Model Accuracy

This tool is a historical comparison, not a guarantee. To check how trustworthy it is, we ran a walk-forward backtest: for each past year we predicted using only the previous year's closing ranks, then compared against that year's actual admission cut-offs (2018–2025). No data from the year being tested was used to predict it. The honest headline:

• The cut-off estimate lands within ±20% of reality about half the time (JEE 49.5%, 12th-% 54.6%).
• The colour bands are odds, not promises — and they describe admissibility by the end of all counselling rounds; the early-round reality is tighter (both shown below).
• Accuracy drops sharply in years when cut-offs shift suddenly.

Honest limitations

• Cut-off drift is the main weakness — the model assumes next year ≈ last year, so a sudden scale shift (e.g. the +62% jump in 12th-% merit ranks in 2021, or ~+30% JEE inflation in 2024) makes it under- or over-shoot.
• End-of-counselling vs round 1. “Admitted” above is judged against the most lenient closing rank across all rounds. If you must commit in an early round, the real cut-off is tighter — which is why the JEE “round 1” column is much lower than “by end of rounds”. (The 12th-% round is a single round, so the two are close.)
• Coverage. ~78% of next year's individual seat pools existed the prior year and so can be predicted; because the rest are tiny/new, that's ~92% by seats. Brand-new branches or colleges can't be predicted at all.
• A measured improvement (queued): for the 12th-% route, averaging the last two years cuts the error ~27%; the JEE route is best left on the most recent year (its cut-offs drift upward).