The default case: seed sets R's own RNG state once | DesignFixedOptimal: deterministic solves, seeded coin and chains | The portable cross-language RNG: edi_rng::RRng | The one documented exception: continuing R's live stream bit-for-bit | Seed-reproducible via per-draw local generators: DesignFixedGreedyDOptimal | Seed-reproducible despite being parallel: greedy-family designs | Randomization inference reuses the design's own draw mechanism | Bootstrap resampling | Machine-dependent performance defaults: tune_EDI_for_this_machine() | Simulation (SimulationFramework): per-replication and per-cache-job seeds | Monte Carlo error | Reproducing a documented example

Setup and a simulated population | A fixed design: everyone is known up front | Inference: covariate-adjusted OLS | Everything at once: InferenceSuite | A sequential design: subjects arrive one at a time | Where to go next

Setup | Fixed design, Poisson regression | Over-dispersion: negative binomial on the same design | Everything at once | Sequential design | Where to go next

Setup | Fixed design, logistic regression | A risk difference instead of an odds ratio | Everything at once | Sequential design: matching on the fly | Where to go next

Setup | Fixed design, proportional odds | Everything at once | Sequential design | Where to go next

Setup | Fixed design, fractional logit | Beta regression on the same design | Everything at once | Sequential design | Where to go next

Setup | Fixed design, recording events and censoring | Inference: Cox proportional hazards | Everything at once | Sequential design | Where to go next

The validation boundary: R6 wrapper vs. raw backend | Argument dimensions and storage order | Numeric domains, overflow, and underflow safeguards | Convergence flags and return-object conventions | Shared cores and wrapper-to-backend equivalence | NA/NaN handling

Setup and a simulated population | A fixed design: everyone is known up front | Inference: covariate-adjusted OLS | Everything at once: InferenceSuite | A sequential design: subjects arrive one at a time | Where to go next

Setup | Fixed design, Poisson regression | Over-dispersion: negative binomial on the same design | Everything at once | Sequential design | Where to go next

Setup | Fixed design, logistic regression | A risk difference instead of an odds ratio | Everything at once | Sequential design: matching on the fly | Where to go next

Setup | Fixed design, proportional odds | Everything at once | Sequential design | Where to go next

Setup | Fixed design, fractional logit | Beta regression on the same design | Everything at once | Sequential design | Where to go next

Setup | Fixed design, recording events and censoring | Inference: Cox proportional hazards | Everything at once | Sequential design | Where to go next

How EDI classes are built | The inference extension contract | Randomization and bootstrap shells | Subclassing rules and capability detection | Custom designs | What the shells deliberately do not cover | Contributing a class to EDI itself

Sample, covariates, and the design matrix | Treatment assignment | Response, censoring, and the y/y_L/y_R schema | Coefficients and treatment effects | Design structure: blocks, matched pairs, clusters, and the reservoir | Resampling and randomization | Treatment-effect scales | Transformed outcomes | Where this glossary does not apply

The default case: seed sets R's own RNG state once | DesignFixedOptimal: deterministic solves, seeded coin and chains | The portable cross-language RNG: edi_rng::RRng | The one documented exception: continuing R's live stream bit-for-bit | Seed-reproducible via a local generator: DesignFixedGreedyDOptimal | Seed-reproducible despite being parallel: DesignFixedGreedy | Randomization inference reuses the design's own draw mechanism | Bootstrap resampling | Machine-dependent performance defaults: tune_EDI_for_this_machine() | Simulation (SimulationFramework): per-replication and per-cache-job seeds | Monte Carlo error | Reproducing a documented example

Continuous | Incidence / binary | Count | Ordinal | Survival | Proportion | GEE / GLMM (matched-design, correlated data) | Numerical/backend utilities | Simulation/calibration checks | Coverage note

The validation boundary: R6 wrapper vs. raw backend | Argument dimensions and storage order | Numeric domains, overflow, and underflow safeguards | Convergence flags and return-object conventions | Shared cores and wrapper-to-backend equivalence | NA/NaN handling

How EDI classes are built | The inference extension contract | Randomization and bootstrap shells | Subclassing rules and capability detection | Custom designs | What the shells deliberately do not cover | Contributing a class to EDI itself

Sample, covariates, and the design matrix | Treatment assignment | Response, censoring, and the y/y_L/y_R schema | Coefficients and treatment effects | Design structure: blocks, matched pairs, clusters, and the reservoir | Resampling and randomization | Treatment-effect scales | Transformed outcomes | Where this glossary does not apply

Continuous | Incidence / binary | Count | Ordinal | Survival | Proportion | GEE / GLMM (matched-design, correlated data) | Numerical/backend utilities | Simulation/calibration checks | Coverage note

Fit a classification forest | Inspect out-of-bag performance | Missing values

Introduction | Overview of BART | The bartMachine package | bartMachine package features for regression | bartMachine package features for classification | Discussion | Sampling to modify tree structure | Bakeoff