Language 0.8 example catalog

Fifty small models, each tested

Copy a complete file, compare modeling patterns, or run the catalog validator to check the parser, model lowering, derivatives, initial state, and generated quantities.

Stan-inspired, not Stan-compatible

01–10 · Foundations

Conjugacy, likelihoods, and constraints

Script	Pattern	Main feature
01 beta-binomial	Coin toss posterior	Bounded probability and predictive binomial RNG
02 A/B test	Two independent arms	Multiple bounded parameters
03 normal mean	Known observation scale	Conjugate location
04 location/scale	Vectorized normal data	Positive scale and prediction
05 robust location	Outlier-resistant model	Vectorized Cauchy likelihood
06 gamma–Poisson	Count-rate posterior	Positive rate and count RNG
07 exponential rate	Waiting-time model	Vectorized positive data
08 Bernoulli	Binary outcome	Beta prior and Bernoulli RNG
09 binomial	Grouped binary outcomes	Binomial likelihood
10 bounded latent	Latent location on [-2,2]	Two-sided transform and uniform prior

11–20 · Regression and structure

Transformed parameters, links, hierarchy, order, and simplex state

Script	Pattern	Main feature
11 precision	Scale-to-precision conversion	Transformed parameter
12 standardized effect	Non-centered latent effect	Derived scientific parameter
13 log response	Normal model on log data	Vector data and prediction
14 Poisson log link	Count regression	`exp` link
15 logistic	Single binary observation	`inv_logit` link
16 two-point logistic	Two predictor values	Multiple transformed probabilities
17 two groups	Difference in means	Two vectorized likelihoods
18 non-centered groups	Partial-pooling skeleton	Standardized effects
19 ordered means	Monotone locations	Ordered vector and indexing
20 three categories	Category probabilities	Simplex and manual log target

21–30 · Priors, transformations, and prediction

Prior predictive simulation and generated quantities

Script	Pattern	Main feature
21 beta predictive	Prior-only probability	Bernoulli prior prediction
22 gamma predictive	Prior-only rate	Poisson prior prediction
23 half-Cauchy scale	Positive constrained scale	Cauchy prior on a transformed parameter
24 bounded location	Finite prior support	Uniform prior and normal likelihood
25 centered data	Subtract a known offset	Transformed data block
26 endpoint sum	Baseline plus change	Transformed parameter block
27 normal RNG	Continuous prediction	`normal_rng`
28 Bernoulli RNG	Binary prediction	`bernoulli_rng`
29 binomial RNG	Future successes	`binomial_rng`
30 Poisson RNG	Future count	`poisson_rng`

31–40 · Extended surface

Remaining RNGs, manual density terms, indexing, and pooled groups

Script	Pattern	Main feature
31 gamma RNG	Positive prediction	Shape/rate convention
32 exponential RNG	Future waiting time	`exponential_rng`
33 uniform RNG	Reference simulation	`uniform_rng`
34 Cauchy RNG	Heavy-tailed prediction	`cauchy_rng`
35 manual target	Explicit prior and likelihood	`target += *_lpdf`
36 nonlinear mean	Exponential response curve	Nonlinear transformed parameter
37 indexed vector	Three explicit observations	One-based indexing
38 ordered prior	Four latent cutpoints	Vectorized prior on ordered state
39 simplex regularization	Balanced weights	Manual simplex prior
40 pooled groups	Grand mean and contrast	Two vectorized group likelihoods

41 · Data and control flow

Start with a real file-backed workflow

Script	Pattern	Main feature
41 CSV normal mean	CSV observations through Java or script	`for`, indexing, external data binding, and normal prediction

Run it with McmcDataIngestionExamples.java or follow the data and compilation tutorial.

42–50 · Version 0.8.2 additions

Apply the expanded language and probability catalog

Script	Pattern	Main feature
42 robust Student-t	Location/scale data with an outlier	`student_t`, transformed degrees of freedom, and RNG
43 Bernoulli-logit	Binary regression	Logit likelihood and RNG without manual probability conversion
44 beta-binomial	Overdispersed grouped successes	Mean/concentration transformation and predictive count
45 negative-binomial log link	Overdispersed count regression	`neg_binomial_2_log` likelihood and RNG
46 piecewise errors	Robustify observations past a cutoff	Scoped locals, `while`, `if`, assignment, and `\|`
47 contamination mixture	Clean and outlier observation regimes	`log_mix` and local log densities
48 Weibull lifetime	Positive event times	Vectorized Weibull likelihood and RNG
49 circular direction	Angular observation	Von Mises likelihood and circular RNG
50 reaction time	Gaussian core with exponential delay	Exponentially modified normal likelihood and RNG

The companion ModelScriptCompilationExamples.java loads a checked-in script and compares in-memory, cached-JDK, and generated-source workflows.

51 · First-passage reaction time

Fit a Wiener drift-diffusion likelihood

Script	Pattern	Main feature
51 Wiener reaction time	Upper-response reaction times	Stable `wiener` likelihood and conditional `wiener_rng`

The four-parameter likelihood follows Stan's upper-response convention. JDistlib's RNG is an extension and samples time conditional on that response.

Validate

Run the semantic catalog gate

./gradlew validateModelScripts

The validator supplies representative typed data for each file, compiles the script, checks its analytic gradient against central finite differences, requires a finite initial density, and evaluates generated quantities.