no quota for that category, where the All-India floor stands in. 19 of the 23 gate cells are indicative, almost all in the North-East, so the seats-bind step below rests mostly on indicative data:
| State (domicile) | Gen | Gen-EWS | OBC | SC | ST |
|---|---|---|---|---|---|
| Andaman & Nicobar | 548 † | 542 † | 547 † | 476 † | 457 † |
| Andhra Pradesh | 512 | 508 | 489 | 457 | 422 |
| Arunachal Pradesh | 213 † | 213 † | 177 † | 177 † | 430 † |
| Assam | 510 | 504 | 385 | 437 | 398 |
| Bihar | 553 | 543 | 541 | 444 | 457 |
| Chandigarh | 532 | 542 | 491 | 421 | 489 |
| Chhattisgarh | 496 | 457 | 501 | 428 | 331 |
| Dadra & Nagar Haveli and Daman & Diu | 548 † | 542 † | 547 † | 476 † | 457 † |
| Delhi | 591 | 568 | 574 | 506 | 453 |
| Goa | 548 † | 542 † | 547 † | 476 † | 457 † |
| Gujarat | 502 | 500 | 497 | 457 | 330 |
| Haryana | 566 | 558 | 557 | 486 | 457 † |
| Himachal Pradesh | 510 | 542 † | 511 | 476 | 479 |
| Jammu & Kashmir | 526 | 352 | 430 | 403 | 388 |
| Jharkhand | 547 | 534 | 544 | 445 | 354 |
| Karnataka | 512 | 512 | 506 | 426 | 445 |
| Kerala | 570 | 539 | 543 | 438 | 360 |
| Ladakh | 438 † | 213 † | 177 † | 177 † | 177 † |
| Lakshadweep | 548 † | 542 † | 547 † | 476 † | 457 † |
| Madhya Pradesh | 519 | 514 | 517 | 439 | 363 |
| Maharashtra | 521 | 504 | 520 | 452 | 370 |
| Manipur | 463 † | 213 † | 422 † | 177 † | 177 † |
| Meghalaya | 498 † | 213 † | 177 † | 424 † | 257 † |
| Mizoram | 213 | 213 | 177 | 177 | 368 |
| Nagaland | 213 † | 213 † | 177 † | 177 † | 381 † |
| Odisha | 534 | 534 | 547 † | 467 | 360 |
| Puducherry | 534 | 346 | 522 | 469 | 288 |
| Punjab | 541 | 522 | 527 | 430 | 457 † |
| Rajasthan | 563 | 559 | 554 | 492 | 432 |
| Sikkim | 548 † | 542 † | 547 † | 476 † | 457 † |
| Tamil Nadu | 551 | 542 † | 521 | 442 | 426 |
| Telangana | 444 | 445 | 434 | 403 | 415 |
| Tripura | 213 † | 213 † | 430 † | 366 † | 333 † |
| Uttar Pradesh | 536 | 535 | 535 | 436 | 305 |
| Uttarakhand | 530 | 465 | 520 | 296 | 457 † |
| West Bengal | 526 | 479 | 523 | 452 | 291 |
3. When the seats bind
Section titled “3. When the seats bind”In the italic cells above, any qualifier could have had a seat, so the bar stops limiting anything. Coached students clear 177 marks easily, and without a limit the model would hand out more seats than exist. Meghalaya, for example, has no state-quota MBBS college of its own. So, in each state, the expected seats among students in those cells are capped at the government seats open to that state’s students:
- in-state colleges: 85% (the state quota) of the government MBBS intake in the state (NMC 2024-25, 49,368 seats nationally). Government-run societies that NMC files as “Society” (NIMSR Nagaland, UP’s Autonomous State Medical Colleges) are counted. Central institutes (AIIMS, JIPMER, NEIGRIHMS, RIMS, AFMC, BHU, AMU) are not, because they have no state quota;
- the Central Pool: MBBS seats MoHFW allots each year to nine “deficient” states and UTs;
- regional institutes: fixed state shares at RIMS Imphal and NEIGRIHMS Shillong.
| Domicile state | In-state government seats (85%) | Central Pool (status) | Regional institutes (status) | Total open to its students |
|---|---|---|---|---|
| Andaman & Nicobar | 97 | 0 (VERIFIED) | 0 (nan) | 97 |
| Andhra Pradesh | 2,771 | 0 (nan) | 0 (nan) | 2,771 |
| Arunachal Pradesh | 85 | 4 (VERIFIED) | 7 (VERIFIED) | 96 |
| Assam | 1,360 | 0 (nan) | 0 (nan) | 1,360 |
| Bihar | 1,292 | 0 (nan) | 0 (nan) | 1,292 |
| Chandigarh | 128 | 0 (nan) | 0 (nan) | 128 |
| Chhattisgarh | 1,216 | 0 (nan) | 0 (nan) | 1,216 |
| Dadra & Nagar Haveli and Daman & Diu | 150 | 0 (nan) | 0 (nan) | 150 |
| Delhi | 948 | 0 (nan) | 0 (nan) | 948 |
| Goa | 170 | 0 (nan) | 0 (nan) | 170 |
| Gujarat | 3,570 | 0 (nan) | 0 (nan) | 3,570 |
| Haryana | 710 | 0 (nan) | 0 (nan) | 710 |
| Himachal Pradesh | 612 | 0 (nan) | 0 (nan) | 612 |
| Jammu & Kashmir | 1,000 | 0 (nan) | 0 (nan) | 1,000 |
| Jharkhand | 578 | 0 (nan) | 0 (nan) | 578 |
| Karnataka | 3,230 | 0 (nan) | 0 (nan) | 3,230 |
| Kerala | 1,492 | 0 (nan) | 0 (nan) | 1,492 |
| Ladakh | 0 | 7 (SECONDARY) | 0 (nan) | 7 |
| Lakshadweep | 0 | 7 (SECONDARY) | 0 (nan) | 7 |
| Madhya Pradesh | 2,189 | 0 (nan) | 0 (nan) | 2,189 |
| Maharashtra | 4,888 | 0 (nan) | 0 (nan) | 4,888 |
| Manipur | 212 | 0 (nan) | 30 (SECONDARY) | 242 |
| Meghalaya | 0 | 0 (MISSING) | 27 (SECONDARY) | 27 |
| Mizoram | 85 | 0 (MISSING) | 7 (SECONDARY) | 92 |
| Nagaland | 85 | 42 (VERIFIED) | 20 (SECONDARY) | 147 |
| Odisha | 1,360 | 0 (nan) | 0 (nan) | 1,360 |
| Puducherry | 323 | 0 (nan) | 0 (nan) | 323 |
| Punjab | 638 | 0 (nan) | 0 (nan) | 638 |
| Rajasthan | 3,656 | 0 (nan) | 0 (nan) | 3,656 |
| Sikkim | 0 | 2 (SECONDARY) | 5 (SECONDARY) | 7 |
| Tamil Nadu | 4,505 | 0 (nan) | 0 (nan) | 4,505 |
| Telangana | 3,562 | 0 (nan) | 0 (nan) | 3,562 |
| Tripura | 149 | 0 (MISSING) | 13 (SECONDARY) | 162 |
| Uttar Pradesh | 4,590 | 0 (nan) | 0 (nan) | 4,590 |
| Uttarakhand | 574 | 0 (nan) | 0 (nan) | 574 |
| West Bengal | 3,146 | 0 (nan) | 0 (nan) | 3,146 |
The status says how solid each count is. VERIFIED means an official document states it. SECONDARY means only news
or coaching sources, used and labelled for now. MISSING means the state is on the beneficiary list but no count
was found, which is counted as 0 and so understates its cap. The cap uses all of a state’s seats, not each
category’s reserved share, so it is an upper bound. Sources are in sources/state_quota_extra_seats.yaml.
4. Validation
Section titled “4. Validation”Each cohort is predicted from the other two. Each student is judged against the B1 cutoff that actually applied to them: national cutoffs to 2025, state-wise by domicile from 2026. So the test is fair across the change.
| Held-out cohort | Group | n | Mean marks (/720) | Mean B1 cutoff that applied | Predicted B1 | Actual B1 |
|---|---|---|---|---|---|---|
| 2024 | CoE | 52 | 386 | 497 | 7.1 | 13 |
| 2024 | Nodal | 299 | 270 | 568 | 7.5 | 4 |
| 2025 | CoE | 40 | 332 | 362 | 25.5 | 13 |
| 2025 | Nodal | 215 | 228 | 421 | 29.7 | 7 |
| 2026 | CoE | 74 | 380 | 297 | 45.1 | 53 |
| 2026 | Nodal | 108 | 296 | 426 | 8.4 | 10 |
| Pooled | CoE | 166 | 77.7 | 79 | ||
| Pooled | Nodal | 622 | 45.5 | 21 |
- Pooled, CoE students are well calibrated: 77.7 predicted against 79 actual.
- Year by year it is not. 2024 is under-predicted and 2025 over-predicted by about two to one, because the year’s paper and cutoffs moved and there are only 100 B1 outcomes to learn from (17, 20 and 63 in 2024, 2025 and 2026).
- Nodal students are over-predicted: 45.5 against 21. Treat Nodal numbers as an upper bound.
- Read any single year or centre to about ±50%, and the pooled CoE total as the most reliable figure.
5. Forecast: this year’s CoE and Nodal NEET students
Section titled “5. Forecast: this year’s CoE and Nodal NEET students”The 2026-27 NEET students at JNV CoEs and Nodals (JNV, ENF and Dakshana programmes): 656 enrolled, 510 with a verified or recorded 10th board mark and a category, all scored as coached at their own centre’s level.
| Cohort | Students scored | Expected government MBBS seats | Rate |
|---|---|---|---|
| FY27 (NEET 2027) | 363 | 93.1 | 26% |
| FY28 (NEET 2028) | 147 | 38.5 | 26% |
By centre:
| Cohort | Programme | Centre | Students | Expected govt MBBS | Rate | Mean board | Mean bar |
|---|---|---|---|---|---|---|---|
| FY27 (NEET 2027) | Dakshana CoE | (all centres) | 112 | 45.6 | 41% | 92 | 456 |
| FY27 (NEET 2027) | JNV CoE | JNV Medak CoE | 40 | 15.9 | 40% | 87 | 425 |
| FY27 (NEET 2027) | JNV CoE | JNV Kohima CoE | 33 | 12.6 | 38% | 76 | 345 |
| FY27 (NEET 2027) | JNV CoE | JNV Kokrajhar CoE | 19 | 8.1 | 43% | 86 | 413 |
| FY27 (NEET 2027) | ENF CoE | (all centres) | 32 | 4.8 | 15% | 85 | 504 |
| FY27 (NEET 2027) | JNV CoE | JNV Chandigarh CoE | 6 | 2.0 | 33% | 92 | 485 |
| FY27 (NEET 2027) | JNV Nodal | JNV Chandrapur Nodal | 13 | 1.0 | 8% | 78 | 459 |
| FY27 (NEET 2027) | JNV Nodal | JNV Bharuch Nodal | 12 | 0.6 | 5% | 70 | 435 |
| FY27 (NEET 2027) | JNV Nodal | JNV Mandya Nodal | 13 | 0.5 | 4% | 82 | 481 |
| FY27 (NEET 2027) | JNV Nodal | JNV South Canara Nodal | 14 | 0.5 | 4% | 76 | 473 |
| FY27 (NEET 2027) | JNV Nodal | JNV Chamarajanagar Nodal | 14 | 0.4 | 3% | 71 | 474 |
| FY27 (NEET 2027) | JNV Nodal | JNV Bhagalpur Nodal | 14 | 0.4 | 3% | 75 | 523 |
| FY27 (NEET 2027) | JNV Nodal | JNV Adilabad Nodal | 9 | 0.3 | 3% | 70 | 421 |
| FY27 (NEET 2027) | JNV Nodal | JNV Udupi Nodal | 15 | 0.2 | 1% | 75 | 491 |
| FY27 (NEET 2027) | JNV Nodal | JNV Wardha Nodal | 9 | 0.2 | 2% | 75 | 464 |
| FY27 (NEET 2027) | JNV Nodal | JNV Vaishali Nodal | 8 | 0.1 | 1% | 69 | 492 |
| FY28 (NEET 2028) | JNV CoE | JNV Medak CoE | 35 | 9.6 | 27% | 84 | 452 |
| FY28 (NEET 2028) | JNV CoE | JNV Kohima CoE | 21 | 9.2 | 44% | 79 | 364 |
| FY28 (NEET 2028) | JNV CoE | JNV Kokrajhar CoE | 16 | 8.0 | 50% | 88 | 393 |
| FY28 (NEET 2028) | JNV CoE | JNV Chandigarh CoE | 15 | 6.4 | 43% | 81 | 378 |
| FY28 (NEET 2028) | JNV CoE | JNV South Canara CoE | 32 | 4.9 | 15% | 80 | 484 |
| FY28 (NEET 2028) | JNV Nodal | JNV Bharuch Nodal | 16 | 0.4 | 3% | 61 | 425 |
| FY28 (NEET 2028) | JNV Nodal | JNV Bhagalpur Nodal | 2 | 0.0 | 1% | 76 | 541 |
| FY28 (NEET 2028) | JNV Nodal | JNV South Canara Nodal | 4 | 0.0 | 0% | 67 | 466 |
| FY28 (NEET 2028) | JNV Nodal | JNV Adilabad Nodal | 1 | 0.0 | 1% | 67 | 403 |
| FY28 (NEET 2028) | JNV Nodal | JNV Udupi Nodal | 5 | 0.0 | 0% | 59 | 471 |
The Class-11 cohort (NEET 2028) is scored on fewer students because most of its 10th marks are still being linked. These are expected seats at CoE-level coaching against the 2026 bars. They will move with each year’s paper and cutoffs, by the margin Section 4 shows.
6. What this model does not do
Section titled “6. What this model does not do”- It does not model BDS, AYUSH or private seats. B1 here is a government MBBS seat only.
- It does not model the All-India quota as a second chance, except where a state has no floor of its own.
- It cannot score students without a 10th board mark or a category.
- Seat counts outside the state are partly secondary. Three Central Pool states have no count at all.
Related analyses
Section titled “Related analyses”| Analysis | Relationship |
|---|---|
| JEE-Main prediction (Model B) | The machinery this model reuses, on JEE percentiles |
| JNV potential | Applies this model to the whole Navodaya cohort to size NEET CoEs |
| Golden student record | The training rows; neet_appeared = marks recorded (#430) |
Replication data
Section titled “Replication data”All outputs are in the private bucket and open to Avanti accounts:
- validate_loco.csv: leave-one-cohort-out validation.
- govt_mbbs_bars.csv: the bar for every state and category.
- capacity_by_state.csv: seats open to each state’s students.
- forecast_by_centre.csv and forecast_by_cohort.csv: the forecast.
- per_student_deid.csv: one row per scored student, no ids or names.
- summary.json: every headline number.