How the Admissions Predictor Works

What goes in

LSAT and GPA are the strongest signals and account for most of the outcome. The model also factors in whether you're a splitter (strong LSAT with a weak GPA, or vice versa), demographic categories schools track (URM, international, military, non-traditional), application timing, and softs captured in your profile. Splitter patterns vary by school, and the model learns each school separately.

The model also uses the target school's published 25/50/75 LSAT and GPA from the latest ABA 509, its acceptance rate, its class size, and year-over-year drift in those numbers. Your distance from a school's medians matters as much as your raw stats — a 170 LSAT means different things at Yale and at a regional school.

Timing matters because most admissions is rolling. September and February submissions convert at different rates with the same numbers. Mid-cycle, the model also tracks how many decisions a school has released, its wave pace, and silent periods that historically precede rejections or acceptances.

What comes out

For every applicant-school pair the predictor returns three probabilities that sum to 100%: acceptance, waitlist, and rejection. A 40% accept probability means that out of 100 past applicants with a profile like yours applying to a school like this one, roughly 40 were accepted. The model is calibrated: when it says 40%, the historical rate has been about 40%.

You'll see two versions of each probability. The unconditional number is your overall chance across the whole cycle. The conditional number answers "given that I'm still pending on day X, what are my odds now?" Early in the cycle they look similar. As time passes without a decision, the conditional number updates — sometimes in your favor (fast rejections have gone out, you're still alive), sometimes against you (the acceptance wave passed you by). The cumulative chart shows this resolution over time.

Known limits

Data volume is the main limit. Top-50 schools have thousands of reported cycles to learn from; a regional school might have a few hundred rows across all years combined. Predictions are still produced for those schools, but there are no per-prediction error bars — treat tight probabilities at sparsely-covered schools with more skepticism than a 65% at Georgetown.

Some schools practice yield protection, waitlisting or rejecting applicants whose numbers are well above their medians on the theory those applicants will go elsewhere. The model learns this pattern where it's consistent, but a year where a school changes its approach reads as noise until the data catches up.

The model doesn't read your application — no personal statement, no letters, no resume. An exceptional soft (Olympic medalist, published author, meaningful military service) can move outcomes beyond what the model anticipates. A character and fitness issue or a weak writing sample can sink an application the numbers say should succeed.

Early-cycle predictions are weaker than late-cycle ones. Before a school releases enough decisions to locate you against its historical pace, the timing features carry less weight and the model leans harder on stats alone. Predictions sharpen as the cycle unfolds.

When to trust it, when not to

Use the predictor to build a school list. Across 196 accredited law schools, it identifies the 15 or 20 worth a serious look given your numbers, separating reaches, targets, and safeties. It also works as a cheap second opinion when a consultant or forum regular labels a school a "safety."

Don't treat any single probability as the truth. A 20% accept probability at a reach still means roughly one in five applicants like you get in — worth applying if the payoff matters to you. A 75% accept probability at a target is not a reason to skip safeties. The value is in the distribution across your full list.

See also: reading a school profile and the glossary.