How the model works

Ranking residency programs is hard because you're juggling a dozen things that pull in different directions — salary, location, prestige, family, workload — and it's genuinely difficult to hold all of that in your head at once. This model doesn't decide for you. It gives your gut feeling a structure: it helps you uncover what you actually care about most (not what you think youshouldsay you care about), scores every program against that, and shows you the difference between the program you want and the program that's realistically within reach.

Think of it as a calculator for a decision you'd otherwise make on vibes. You still make the final call — the model just makes sure it's a defensible one.

Asking you to “rate salary from 1 to 10” doesn't work — everyone says 8. What works is forcing a choice: salary or being close to home? Prestige or wellbeing?

Think of a sports league: you don't know which team is better from what the fans say — you know from the results. Each head-to-head in onboarding (we call this the Decision Tiebreaker) is a game between two criteria, and the intensity slider is the score: winning 1-0 is not the same as winning 5-0.

From the 24 results, the Bradley-Terry model— the same kind of model used to rank chess players from head-to-head games — estimates each criterion's “strength” (the probability it beats any other) and turns it into your weight vector: 11 numbers that sum to 100%. The matchups are generated with a balanced round-robin — like a mini-league where every criterion plays 4-5 games and everyone eventually faces everyone, directly or through a common opponent — so no criterion gets an easier draw than another.

Every program is scored on an absolute scale, criterion by criterion, with a utility— a score from 0 to 1 that says how good that one aspect of the program is for you, independent of every other program. We don't compare programs to each other — we compare each one against a fixed yardstick. This whole approach (score each factor separately, then combine them with your weights) is a well-known decision technique called MAUT (Multi-Attribute Utility Theory). The curves aren't straight lines because reality isn't either:

• Salary (concave curve): the first dollars matter more than the last. Going from 45k to 55k changes your life; from 75k to 85k, less so — so the utility curve rises fast at first and then flattens out, like a curve that's bending downward. We also adjust for cost of living: 60k in Houston goes further than 70k in New York.
• Boards (sigmoid curve): a 90% pass rate and a 95% are nearly the same (both excellent); 80% is the frontier; below 75% is a red flag. An S-shaped curve captures that — flat at the extremes, steep in the middle — because “almost everything that matters happens in that middle zone.”
• Clinical purpose and climate (ideal point): here there's no “more is better.” You set your ideal (county hospital or private? cold or warm?) and the program scores by closeness to it, like a thermostat — too far in either direction scores worse.
• Compensation rules: a program far from home but with a direct flight recovers most of the hit (real distance is door-to-door time). And if your calling is maximum social impact, the penalty for a rough neighborhood softens: you know exactly what you're signing up for, and why.

Your desirability index— how much you'd want this program, all things considered — is the sum of each utility times your weight for that criterion, shown as 0-100%.

Wanting a program isn't enough; the program has to want you back. After each interview you fill in a checklist of objective signals (resident and faculty follow-up, IMG/out-of-state recruitment history, prior ties, spots, bilingual demand…) and out comes a viability factor — a number between 0.2 and 1.0 that estimates your real odds of matching there — which multiplies your desirability.

The uncomfortable part: fame works. AGAINSTyou. A Harvard-type program is harder to match at than a little-known one, so reputation penalizes viability (less so the stronger your application). Your dream program can drop in the strategic ranking even as it rises in desirability. That's not a bug — it's exactly the difference between what you want and what you can get, and you need to see both to decide well.

It's curiosity and context, not a decision signal: there is no “correct” set of preferences. While there are few real users, the average shown is a synthetic baseline — a stand-in average built from an archetypal profile (a Puerto Rico candidate: proximity, community, family), clearly labeled as such, rather than a real crowd. As real users come in, its weight fades away and the community average becomes genuinely yours.