What actually moves your chances
Anyone can guess a percentage. This estimates the tradeoffs: how many points of admission probability a test score, an award, or applying early is actually worth at a given school — and how much of the decision nobody outside the admissions office can see.
Upload your Common App PDF to see your odds at any of the 403 schools. The file is read in your browser and never uploaded.
The finding that shapes everything else
Harvard's admissions model, published in the SFFA litigation, reaches a McFadden R² of 0.343 using only what an applicant knows about themselves. Add Harvard's own internal reader ratings — essays, recommendations, the reader's personal impression — and it jumps to 0.649.
Solving for what that gap implies gives an unobserved index with a standard deviation of 2.88 log-odds, larger than the logistic error term itself. In other words:
36% of what systematically drives an admissions decision is invisible to any model an applicant can feed.
That is not a shortcoming we plan to fix with more scraping. It is a property of the problem, it is measured rather than asserted, and it is why every number on this site is an interval and why a second, wider band sits next to it.
Schools
Tier A — coefficients estimated on individual applicant records for that school. Tier B — published aggregates plus a validated Common Data Set C7, anchoring a shape transported from elsewhere. Tier C — too thin; we refuse to give per-attribute effects rather than manufacture them.