Architecture overview

This page is the contributor mental model: what depends on what, and where to make changes safely.

For user-facing examples and guides, prefer wrapper entry points (dpmix() / dpmgpd() for one-arm and dpmix.causal() / dpmgpd.causal() for causal). This page focuses on the internal execution path (bundle() -> mcmc()) used by contributors.

Layers

1. Spec layer: compiles user inputs into a normalized internal specification.
2. Contracts: enforce legal configurations and guard invariants early.
3. Registry: routes kernel/tail choices to concrete implementations.
4. Bundle + codegen: builds Nimble model code, constants/data, monitors, inits.
5. MCMC: runs the sampler, returning fit objects with standardized slots.
6. S3 surface: predict/plot/summary that should never need to know implementation details.

Call flow (non-causal)

Default user path: dpmix(...) / dpmgpd(...). Internal path shown below: bundle() then mcmc().

sequenceDiagram
  participant U as User
  participant S as compile_model_spec()
  participant R as kernel registry
  participant B as bundle()
  participant M as mcmc()
  participant F as mixgpd_fit
  U->>B: bundle(...)
  B->>S: compile spec
  S->>R: validate + route kernel/tail
  B->>M: bundle (code/data/constants)
  M->>F: fit object
  U->>F: summary/plot/predict

Call flow (causal)

Default user path: dpmix.causal(..., treat=...) / dpmgpd.causal(..., treat=...). Internal path shown below: bundle() then mcmc().

sequenceDiagram
  participant U as User
  participant C as bundle()
  participant P as PS sub-bundle
  participant M as mcmc()
  participant Q as qte()/ate()
  U->>C: bundle(...)
  C->>P: build + run PS (optional)
  U->>M: mcmc(...)
  M->>Q: derive estimands

Prereqs

• Required packages and data for this page are listed in the setup chunks above.

Outputs

• This page renders model fits, diagnostics, and summary artifacts generated by package APIs.

Interpretation

• Canonical concept page: Spec And Contracts
• Treat this page as an application/example view and use the canonical page for core definitions.

Next

• Continue to the linked canonical concept page, then return for implementation-specific details.