Glossary

The words the simulator uses.

Agreement ( $ρ$ ): The average correlation between two agents' true utilities for the other side. 1: everyone ranks the other side the same way; 0: preferences are purely personal.
Applicants per position: Applicants divided by positions: how tight the market is. The NRMP's 2026 Main Match had about 1.08.
Blocking pair: An applicant and a program that both rank each other above what the match gave them. A stable match has none.
Certified applicant: An applicant with a non-empty rank order list; only they take part in the match.
Deferred acceptance: The matching algorithm: one side proposes down its lists, the other holds its best proposals so far and rejects the rest, until no one is rejected.
Fidelity: How well a view matches the truth: the correlation of true and observed utility, or, per agent, the rank correlation of its true and observed orderings.
Fit shock: The part of an applicant–program match that only an interview reveals.
Interview informativeness ( $κ$ ): How much of the pre-interview error an interview removes: 0 none, 1 all.
Match rate: The share of certified applicants who matched.
Noise ( $σ$ ): The size of the error in a pre-interview view, in utility SD units.
Post-interview view: What an agent believes about the other after the interview: the realised utility plus what is left of its pre-interview error.
Pre-interview view: What an agent believes about the other before interviews: the true utility plus error.
Realised utility: The true utility plus the fit shock an interview reveals.
Rank order list: An applicant's programs, or a program's applicants, in order of preference; the input of the match.
Regret: For a matched applicant, how much better the best program they interviewed at turned out to be than the one they matched to.
Rural hospitals theorem: Every stable match leaves the same applicants unmatched and fills each program to the same number of positions.
Seed: The number that fixes every random draw of a run; the same seed and parameters give the same results.
Signal: A preference signal an applicant sends with an application; programs that read signals favour signalled applications when inviting.
Stable match: A match without blocking pairs.
True utility: How much an agent really values the other; the model's ground truth.
Yield protection: A program's reluctance to invite applicants who look too strong for it, because they may go elsewhere.
z-scale: Values in standard deviations from the average: 0 is average, 1 one standard deviation above.