Admissions Predictor Update: Same Look, New Engine
Our admissions predictor looks the same on the outside, but we rebuilt the algorithm behind it from scratch.
What changed
- Newer data. It's now trained on the three most recent cycles, including 2025. The old version was learning from an era that was meaningfully less competitive than the one you're applying into.
- Stats are measured against current medians. Your GPA and LSAT are now evaluated relative to each school's medians rather than in absolute terms, so median drift is built in instead of ignored.
- Your numbers are treated as a package, not a checklist. The old model added up LSAT, GPA, and URM status as independent boosts. The new one learns how those factors interact with each other and with the tier of school you're applying to — the same LSAT can mean very different things depending on the rest of your profile and where you're applying.
- Fewer extreme predictions. The old model was happy to tell you 95% at a T14 on numbers alone. It shouldn't have, and the new one won't.
What that means for you
Most people will see lower numbers, and the drop will be largest at reach schools and for profiles that were previously getting very high admit odds. Nothing about your candidacy changed — the old figures were just too generous.
How we know it's better
Tested against actual 2025 outcomes, log loss (a standard measure that punishes confident wrong answers) fell from 1.35 to 0.85, and predicted probabilities now track observed results closely across the whole range. For scale: applicants the old model gave a ~90% admit chance were actually admitted roughly half the time. That gap is largely gone.
The usual caveat
It still only sees GPA, LSAT, and URM status. Essays, softs, work experience, letters, and school-specific priorities are invisible to it. Treat the output as a well-calibrated starting point for building a school list, not a verdict.