Parameter reference
Every parameter of the simulation, its range and default.
Every parameter, its meaning, its range and its default, generated from the parameter schema the app itself uses. Parameters marked Not used yet belong to features that are not implemented (replicates and interview dates); they are saved but have no effect. Applicants × programs is limited to 1,000,000 pairs per run for now.
Run
| Parameter | Meaning | Range | Default |
|---|---|---|---|
| Random seed | Runs with the same seed and parameters give identical results, and changing one parameter then changes only what depends on it. Leave blank to draw a new seed for every run (the seed used is stored with the results). | 0–9.22337e+18 | blank |
| Replicates Not used yet | Independent repeats of the random stages, for averages and uncertainty bands. | 1–1000 | 1 |
| New population per replicate Not used yet | Off: every replicate keeps the same applicants, programs and true preferences, so only process noise varies. On: every replicate draws a new market. | off | |
| Confidence level Not used yet | Confidence level of the intervals reported over replicates. | 0.5–0.99 | 0.95 |
Market
| Parameter | Meaning | Range | Default |
|---|---|---|---|
| Applicants | Number of applicants in the market. | 10–60000 | 1000 |
| Applicants per position | Market tightness: applicants divided by positions, which sets the number of positions (NRMP 2026 Main Match: 48,050 applicants for 44,344 positions, about 1.08). | 0.5–3 | 1.08 |
| Programs | Number of programs. Leave blank to derive it from the positions and the mean program size. | 1–7000 | blank |
| Program sizes | How the positions are spread over programs; every program has at least one. Fixed: as equal as possible. Lognormal: right-skewed, like real programs. Shifted multinomial: random, close to 1 + Poisson. | Fixed, Lognormal, Shifted multinomial | Lognormal |
| Mean positions per program | Average program size, used to derive the number of programs when it is blank (NRMP 2026: 44,344 positions in 6,809 programs, about 6.5). (positions) | 1–60 | 6.5 |
| Program size spread | Spread of lognormal program sizes (SD of the log size); 0 makes them as equal as possible. | 0–3 | 0.8 |
Applicants
Applicant groups
Applicant types, their shares and strength distributions.
| Column | Meaning | Range |
|---|---|---|
| Name | Group name, for example us_md (1 to 40 lowercase letters, digits or underscores, starting with a letter). | |
| Share | Fraction of applicants in this group; the shares add up to 1. | >0–1 |
| Strength mean | Mean latent strength of the group, in SD units; only differences between groups matter. | -3–3 |
| Strength SD | Spread of strength within the group. | >0–3 |
| Applications (mean) | Mean applications for this group; blank uses the overall mean. | 1–7000 |
| Signals | Most signals an applicant in this group sends; blank uses the signal tiers in full. | 0–100 |
Default:
us_md 0.44 0.5 1.0; us_do 0.18 0.35 1.0; us_img 0.09 -1.0 1.0; non_us_img 0.25 -1.3 1.0; other 0.04 -0.5 1.0
Applicant attributes
What programs evaluate about applicants.
| Column | Meaning | Range |
|---|---|---|
| Attribute | Attribute name, for example board_scores (1 to 40 lowercase letters, digits or underscores, starting with a letter). | |
| Correlation with strength | How closely this attribute tracks the applicant's overall strength (0 = unrelated, 1 = the same). | 0–1 |
| Average weight | How much programs value this attribute on average, relative to the others (renormalised to 1). | >0–100 |
Default:
board_scores 0.8 0.5; research 0.5 0.3; honors 0.6 0.2
Programs
| Parameter | Meaning | Range | Default |
|---|---|---|---|
| Quality SD within tiers | Spread of latent quality within a tier. Without tiers it has no effect on results. (SD) | >0–3 | 1 |
Program tiers
Optional quality tiers; leave empty for one tier with mean 0.
| Column | Meaning | Range |
|---|---|---|
| Name | Tier name, for example top (1 to 40 lowercase letters, digits or underscores, starting with a letter). | |
| Share | Fraction of programs in this tier; the shares add up to 1. | >0–1 |
| Quality mean | Mean latent quality of the tier, relative to the within-tier spread. | -3–3 |
Program attributes
What applicants evaluate about programs.
| Column | Meaning | Range |
|---|---|---|
| Attribute | Attribute name, for example reputation (1 to 40 lowercase letters, digits or underscores, starting with a letter). | |
| Correlation with quality | How closely this attribute tracks the program's overall quality. Ignored for program_size, which is computed from the program's positions. | 0–1 |
| Average weight | How much applicants value this attribute on average, relative to the others (renormalised to 1). | >0–100 |
Default:
reputation 0.9 0.5; program_size 0.0 0.2; location 0.0 0.3
Preferences
| Parameter | Meaning | Range | Default |
|---|---|---|---|
| Applicant agreement | How similarly applicants rank programs: the average correlation between two applicants' true utilities (1 = everyone agrees, 0 = purely personal tastes). | 0–1 | 0.6 |
| Program agreement | How similarly programs rank applicants (1 = all programs agree, 0 = purely personal). | 0–1 | 0.7 |
| Attribute share of the common view | Mixing weight of the average-weighted attributes, rather than latent quality or strength, in the view everyone shares. | 0–1 | 0.3 |
| Taste share of personal preferences | Part of the personal (non-shared) utility that comes from individual attribute weights rather than pure fit. | 0–1 | 0.5 |
| Weight concentration | Shape of individual attribute weights: low values give sparse tastes (one attribute dominates), high values keep everyone's weights close to the average. It does not change how much agents agree. | 0.1–1000 | 10 |
Information
| Parameter | Meaning | Range | Default |
|---|---|---|---|
| Applicant pre-interview noise | Error in applicants' view of programs before interviews, in utility SD units: 0 = perfect information, 0.5 keeps a correlation of about 0.89 between true and perceived utility. (SD) | 0–3 | 0.5 |
| Program pre-interview noise | Error in programs' view of applicants from the application alone, in utility SD units. (SD) | 0–3 | 0.5 |
| Interview informativeness | How much an interview shrinks the pre-interview error (0 = not at all, 1 = reveals the truth). | 0–1 | 0.6 |
| Interview fit shock | New fit, good or bad, that only shows at the interview. (SD) | 0–2 | 0.3 |
| Visibility effect | Extra pre-interview error for less-known programs and applicants (0 = the same error for all). | 0–2 | 0 |
| Shared error (halo) | Part of the pre-interview error that everyone makes about the same program or applicant, so errors herd (0 = independent errors). | 0–1 | 0 |
Applications
| Parameter | Meaning | Range | Default |
|---|---|---|---|
| Application count | Distribution of the number of applications per applicant. | Fixed, Poisson, Negative binomial | Negative binomial |
| Applications (mean) | Mean applications per applicant; a group's own mean replaces it (ERAS 2025-26 real average: about 82). | 1–7000 | 30 |
| Application dispersion | Spread of the negative-binomial application count: the variance is mean + dispersion x mean². | 0.01–10 | 0.5 |
| Application strategy | How applicants choose programs: the best by their pre-interview view (top n), a reach / target / safety portfolio, every program, or at random. | Top n, Portfolio, All, Random | Portfolio |
| Reach | Share of applications to reach programs. | 0–1 | 0.25 |
| Target | Share of applications to target programs. | 0–1 | 0.5 |
| Safety | Share of applications to safety programs. | 0–1 | 0.25 |
| Target band | How close a program's prestige percentile must be to the applicant's self-assessed standing to count as a target (portfolio strategy and realistic signals). | 0–0.5 | 0.15 |
| Self-assessment error | Error in applicants' estimate of their own competitiveness (0 = perfect self-knowledge). (SD) | 0–2 | 0.5 |
Signals
| Parameter | Meaning | Range | Default |
|---|---|---|---|
| Signal allocation | Where applicants send their signals. | Top utility, Realistic, Random | Realistic |
| Programs using signals | Share of programs that use signals when screening. | 0–1 | 1 |
| Signals affect rank lists | Whether a signal also affects programs' rank lists. | off |
Signal tiers
Signal tiers, for example gold 3 / boost 0.8 and silver 12 / boost 0.4.
| Column | Meaning | Range |
|---|---|---|
| Name | Tier name, for example gold (1 to 40 lowercase letters, digits or underscores, starting with a letter). | |
| Signals | Signals per applicant in this tier. | 1–100 |
| Screening boost | Boost to the program's screening score. | 0–3 |
Invitations
| Parameter | Meaning | Range | Default |
|---|---|---|---|
| Interviews per position | Interview slots per position: each program has ceil(ratio x positions) slots. | 1–30 | 10 |
| Invitation strategy | How programs choose whom to invite: by screening score; after the hard screen by score or at random; or signalled applicants first. | Top score, Threshold then top, Threshold then random, Signal first | Top score |
| Screening attribute | Applicant attribute used as a hard screen (blank = none). | blank | |
| Screening minimum percentile | Applicants below this percentile of the screening attribute are not invited (threshold strategies). | 0–1 | 0 |
| Yield protection | Screening penalty per percentile an applicant stands above the program, unless they signalled it. | 0–2 | 0 |
| Invitation rounds | Invitation waves; programs invite again to fill the slots that were declined. | 1–10 | 3 |
Interviews
| Parameter | Meaning | Range | Default |
|---|---|---|---|
| Interviews per applicant (max) | The most interviews an applicant accepts and attends. | 1–50 | 12 |
| Acceptance order | In which order applicants accept invitations: the best first (by pre-interview view), or in the order they arrive. | Best first, First come | First come |
| Interview dates Not used yet | Number of interview dates for scheduling conflicts (0 = no conflicts). | 0–60 | 0 |
| Dates per program Not used yet | Interview dates each program offers. | 1–10 | 3 |
Rank order lists
| Parameter | Meaning | Range | Default |
|---|---|---|---|
| Applicant rank lists | Which interviewed programs applicants rank: all, the top k, those above the reservation utility, or all with the ones they are less likely to match moved down (likelihood weighted). | All interviewed, Top k, Above reservation, Truncate to k, Likelihood weighted | All interviewed |
| Applicant list length | List length for the top-k policies. | 1–300 | 20 |
| Reservation utility | Applicants (above-reservation policy) and programs (do-not-rank threshold) rank only above this post-interview utility. (SD) | -5–5 | -1 |
| Program rank lists | Which interviewed applicants programs rank: all, all but the bottom share, or those above the reservation utility. | All interviewed, Do-not-rank quantile, Do-not-rank threshold | Do-not-rank quantile |
| Do-not-rank share | Bottom share of interviewees a program does not rank. | 0–0.9 | 0.1 |
Match
| Parameter | Meaning | Range | Default |
|---|---|---|---|
| Algorithm | Deferred acceptance proposed by applicants (as the NRMP) or by programs. | Applicant proposing, Program proposing | Applicant proposing |
| Compare both sides proposing | Also run the other proposing side and report the differences. | off |