Programs ranked by empirical signal lift — how much sending a signal moves your interview probability, computed from real applicant data across 22 specialties.
Separate from program signals. Pick regions you’re flexible about. Programs in those regions weight your application favorably.
Lift = P(invite | signaled in your stratum) − P(invite | unsignaled in your stratum). When the signaled rate is lower than unsignaled in the source (small-N sample noise), we floor the lift at 0 — signaling cannot credibly hurt your odds, and the UI shouldn’t lie.
Stratum = applicant type (US MD / DO / IMG) × Step 2 CK band (<240 / 240-249 / 250-259 / 260-269 / 270+). When n < 10 in your exact stratum, we fall back to the program’s overall rollup so you still get a number, just less personalized.
Tier: Reach (matched-invitee mean is 8+ above your Step 2, or signal can’t rescue), Target (closer fit), Safety (baseline P(invite) > 50% — you’ll interview without a signal). The signal slider hides on Safety and on unrescuable Reach rows where worthSignaling is false.
Refine your estimate: each credential you set shifts P(invite) by its modeled effect for this specialty, from a ridge logistic regression on real applicant data. Credentials with no statistically measurable effect leave the number unchanged.
Caveats: lift is observational (signaled applicants may differ in unmeasured ways like geographic ties, mentorship, prior aways); programs with fewer than ~30 reports have wide CIs.
/match-plan takes your saved programs and computes P(invite), P(match), and tier per program for your specific stats — with primary + backup specialty support and signal-budget allocation.