Documentation

Start with the README or choose a path in Concepts and Reading Paths.

Tutorials

Clay notebooks in notebooks/foerster/, rendered at replikativ.github.io/foerster. Each runs as it renders, in CI on every pull request.

Notebook	Topics
Getting Started	a first model, sample and observe, SMC, the posterior and the evidence
Inference in a Running Application	a live queue, speculative application state, SMC checked against enumeration
Choosing an Inference Algorithm	importance sampling, SMC, particle MCMC, MCMC and BBVI on known posteriors; which to use when
Inference Methods	every inference method foerster implements — what it does, when to use it, its options and cost, a run on a model with a known answer, and the papers it comes from
Models in Worlds	:world-policy :fork, systems in the world, budgets, what cannot be copied
Blocks and HMC	numerical blocks, HMC-within-Gibbs, checking a gradient
Streaming SMC	online filtering against the Kalman filter
Trace Programming and Custom Moves	the generative function interface, MH by selection, involutive MCMC
Interventions and Counterfactuals	seeing and doing, interventions by selector, twin-world counterfactuals
From Data to Counterfactual Repairs	native mechanism fitting, held-out checks, bootstrap effect/attribution uncertainty and incident repairs against known truth
A Gallery of Classic Models	small models checked against exact answers, including branching and label switching
From the PyMC Gallery	golf putting (logistic and geometric models compared by LOO) and radon (partial pooling, NUTS on 89 parameters), each against an exact posterior; stochastic volatility (a non-centred path by NUTS, checked by a particle filter)
Marketing Mix Modeling	adstock and saturation fit by NUTS on a block, ROAS by counterfactual and in the twin world, a lift test, budget splits by expected response and CVaR — against a known truth
Steering a Process	SMC over scored steps of a process without a density, value estimates as twists, trajectories as training data

Guides

Guide
Concepts and Reading Paths	worlds, traces, measures, populations, blocks and their boundaries
Incremental Trace Inference	complete keyed traces, finite Gibbs, resumable SCMC and the proposed reactive boundary
Causal Queries and Identifiability	reusable abduction, nested worlds, ID/IDC, attribution, repairs and actual causality
Amortized Inference	automatic RV schemas and masks, conditional sampling artifacts, causal reconstruction and intervention inference
Language	sites and their options, addresses, selectors, the rules of the spin macro
Algorithms	every inference entry point and its options; foerster in an interpreter sandbox
Posteriors	reading a measure: particles, traces, summaries, ESS, evidence, training data
Checking a Model and its Inference	prior predictive simulation, several methods, R-hat/ESS/MCSE and SMC diagnostics, posterior predictive checks, model comparison, simulation-based calibration
Coming from Other PPLs	Stan, PyMC, Turing, Gen, Anglican and WebPPL concepts in foerster, and what foerster does not have yet
Inference Programs	scoped custom population programs, targets, weight accounting and Markov transitions
Extending	your own distributions, policies, proposals, MH moves and kernels; learned twists and proposals
Worlds	canonical worlds, their lifecycle, failure and recovery, resources
Distributions and Reproducibility	foerster.dist, writing a distribution, seeds and streams
Design	savepoint-based traces, numerical capabilities and the shared inference engines
Literature	the systems and papers foerster builds on, by method

Reference

The API reference is on cljdoc, generated from the docstrings.

• Incremental trace inference: keyed execution, finite Gibbs, ownership restrictions and the proposed reactive-data boundary.