Contract-First Bayesian Workflows for Hierarchical Behavioural Data
Overview
Provides transparent, contract-first infrastructure for Bayesian analysis of repeated-measures and hierarchical behavioural data. It supports approved Bernoulli-logit, positive lognormal duration, and governed Gaussian dynamic-pupillometry workflows with strict readiness auditing, deterministic simulation, explicit preparation and transformation replay, inspectable scale-aware priors, prior and posterior predictive checks, restricted optional fitting through 'brms' with either 'rstan' or 'cmdstanr', sampling and temporal diagnostics, explicit posterior estimands, sensitivity analysis, target-specific predictive validation, simulation-based calibration, and conservative reporting. Core contracts and validation remain backend-independent. Version 0.5 adds governed robust and distributional dynamic pupillometry, bounded ARMA residual structures, Gaussian-process trajectories, explicit measurement uncertainty and missing-data models, joint binocular analysis, predictive model comparison, functional posterior estimands, and experimental nonlinear response-shape models while preserving explicit scientific and computational governance boundaries.
Advanced dynamic and measurement-aware Bayesian pupillometry
- Adds an additive advanced pupil-model specification layer while preserving the frozen 0.4 public API.
- Adds Gaussian and Student-t observation models with constant, condition-dependent, time-dependent, and condition-by-time residual scale specifications.
- Adds bounded governed residual temporal dependence through AR(1), AR(2), ARMA(1,1), and explicit ARMA specifications constrained to p <= 3 and q <= 2.
- Adds governed Gaussian-process trajectories with Matérn 3/2, Matérn 5/2, and exponentiated-quadratic kernels; approximate Hilbert-space GP bases are the scalable default and exact GP requests are subject to a complexity audit.
- Adds explicit known measurement-uncertainty declarations for pupil responses and covariates and MAR-oriented joint missing-value models without automatic interpolation.
- Adds joint binocular pupil preparation, multivariate modelling, residual eye correlation, eye-difference estimands, and agreement summaries without requiring upstream eye averaging.
- Adds governed PSIS-LOO, exact K-fold, predictive model weights, and explicit leave-future-out validation plans. Model comparison never chooses a substantive winner automatically.
- Adds an experimental nonlinear pupil response-shape model with interpretable baseline, amplitude, onset, rise, duration, and decay parameters.
- Adds posterior temporal derivative, dynamic condition-contrast, and prespecified threshold-duration estimands without automatic onset, changepoint, or favorable-window detection.
- Adds computational-budget, design-support/identifiability, temporal-dependence, measurement, missingness, predictive-calibration, and advanced-fit diagnostics.
- Adds posterior residual-scale, GP hyperparameter, residual-spectrum, model-card, sensitivity-suite, and publication-oriented tables and graphics.
Show all changes in 0.5.0Hide the rest of the changes in 0.5.0
- Adds deterministic advanced and binocular simulation utilities with truth stored separately from analysis data.
- Adds backend-free examples, plotting galleries, focused tests, and eleven advanced pupillometry articles.
Governance boundaries
- gp3bayes 0.5 does not interpolate or otherwise preprocess missing pupil samples automatically.
- MAR-oriented missing-data modelling is assumption-conditional and does not establish that MAR is true.
- Student-t robustness does not mark individual observations as invalid outliers.
- Predictive comparison, stacking weights, temporal derivatives, threshold durations, and nonlinear response parameters do not establish model adequacy, causality, or cognitive/physiological states.
- MNAR models, overlapping-event deconvolution, automatic changepoint detection, multimodal psychophysiological latent-state inference, automated model search, and automatic cognitive-state inference remain outside the 0.5 scope.
From NEWS.md in the release
Install
Health
- OK13 OKNow
- OK8 OK2026-08-10
Documentation
- Help pages
- 465 (5 internal)includes a package overview page
- Examples that run
- 53%40 run in a plain check (3 of them under a condition) · 1 only when donttest examples are enabled (R CMD check --as-cran) · 34 hold no code
- Exported functions with examples
- 16%
- Documented parameters
- 94%
- Return-value docs
- 94%
- Help pages citing references
- 0%
- A Reproducible 0.2.0 Release Case Study HTML
- A Stable Unified Workflow API HTML
- Advanced Dynamic Pupillometry in gp3bayes 0.5 HTML
- Advanced Optional Bayesian Workflows HTML
- Advanced Predictive Diagnostics HTML
- Analysis Manifests and Reproducible Bayesian Workflows HTML
- Backend Portability and Installation HTML
- Backend Reliability, Parity and Object Schemas HTML
- Baseline, gaze/PFE, and luminance sensitivity HTML
- Bayesian dynamic pupillometry: governed foundation HTML
Show 49 moreHide 49 vignettes
- Bounded ARMA and Temporal Diagnostics HTML
- Complete Public API Map HTML
- Computational Governance and Model Cards HTML
- Declared Priors versus Fitted Posteriors HTML
- End-to-End Evidence and Publication Showcase HTML
- End-to-End Hierarchical Binary Workflow HTML
- End-to-End Hierarchical Lognormal Duration Workflow HTML
- Evidence Graphics and Governance HTML
- Experimental Interpretable Pupil Response Shape HTML
- First-Class Estimands and Sensitivity Workflows HTML
- Fitting hierarchical pupil time-course models HTML
- Functional Dynamics and Predictive Calibration HTML
- Gaussian-Process Pupil Trajectories HTML
- Gazepoint pupil-data interoperability HTML
- Governed Predictive Model Comparison HTML
- Grouped PSIS-LOO Influence HTML
- Hierarchical Effect and Variance Atlases HTML
- Hierarchical Effects and Predictive Uncertainty HTML
- Joint Binocular Pupil Models HTML
- LOO Influence and Predictive Model Comparison HTML
- Measurement Uncertainty and Missing Pupil Data HTML
- Model Cards and Reporting Inventories HTML
- Optional Bayesian Backend Installation HTML
- Parameter Recovery Diagnostics for Publication HTML
- Pathological Simulation Scenarios HTML
- Pointwise LOO Influence Atlases HTML
- Posterior Exploration and Publication Graphics HTML
- Posterior pupil trajectories and declared estimands HTML
- Pre-fit Design-Support Diagnostics HTML
- Prediction Contrasts, Rankings, and Groups HTML
- Prediction Profiles, Surfaces, and Contrast Profiles HTML
- Prediction, Calibration, and Scoring HTML
- Predictive Distribution and Calibration Uncertainty HTML
- Preparing and auditing pupil time courses HTML
- Prior Sensitivity and Simulation-Based Recovery HTML
- Publication Registries and Diagnostic Dashboards HTML
- Publication-Ready Analysis Bundles HTML
- Pupil posterior predictive checks and temporal diagnostics HTML
- Quality Hardening and Failure Contracts HTML
- Robust and Distributional Pupil Models HTML
- Sampling Diagnostics and Conservative Decisions HTML
- Sensitivity Atlases HTML
- Simulation-Based Calibration Diagnostics HTML
- Specification Closure: Strict Readiness and Governed Validation HTML
- Synthetic Advanced Pupillometry Gallery HTML
- Synthetic Gazepoint pupillometry case study HTML
- Transformation Replay and Detailed Posterior Predictive Checks HTML
- Unified Sensitivity Suites and Evidence Inventories HTML
- Validation for temporally dependent pupil data HTML
Downloads
Repository
- PRs merged (all time)
- 100%
- Median days to close, 50 most recently updated closed issues
- not measured
- Median days to close, 50 most recently updated closed PRs
- 0
Releases over time
Issues over time
PRs over time
Commits over time
Commit identities over time
9 development-tooling and community-health practices detected across 5 families in the upstream repository
Checks run against github.com/stefanosbalaskas/gp3bayes on 2026-10-11.
The R CMD check workflow runs on Linux, macOS and Windows, and on R-devel
GitHub Pages site last deployed 2026-10-09 · The repository has a site deploy workflow
pkgdown site last built 2026-10-09
Show all practicesHide 1 practice
Dependencies
Code & Tests
People & History
Pick two releases to compare their code metrics.
- 0.5.0Latest2026-08-23 · current release · diff ↗ ·
added 427 exports
advanced_pupil_trajectory_table, analysis_bundle_table, analysis_manifest_table, apply_transformation_recipe, as_pupil_prediction_draws, assess_powerscaled_sensitivity, audit_advanced_pupil_identifiability, audit_backend_parity, audit_binocular_pupil_readiness, audit_design_support, audit_duration_boundaries, audit_duration_unit_invariance, audit_estimand_invariance, audit_fixed_effect_design, audit_missingness_structure, audit_model_readiness_strict, audit_prediction_support, audit_pupil_computational_budget, audit_pupil_measurement_context, audit_pupil_measurement_model, audit_pupil_missingness, audit_pupil_predictive_calibration, audit_pupil_readiness, audit_pupil_temporal_dependence, audit_random_effects_support, backend_capabilities, backend_environment_table, backend_parity_table, bayesian_backend_capabilities, binary_calibration_error and 397 more - 0.1.12026-08-09
- RR 4.6.0 released · 2026-04-24
Package metadata
- First published
- 2026-08-09
- Total releases
- 2 / 1 yrs
- License
- MIT + file LICENSE OSI
- Additional repositories
- stan-dev.r-universe.dev, r-multiverse.r-universe.dev
- Download size
- 1.0 MB
Cite
Cite this package
Run in R for the authors' entry:
citation("gp3bayes")The package provides a citation file, so citation() returns the authors' own entry.
The repository also keeps a CITATION.cff.
Cite the R Observatory
For a number measured here: a download total, a coverage figure, an archival date.
From data release v2026-10-19, which the citation names so these numbers can be found later. More on citing and the projects behind them.