fitPS development news

Purpose

This file records user-facing and developer-facing changes for fitPS. It is a release-note summary, not a commit-by-commit history.

fitPS 1.1.3.031

• Restore xtable for the two getting-started vignette tables to recover the cleaner LaTeX presentation.
• Suppress xtable generator comments while retaining explicit captions, mathematical probability headings, and stable LaTeX labels.
• Keep the probability table to four meaningful columns and retain clear Lau-Jackson survey headings.
• Preserve package behaviour and validate the rendered getting-started PDF plus all four vignette sources.

fitPS 1.1.3.028

• Replace the two xtable-based tables in the getting-started vignette with native knitr tables that render cleanly in PDF output.
• Restore clear mathematical probability headings in the raw, zeta, and ZIZ comparison table without exposing LaTeX generator comments.
• Give the Lau-Jackson survey comparison table explicit column headings and keep its caption with the table.
• Preserve package behaviour and validate the rendered getting-started PDF plus all four vignette sources.

fitPS 1.1.3.027

• Refocus the uncertainty vignette on confidence and credible regions as statistical methods rather than on numerical extractor terminology.
• Stop printing full confidence- and credible-region coordinate sets in the vignette while retaining the programmatic region objects for users who need them.
• Restore the four-way comparison of fitted P probabilities across MLE, ordinary bootstrap, parametric Bayes, and Rubin Bayesian Bootstrap.
• Add a direct visual comparison of probability uncertainty while keeping plotting and cached-computation plumbing out of the user narrative.
• Preserve package behaviour and validate all four vignette sources with the lightweight knit-based documentation check.

fitPS 1.1.3.026

• Rework all package vignettes to read as user-facing statistical documentation rather than development notes.
• Rename and restructure the uncertainty vignette around parameter uncertainty, probability uncertainty, method choice, numerical extraction, plotting, and practical interpretation.
• Simplify the basics and model-comparison narratives and reframe the extension vignette as a practical guide for authors adding statistical models.
• Preserve the existing fitting APIs, examples, cached uncertainty figure, and inferential behaviour; this stage changes documentation only.
• Validate the four vignette sources with a lightweight knit-based documentation check.

fitPS 1.1.3.025

• Repaired two-dimensional ordinary and Rubin Bayesian Bootstrap uncertainty regions to use an unconstrained bivariate KDE rather than the ks positive-support transformation.
• Preserved Hscv bandwidth selection and probability-content contour extraction while removing a boundary transformation that rotated and over-smoothed the observed joint replicate density.
• Updated the deprecated bootCI compatibility path to use the same corrected unconstrained ZIZ bootstrap KDE.
• Added regression checks for unconstrained KDE coordinates and contour orientation relative to the bootstrap replicate cloud.
• Regenerated the fixed-seed 1,000-replicate cached uncertainty comparison and empirical-containment assets used by the vignette.
• Preserved profile-likelihood and parametric-Bayesian uncertainty calculations unchanged and validated through the strict fitPS documentation, test, package-check, vignette-build, build, and installation workflow.

fitPS 1.1.3.024

• Carried forward the verified Stage 11.8.6 bootstrap-geometry and cached-vignette asset work after its changed-path audit stopped on a collapsed untracked directory entry.
• Made the stage audit enumerate untracked files individually so generated files under vignettes/figures are checked and staged by their exact package-relative paths.
• Preserved the historical bootCI geometry regression, 1,000-replicate cached bootstrap assets, shared-scale four-panel comparison, and containment diagnostics unchanged.
• Validated through the strict fitPS documentation, test, package-check, vignette-build, build, and installation workflow.

fitPS 1.1.3.023

• Verified that ordinary-bootstrap plotUncertainty() contours reproduce the historical bootCI() Hscv/KDE/contourLines geometry from identical parameter replicates.
• Added deterministic regression coverage that locks the modern sample-region coordinates to the historical smoothed-bootstrap construction.
• Precomputed the 1,000-replicate ordinary- and Rubin Bayesian-Bootstrap vignette assets with fixed seeds so routine vignette builds no longer rerun the expensive bootstrap fits.
• Stored the four-panel shared-scale uncertainty comparison and empirical containment table as reproducible vignette assets, with a development regeneration script retained in the source repository.
• Validated through the strict fitPS documentation, test, package-check, vignette-build, build, and installation workflow.

fitPS 1.1.3.022

• Restored the uncertainty vignette’s single four-panel ZIZ comparison by retaining only the final base-graphics device state from the shared-scale plotting chunk.
• Kept the visible vignette code focused on the four plotUncertainty() calls while leaving panel layout and shared-axis plumbing hidden.
• Preserved the Stage 11.8.4 bootstrap containment diagnostics and 1,000-realization uncertainty calculations unchanged.
• Validated through the strict fitPS documentation, test, package-check, vignette-build, build, and installation workflow.

fitPS 1.1.3.021

• Increased ordinary and Rubin Bayesian Bootstrap vignette runs from 100 to 1,000 realizations for more stable two-dimensional KDE uncertainty regions.
• Added empirical point-in-contour containment diagnostics for sample-based uncertainty regions without adding a spatial-package dependency.
• Reported nominal 80% and 95% region content beside the observed fraction of bootstrap realizations enclosed by each displayed contour.
• Preserved the historical ks contour-level convention for this diagnostic stage so containment results can determine whether a later semantic correction is required.
• Added deterministic tests for polygon membership, containment calculation, and stored sample-region diagnostics.
• Repaired the stage audit to accept and commit the roxygen-generated plotUncertainty help file.
• Validated through the strict fitPS documentation, test, package-check, vignette-build, build, and installation workflow.

fitPS 1.1.3.019

• Corrected the four-panel uncertainty vignette so shared axes include the final bootstrap KDE contour coordinates as well as the realization clouds, preventing contour clipping.
• Kept the four ZIZ uncertainty panels on exactly the same padded pi and shape scales for direct visual comparison.
• Hid panel-layout, KDE-boundary, and shared-axis bookkeeping from the rendered vignette while leaving only the four meaningful plotUncertainty() calls visible to readers.
• Preserved the established profile-likelihood, ordinary-bootstrap, parametric-Bayesian, and Rubin Bayesian-Bootstrap interpretations and plotting behavior.
• Validated through the strict fitPS documentation, test, package-check, vignette-build, build, and installation workflow.

fitPS 1.1.3.018

• Made the uncertainty vignette compare MLE, ordinary bootstrap, parametric Bayes, and Rubin Bayesian Bootstrap ZIZ regions in a single 2 by 2 panel with shared parameter axes.
• Showed ordinary- and Bayesian-Bootstrap parameter realizations beneath their KDE contours by default while leaving parametric Bayesian point displays opt-in.
• Preserved the original ZIZ smoothed-bootstrap construction by using Hscv() KDE smoothing with positivity enforced for the pi/shape parameterization.
• Standardized user-facing terminology on profile-likelihood confidence intervals and regions while retaining the likelihood-ratio cutoff as their mathematical construction.
• Added regression coverage for method-specific bootstrap realization defaults and validated through the strict fitPS package workflow.

fitPS 1.1.3.017

• Made plotUncertainty() the documented parameter-uncertainty plotting interface across MLE, ordinary bootstrap, parametric Bayes, and Rubin Bayesian Bootstrap fits.
• Kept credint()/credInt() as the canonical Bayesian credible-interval and credible-region extractor, extending it across stored posterior representations and Rubin Bayesian Bootstrap results.
• Kept confint() as the frequentist numerical confidence-interval and likelihood-ratio contour extractor, with credint() as its Bayesian counterpart and plotUncertainty() as the common visual layer.
• Updated the basics, uncertainty, and README documentation so plotUncertainty() examples are rendered and the numerical-versus-visual uncertainty interfaces are explicit.
• Deprecated only the redundant bootCI() compatibility path and added regression coverage that credint() remains an active Bayesian extractor.
• Validated with the strict fitPS documentation, test, package-check, build, and installation workflow.

fitPS 1.1.3.016

• Repaired the cumulative posterior integration helper to use direct vector indexing instead of unqualified head() and tail() calls.
• Removed the resulting R CMD check NOTE without adding unnecessary package imports or changing numerical results.
• Carried forward the Stage 11.7 uncertainty-plotting interface, Simpson cumulative integration, and weighted KDE repair unchanged.
• Validated with the strict fitPS documentation, test, package-check, build, and installation workflow.

fitPS 1.1.3.013

• Corrected plotPosterior() credible-region hatching so the shaded polygon follows the posterior density curve rather than joining only the interval endpoint heights.
• Added deterministic regression coverage for the posterior interval polygon geometry.
• Modernized plotPosterior() documentation and examples around the fit() Bayesian interface.
• Clarified why the uncertainty vignette uses the zero-inflated zeta model as a rich worked example and stated that model choice is a separate task.
• Validated with the strict fitPS documentation, test, and package-check workflow.

fitPS 1.1.3.012

• Rewrote README.md around fit() as the canonical interface for MLE, parametric Bayes, the ordinary bootstrap, and Rubin’s Bayesian Bootstrap.
• Audited the user-facing vignettes for stale API language and added foundational references for forensic P/S modelling, model-comparison criteria, and uncertainty methods.
• Added Curran, Buzzini and Trejos (2024), Estimating probability terms for the background presence of glass when considering activity in forensic casework, as the forensic sampling-uncertainty reference and retained Rubin (1981) for the Bayesian Bootstrap.
• Modernized the package title, description, and CITATION entry and made the cited package version follow DESCRIPTION automatically.
• Corrected the Coulson et al. bibliography year typo and normalized stored DOI fields.
• Kept the extension vignette focused on the fitPS public model contract rather than adding unnecessary external citations.
• Validated with the strict fitPS documentation, test, and package-check workflow.

fitPS 1.1.3.011

• Made fit() the canonical public entry point for maximum likelihood, parametric Bayesian inference, the ordinary nonparametric bootstrap, and Rubin’s Bayesian Bootstrap.
• Added method = “bootstrap” to return the maximum-likelihood psFit with psBootstrap uncertainty attached, and method = “bayesianBootstrap” to return psBayesianBootstrap inference.
• Removed the standalone bayesianBootstrap() public wrapper while retaining the psBayesianBootstrap class and its summary/print methods.
• Deprecated bootstrapFit() with guidance to use fit(…, method = “bootstrap”) while preserving its historical compatibility behaviour.
• Updated bootstrap diagnostics, examples, README material, and uncertainty/model-comparison vignettes to use the unified fit interface and explain all four inferential routes.
• Added deterministic regression coverage for fit-based bootstrap dispatch, deprecation behaviour, Bayesian Bootstrap dispatch, and unsupported ordinary-bootstrap model coverage.
• Validated with the strict fitPS documentation, test, and package-check workflow.

fitPS 1.1.3.007

• Added the public bayesianBootstrap() interface for Rubin Bayesian Bootstrap inference with built-in zeta, zero-inflated zeta, and logarithmic models.
• Added print() and summary() methods for psBayesianBootstrap while keeping raw parameter/probability replicates and failed-fit diagnostics available on the object.
• Documented that Rubin Bayesian Bootstrap uncertainty is over empirical observation weights and is distinct from the parametric Bayesian psPosterior architecture.
• Added deterministic public-API tests covering object identity, summaries, printing, validation, and separation from psPosterior and psBootstrap.
• Recorded future work on likelihood-matched conjugate zeta priors, including effective-sample-size and prior-geometric-mean interpretations, without expanding Stage 11 scope.
• Validated with the strict fitPS documentation, test, and package-check workflow.

fitPS 1.1.3.006

• Added an internal Rubin Bayesian Bootstrap replicate engine using grouped Dirichlet category weights derived directly from observed survey counts.
• Scaled grouped weight draws to the observed sample size and refitted built-in zeta, zero-inflated zeta, and logarithmic models through the Stage 11 weighted-MLE boundary.
• Added the provisional psBayesianBootstrap object with parameter/probability replicates, equal-tail summaries, seed metadata, and explicit failed-fit diagnostics.
• Kept failed weighted fits as missing replicate rows rather than redrawing or smoothing them, including genuinely singleton-support surveys.
• Added deterministic tests for positive normalized weights, seeded reproducibility, grouped-Dirichlet mean behaviour, replicate dimensions, failure retention, and built-in model coverage.
• No exported Bayesian Bootstrap user-facing function is introduced yet; public API integration is deferred to the next stage.
• Validated with the strict fitPS documentation, test, and package-check workflow.

fitPS 1.1.3.004

• Added an internal weighted-MLE fitting boundary for zeta, zero-inflated zeta, and logarithmic models using arbitrary positive real category weights.
• Kept observed psData counts unchanged by substituting weights only in a private likelihood copy, avoiding makePSData() rounding and rep()-based expansion.
• Reused the established model log-likelihood and probability contracts so weighted fits are ready for grouped Dirichlet Bayesian-Bootstrap replicates.
• Added deterministic regression coverage for ordinary-count equivalence, fractional weights, common-scale invariance, invalid weights, and sparse-support handling.
• Corrected the psData rounding regression test so warning detection and returned-object inspection are tested separately across testthat versions.
• No public Bayesian Bootstrap API or uncertainty class is introduced in this stage.
• Validated with the strict fitPS documentation, test, and package-check workflow.

fitPS 1.1.3.001

• Audited Bayesian Bootstrap weighted fitting before introducing any public Bayesian Bootstrap API.
• Confirmed that zeta, zero-inflated zeta, and logarithmic core log likelihoods already accept arbitrary positive real weights mathematically.
• Identified public psData construction and legacy rep()-based expansion paths as integer-count boundaries that should remain separate from Bayesian-Bootstrap weights.
• Established grouped Dirichlet draws with parameters equal to observed category counts as the efficient observation-level-equivalent representation.
• Recommended a small internal weighted-MLE layer for the next stage while keeping psBootstrap, psPosterior, and the future psBayesianBootstrap statistically distinct.
• Package validation was intentionally skipped because this stage changes only development audit documentation.

fitPS 1.1.2.008

• Repaired Stage 10.4 source-package completion after verbose devtools::build() output obscured the returned archive path.
• Preserved the completed unconstrained-parameter API rename and extension-vignette documentation from Stage 10.4.1.
• Recorded the exact source archive path directly from R before installing that verified build.
• Retained portable cleanup of older completed repository archives after the new archive is verified.
• Revalidated documentation, the extending vignette, tests, package checks, source build, and installation.

fitPS 1.1.2.007

• Renamed the public MCMC transformation contract to modelToUnconstrained(), modelFromUnconstrained(), and modelLogJacobian() so the API states its mathematical purpose directly.
• Removed the unreleased modelToWorking(), modelFromWorking(), and modelWorkingLogJacobian() names rather than retaining compatibility aliases.
• Expanded the extending vignette to explain natural and unconstrained parameter scales, inverse transformations, Jacobian direction, and one- and multi-parameter examples.
• Updated built-in models, generic MCMC, external-model fixtures, and public-contract regression tests to use the new terminology.
• Replaced non-portable completed-archive cleanup with a macOS-safe Bash glob loop in the Stage 10.4.1 runner.
• Loaded the changed source namespace before direct vignette rendering so extension-method registration validates the renamed generics before package installation.
• Validated by roxygen regeneration, direct rendering of the extending vignette, strict tests, package check, source build, and installation.

fitPS 1.1.2.005

• Simplified vignette filenames to basics, model-comparison, uncertainty, and extending, while removing the redundant simple_P_fit vignette.
• Kept Roux as the worked dataset inside the model-comparison vignette rather than encoding the dataset name in the filename.
• Added .gitignore and .Rbuildignore rules for transient LaTeX, knitr, and rmarkdown files created while rendering vignettes.
• Cleaned ignored vignette rendering products after direct rendering so validation leaves only deliberate source changes.
• Validated by direct vignette rendering followed by the full roxygen, strict test, package check, build, and install workflow.

fitPS 1.1.2.003

• Allowed Bayesian fitting to use surveys with a single occupied support value instead of rejecting them during shared observation conversion.
• Preserved the historical MLE singleton-support rejection while moving that restriction into the MLE fitting layer.
• Added an explicit Bayesian warning that singleton-support posterior inference may be strongly influenced by the prior.
• Added regression coverage using the Pettard S survey, whose six observations are all S1, across zeta, logarithmic, and zero-inflated zeta Bayesian paths.
• Stage 10.2.1 repaired the warning-capture and numeric-type assertions exposed by the initial validation attempt.
• Validated with roxygen regeneration, strict tests, and package check under the full fitPS stage workflow.

fitPS 1.1.2.001

• Added a Roux-data model-comparison vignette covering the built-in zeta, zero-inflated zeta, and logarithmic models through the public fit() API.
• Compared maximum-likelihood fits with log-likelihood, AIC, and BIC, and Bayesian numerical fits with DIC without treating the criteria as interchangeable.
• Added observed-versus-fitted probability diagnostics and clarified plug-in probabilities, posterior parameter uncertainty, and posterior predictive probabilities.
• Kept Stage 10.1 documentation-only; Bayesian Bootstrap implementation remains deferred to later Stage 10 work.
• Validated by directly rendering the executable vignette with the package vignette toolchain; full package-wide validation was intentionally skipped for this vignette-only stage.

fitPS 1.1.1.021

• Completed the Stage 9 Bayesian architecture audit and confirmed that external models can use fitPS-owned numerical and MCMC posterior engines through the public model mathematics contract.
• Recorded the numerical-engine policy of deterministic integration for one- and two-parameter models and MCMC for models with three or more parameters.
• Recorded specialised built-in MCMC methods as compatibility overrides rather than extension requirements, and kept Laplace and importance as secondary existing capabilities.
• Confirmed external Poisson and Poisson-normal regression coverage for posterior summaries, DIC, P/S support mapping, serialization, and deliberate unsupported-engine failures.
• No installed package behaviour changed; package validation was intentionally skipped because this stage changes only development audit documentation.

fitPS 1.1.1.020

• Confirmed that zeta, logarithmic, and zero-inflated zeta models all satisfy the generic MCMC model contract.
• Retained specialised built-in MCMC samplers as compatibility overrides where established stochastic behaviour remains part of the regression contract.
• Clarified that external and future models do not need model-specific sampler implementations because the generic MCMC fallback is complete.
• Added deterministic regression coverage for the built-in generic MCMC fallback and documented the Stage 9.8 consolidation decision.

fitPS 1.1.1.019

• Updated all executable vignettes to teach the generic fit() model-descriptor API rather than deprecated distribution-specific fitting wrappers.
• Updated the Bayesian vignette to present deterministic numerical integration for one- and two-parameter models and MCMC as the core simulation route, while keeping Laplace and importance sampling secondary.
• Expanded the external-model vignette to document the public Bayesian prior, control, transform, numerical, and MCMC extension contract with Poisson and Poisson-normal examples.
• Documented stable standard-Normal evaluation of the Poisson-normal marginal probability and the numerical-engine policy that models with three or more parameters use MCMC.
• Validated the documentation-only change by rendering every changed vignette directly with its established R Markdown output format.

fitPS 1.1.1.018

• Consolidated fragmented psData methods and operations into R/psData-methods.R without changing their interfaces or calculations.
• Consolidated small psFit fitted, log-likelihood, print, and summary methods into R/psFit-methods.R.
• Grouped survey-comparison helpers and bootstrap inference wrappers into cohesive source files while preserving public behaviour.
• Left the large zeta and ZIZ model files unchanged pending a separate decision about specialised versus generic MCMC implementations.
• Regenerated documentation and validated the behaviour-preserving source-layout cleanup with the strict fitPS test and check workflow.

fitPS 1.1.1.017

• Extended generic numerical posterior fitting to two-parameter models with adaptive cubature::hcubature() integration over model-supplied natural-scale bounds.
• Adopted a dimensional policy of deterministic numerical fitting for one and two parameters and MCMC for models with three or more parameters.
• Migrated ZIZ posterior normalization, moments, and DIC to the generic two-dimensional engine while retaining a compatibility grid for established plotting helpers.
• Stabilized the external Poisson-normal proof likelihood by integrating over a standard-Normal latent variable rather than a collapsing narrow Normal density.
• Added targeted numerical regression validation through devtools::test() so the package and test helpers are loaded before the full suite runs.
• Added cubature as a package import for core two-dimensional numerical integration and kept Laplace and importance as secondary existing capabilities.
• Validated through targeted numerical tests, roxygen regeneration, strict full tests, package check, source build, and installation.

fitPS 1.1.1.011

• Made one-dimensional numerical posterior fitting generic through the public model likelihood, prior, and Bayesian-control contract.
• Migrated zeta and logarithmic numerical fitting and probability summaries onto the shared numerical engine while preserving their model mathematics.
• Added external Poisson numerical fitting with infinite natural-scale support and conjugate posterior regression checks.
• Updated the external-model regression contract to recognise both numerical and MCMC support for the Poisson proof model.
• Added the explicit stats::setNames roxygen import required by the generic numerical engine.
• Kept ZIZ numerical fitting model-specific because its established posterior representation is two-dimensional.
• Validated through roxygen regeneration, strict tests, package check, source build, and installation.

fitPS 1.1.1.008

• Completed the multi-parameter external Bayesian proof using the Poisson-normal test model through the generic MCMC contract.
• Verified posterior parameter summaries, generic derived probability summaries, fitted probabilities, and DIC for an external two-parameter model.
• Verified matched P- and S-survey data retain identical zero-based support behaviour through Bayesian fitting.
• Verified serialization preserves the external model descriptor, MCMC representation, posterior summaries, and fitted values.
• Kept Laplace and importance as secondary existing capabilities without expanding their public extension role.
• Validated through roxygen regeneration, strict tests, package check, source build, and installation.

fitPS 1.1.1.007

• Added a model-neutral random-walk Metropolis fallback so external psModel classes can use fitPS-owned MCMC posterior orchestration.
• Let external Bayesian models supply model-specific prior objects through modelLogPrior() rather than requiring the legacy one-dimensional psPrior representation.
• Proved the generic Bayesian path with deterministic external Poisson and two-parameter Poisson-normal models, including transformed positive parameters.
• Updated the public-model regression expectation so missing external priors now fail with the new explicit model-specific-prior contract rather than the retired MLE-only error.
• Preserved the specialised zeta, ZIZ, and logarithmic MCMC implementations and kept Laplace and importance as secondary existing capabilities.
• Validated through roxygen regeneration, strict tests, package check, source build, and installation.

fitPS 1.1.1.005

• Added a public Bayesian model contract for complete prior evaluation, model-owned Bayesian controls, parameter transformations, and working-scale Jacobians.
• Added identity transform defaults so unconstrained external models can use the contract without transformation boilerplate.
• Exposed existing zeta, logarithmic, and zero-inflated zeta prior and transformation mathematics through the new public S3 methods without changing their current Bayesian fitting paths.
• Added deterministic contract tests for external methods, built-in prior evaluation, transform round trips, Jacobians, and Bayesian starting values.
• Documented method-specific Bayesian prior and starting-value arguments so generated Rd usage and argument sections remain synchronized.
• Generic external-model Bayesian fitting remains intentionally deferred until posterior engines are migrated to consume the new contract.

fitPS 1.1.1.002

• Repaired the Stage 9.1 version formatter, which had recycled the build component and produced an invalid six-part version; build 001 remains consumed by that attempted stage.

• Audited the Bayesian architecture for external model extensibility and defined the minimum public model contract needed for fitPS-owned posterior engines.

• Identified model-specific posterior fitting dispatch, scalar prior assumptions, and built-in parameter transformations as the main barriers to third-party Bayesian models.

• Recorded a staged implementation path beginning with public prior, Bayesian-control, and parameter-transformation methods, followed by generic MCMC proof models.

• No installed package behaviour changed; package validation was intentionally skipped because this stage changes only development audit documentation.

fitPS 1.1.0.018

• Completed the Stage 8 public model API audit and repaired the multi-parameter external-summary regression assertion.
• Verified external Poisson and integral-based Poisson-normal models use only the public extension contract, including P/S support shifting without truncation.
• Preserved deprecated compatibility fitters while keeping zeta, ZIZ, and logarithmic implementation consolidated in model-specific source files.
• Documented that generic third-party fitting is MLE-only at this boundary; unsupported Bayesian external fitting fails explicitly.
• Preserved generic print/summary presentation for external MLE models and validated with roxygen regeneration, strict tests, package check, source build, and installation.

fitPS 1.1.0.015

• Deprecated fitDist(), fitZIDist(), and fitlogDist() as compatibility entry points and direct users to the generic fit() interface.
• Kept legacy fitters functional through common model-oriented dispatch and explicitly regression-tested their retained model metadata.
• Consolidated distribution-specific zeta, zero/one-inflated zeta, and logarithmic implementations into model-zeta.R, model-ziz.R, and model-logarithmic.R.
• Renamed shared numerical, MCMC, Laplace, and importance posterior code as model-neutral posterior-engine source files.
• Corrected the external Poisson and Poisson-normal examples so S surveys shift the zero-based probability sequence from P0, P1, … to S1, S2, … without truncation or renormalisation.
• Expanded extension documentation and tests to make P/S support mapping an explicit part of the public model contract.
• Validated with roxygen regeneration, strict tests, a strict package check, source build, and installation.

fitPS 1.1.0.010

• Corrected the external Poisson-normal example to use a latent normal log-rate integrated out of the Poisson likelihood rather than the Hermite distribution.
• Added the Poisson probability mass function, weighted likelihood, Poisson-normal marginal integral, marginal likelihood, and moment relationships to the public extension vignette.
• Updated the external Poisson-normal tests to evaluate marginal probabilities by numerical integration while keeping all distribution-specific code outside R/.
• Preserved the public mu and sigma parameterization for the latent normal distribution and the generic fitPS MLE, prediction, serialization, and model-comparison interfaces.
• Validated with roxygen regeneration, strict tests, a strict package check, source build, and vignette build during package construction.

fitPS 1.1.0.009

• Corrected the external Poisson-normal example to use a latent normal log-rate integrated out of the Poisson likelihood rather than the Hermite distribution.
• Added the Poisson probability mass function, weighted likelihood, Poisson-normal marginal integral, marginal likelihood, and moment relationships to the public extension vignette.
• Updated the external Poisson-normal tests to evaluate marginal probabilities by numerical integration while keeping all distribution-specific code outside R/.
• Preserved the public mu and sigma parameterization for the latent normal distribution and the generic fitPS MLE, prediction, serialization, and model-comparison interfaces.
• Validated with roxygen regeneration, strict tests, a strict package check, source build, and vignette build during package construction.

fitPS 1.1.0.008

• Replaced the earlier maintainer-facing extension vignette with a public guide for defining models outside fitPS.
• Added worked external Poisson and Poisson-normal examples using only the exported psModel contract and generic fit() interface.
• Documented downstream S3 registration, model-specific MLE controls, retained model objects, prediction, serialization, and likelihood-based model comparison.
• Kept all Poisson and Poisson-normal distribution mathematics outside fitPS package implementation code.
• Validated with roxygen regeneration, strict tests, a strict package check, source build, and vignette build during package construction.

fitPS 1.1.0.007

• Demonstrated the public model-extension contract with a second distribution defined entirely outside fitPS package code.
• Added an external two-parameter Poisson-normal (Hermite) test model parameterized by mu and sigma, with model-specific mathematics and S3 registration confined to test helper code.
• Confirmed the unchanged generic MLE path handles coupled parameter constraints, fitted probabilities, prediction, log likelihood, deviance, AIC, BIC, and serialization for a multi-parameter external model.
• Added no distribution-specific implementation under R/; Stage 8.4 therefore tests the Stage 8.3 public API without extending fitPS internals.
• Validated with roxygen regeneration, strict tests, and a strict package check.

fitPS 1.1.0.006

• Added a supported public psModel() constructor and exported the minimal S3 generics needed for third-party model extensions.
• Added fitPS-owned generic maximum-likelihood optimisation using model-declared starting values and bounds, with optional modelMleControl() overrides for more demanding models.
• Demonstrated true external extensibility with a Poisson model defined entirely in test helper code using public APIs only.
• External Poisson fits now participate in fitted probabilities, point prediction, log likelihood, deviance, AIC, BIC, and fitted-object serialization without modifying fitPS source code.
• Kept Bayesian support for arbitrary external models deliberately out of the initial contract until posterior-engine compatibility is defined explicitly.
• Validated with roxygen regeneration, strict tests, and a strict package check.

fitPS 1.1.0.004

• Added the model-oriented fit() entry point for built-in zeta, zero-inflated zeta, and logarithmic model objects.
• Fixed generic fit() method dispatch so its default method vector is resolved to the scalar mle choice before delegation to model-specific fitters.
• Corrected the built-in model interface regression test so list names retained by vapply() do not cause a false failure when the fitted model identifiers are otherwise identical.
• New psFit objects retain their originating model descriptor while preserving the established character model identifier.
• Model-aware operations now prefer the retained descriptor and fall back to built-in name reconstruction for legacy fitted objects.
• Exported the built-in model constructors needed by the new fitting interface; third-party model construction remains deferred until the external Poisson contract is validated.
• Validated with roxygen regeneration, strict tests, and a strict package check.

fitPS 1.1.0.001

• Audited the internal model protocol and defined the Stage 8 public-extension direction.
• Established external Poisson and Poisson-normal models as proof-of-concept targets that must work without rebuilding or modifying fitPS.
• Chose retained model descriptors with legacy name fallback as the preferred route away from hard-coded model reconstruction.
• Started the Stage 8 development series at 1.1.0.xxx; package validation was intentionally skipped because this stage changes only a development audit document.

fitPS 1.0.9.007

• Completed the Stage 7 model-comparison and distribution-extensibility audit across zeta, ZIZ, and logarithmic models.
• Corrected the cross-model audit test to accept the numeric nobs attribute returned by the shared logLik contract.
• Added durable cross-model regression tests for the shared MLE comparison contract and intentional posterior-engine support matrix.
• Confirmed that logarithmic integration reuses the model/engine abstractions without distribution-name branches in generic posterior engines.
• Documented that the current extension contract remains internal and recommended a later deliberate third-party model API rather than exporting evolving internals piecemeal.
• Recorded a future thin fit(data, model = …) front end as the natural place to accept external model objects while retaining existing fitters.
• Validated with vignette rendering, roxygen, strict tests, and strict package check.

fitPS 1.0.9.005

• Added the developer-facing Extending fitPS with a New Distribution vignette using the logarithmic-series model as the worked example.
• Documented the psModel extension contract for parameters, probabilities, likelihoods, supported posterior engines, fitting, posterior summaries, and model comparison.
• Documented the expected behavioural tests for new distributions and the current boundary between internal package extensibility and a future public model-registration API.
• Recorded the possible future fit(data, model = …) front end while retaining fitDist() and fitZIDist() as compatibility and convenience interfaces.
• Validated the vignette by rendering it in addition to the strict fitPS test and package-check workflow.

fitPS 1.0.9.004

• Added the logarithmic-series distribution as a psModel with shared P/S support, probability, and likelihood dispatch.
• Added numerical and MCMC Bayesian logarithmic fits through the existing posterior-engine extension points and common psPosterior finalisation path.
• Enabled logarithmic MLE fits to participate in logLik, deviance, AIC, BIC, fitted probabilities, prediction, and standard psFit summaries.
• Enabled DIC for one-parameter numerical and MCMC models by removing residual zeta-specific parameter-name assumptions.
• Generalised uniform/custom prior ranges while preserving zeta-domain safeguards for zeta and ZIZ posterior fits.
• Replaced the provisional logarithmic fitter internals rather than treating them as a compatibility contract.
• Validated by the strict fitPS stage workflow.

fitPS 1.0.9.002

• Added a shared model-comparison contract for likelihood, deviance, and parameter counting.
• Enabled AIC and BIC for maximum-likelihood psFit objects through the common logLik interface.
• Added DIC for numerical, MCMC, and importance Bayesian posterior representations; Laplace fits fail explicitly until a posterior deviance expectation is available.
• Preserved the Bayesian-bootstrap proposal as future work under a non-stage-specific context and recorded the future direction toward a model-oriented fit interface.
• Corrected the DIC contract test to compare the numeric criterion separately from its Dbar, Dhat, and pD attributes.
• Validated by the strict fitPS stage workflow.

fitPS 1.0.8.014

• Completed the final Bayesian architecture and regression audit for the model/engine migration.
• Added durable tests for the supported model/engine matrix, removal of provisional fitter paths, and model-descriptor extensibility.
• Confirmed all named R functions remain documented and removed the last migration-stage wording from internal roxygen text.
• Recorded logarithmic-distribution extensibility and AIC/DIC model comparison as post-migration follow-on work.
• Re-ran the full fitPS validation workflow before closing Stage 6.

fitPS 1.0.8.013

• Completed the provisional Bayesian cleanup and repaired controlled Git staging for deleted generated documentation.
• Kept psPosterior as the source of truth for ZIZ posterior representations, probability summaries, and diagnostics.
• Preserved established pre-1.0.7 plain-zeta Bayesian compatibility fields behind the model-specific compatibility method.
• Carried forward the full Stage 6.10 cleanup because the failed attempt did not commit.
• Re-ran the full fitPS validation workflow before committing the repaired cleanup.

fitPS 1.0.8.012

• Removed superseded provisional Bayesian fitter wrappers and routed fitDist() and fitZIDist() through the shared model/engine orchestration.
• Made psPosterior the source of truth for ZIZ posterior representations, probability summaries, and diagnostics instead of duplicating native engine payloads at psFit top level.
• Preserved established pre-1.0.7 plain-zeta Bayesian compatibility fields behind a model-specific compatibility method.
• Removed dead migration helpers and consolidated inflation calculations directly into representation-specific S3 methods.
• Validated with the full fitPS stage workflow because package R source, S3 registration metadata, generated documentation, and tests changed.

fitPS 1.0.8.011

• Consolidated Bayesian psFit finalisation and psPosterior construction across plain zeta and zero-inflated zeta models.
• Added the common psPosterior contract and posterior mean probability summaries to plain-zeta numerical and MCMC fits.
• Standardised psPosterior representation storage on typed engine wrappers while retaining legacy top-level representation fields during migration.
• Replaced inflation method-name switching with S3 dispatch on posterior representation classes and centralised derived probability summarisation.
• Validated with the full fitPS stage workflow because package R source, S3 registration metadata, and tests changed.

fitPS 1.0.8.010

• Migrated the existing zero-inflated zeta Laplace and importance-sampling approximations onto the shared model and posterior-engine architecture.
• Preserved the transformed-coordinate Laplace optimisation, Hessian/delta-method covariance, importance proposal, seeded weighted sampling, posterior summaries, and fitted probability behaviour.
• Added engine-level fitting, summary, diagnostic, and point-estimate methods while keeping plain-zeta Laplace and importance combinations explicitly unsupported.
• Added deterministic P and S regression coverage using stable behaviour-based test names and descriptions.
• Validated with the full fitPS stage workflow because package R source, S3 registration metadata, and tests changed.

fitPS 1.0.8.009

• Completed the zero-inflated zeta MCMC migration onto the shared model and posterior-engine architecture while preserving the established Metropolis-Hastings sampler and seeded RNG sequence.
• Registered the internal Stage 6 S3 methods explicitly so roxygen documentation generation succeeds without exposing those helpers as user-facing API.
• Preserved the historical ZIZ MCMC fitted-probability component as a named numeric vector rather than a one-row matrix.
• Kept the behaviour-based test filename cleanup and strengthened deterministic P and S regression coverage for the migrated ZIZ MCMC path.
• Validated with the full fitPS stage workflow because package R source, S3 registration metadata, and tests changed.

fitPS 1.0.8.007

• Migrated numerical zero-inflated zeta Bayesian fitting onto the shared Stage 6 model and posterior-engine protocols while preserving the existing two-dimensional posterior grid calculation.
• Added the ZIZ model log-likelihood method and numerical-engine model dispatch while reusing the shared numerical summary, diagnostics, and point-estimate methods.
• Retained the legacy posterior grid, marginal densities, posterior probability summaries, parameter moments, fitted probabilities, and psPosterior representation while also storing the engine wrapper.
• Added deterministic P and S regression tests against the pre-migration numerical ZIZ orchestration; ZIZ MCMC, Laplace, and importance fitting remain unmigrated.
• Validated with the full fitPS stage workflow because package R source and tests changed.

fitPS 1.0.8.006

• Migrated the plain-zeta MCMC Bayesian fit onto the Stage 6 model and posterior-engine protocols while preserving the existing sampler calculation and RNG ordering.
• Added MCMC-engine fit, summary, diagnostics, point-estimate, and representation methods for zetaModel.
• Retained the legacy chain, posterior-mean shape, variance, fitted probabilities, and spline density while also storing the engine representation.
• Added deterministic P and S regression tests against the pre-refactor MCMC algorithm; ZIZ MCMC fitting remains explicitly unmigrated.
• Validated with the full fitPS stage workflow because package R source and tests changed.

fitPS 1.0.8.005

• Migrated the plain-zeta numerical Bayesian fit onto the Stage 6 model and posterior-engine protocols while preserving the existing integration calculation.
• Added the zeta model log-likelihood method and numerical-engine fit, summary, diagnostics, point-estimate, and representation methods.
• Kept fitted P/S probabilities evaluated at the posterior mean shape and retained the legacy numerical posterior density component for compatibility.
• Added deterministic regression tests against the pre-refactor numerical algorithm for both P and S data; ZIZ numerical fitting remains explicitly unmigrated.
• Validated with the full fitPS stage workflow because package R source and tests changed.

fitPS 1.0.8.004

• Added internal S3 psModel descriptors for plain zeta and zero-inflated zeta models.
• Added posterior-engine descriptors and common fit, summary, diagnostics, point-estimate, and representation protocols without migrating existing fitting paths yet.
• Declared the supported model/engine matrix explicitly so unsupported combinations fail clearly rather than being inferred from method switches.
• Consolidated duplicated P/S probability-index validation, latent-support conversion, term naming, and observation mapping shared by zeta and ZIZ code.
• Added deterministic offline protocol tests; validated with the full fitPS stage workflow because package R source and tests changed.

fitPS 1.0.8.003

• Completed the Stage 6.2 internal roxygen2 documentation baseline and Bayesian rationale comments.
• Repaired shared summary help topics by documenting the x arguments contributed by print.summary.psBootstrap() and print.summary.psPosterior().
• Preserved executable R statements while carrying forward the complete Stage 6.2 documentation change set after its pre-commit check failure.
• Added the Stage 6.2.1 repair note explaining the roxygen @describeIn argument-documentation issue.
• Validated with the full fitPS stage workflow because package R source files changed.

fitPS 1.0.8.001

• Began Stage 6 as the fitPS Bayesian architecture, documentation, and consolidation series.
• Added the Stage 6.1 audit defining orthogonal S3 model and posterior-engine abstractions for plain zeta, zero-inflated zeta, and future distributions.
• Recorded consolidation targets for shared observation handling, probability naming, prior validation, posterior summaries, and Bayesian psFit construction.
• Reserved the 1.0.8.xxx version series for Stage 6, with every runner attempt consuming a new build number.
• Renumbered the proposed Bayesian-bootstrap context as future Stage 7.1 work so it remains separate from the architecture refactor.
• Stage 6.1.1 repairs the Stage 6.1 delivery workflow and installs the same planning documentation; no R package validation is claimed for this planning stage.

fitPS 1.0.7.018

• Added posteriorInflation() for Pr(pi < epsilon | data) under Bayesian zero-inflated zeta fits.

• Implemented the practical-inflation probability consistently for numerical, MCMC, importance, and Laplace posterior engines.

• Added inflationEpsilon to summary.psPosterior(), defaulting to 0.01, and report the resulting practical-negligibility probability.

• Replaced the vignette inflation discussion with an epsilon-based interpretation using the Roux footwear example and epsilon = 0.01.

• Documented that epsilon is application-specific and that the diagnostic differs from assigning posterior mass to the exact no-inflation model.

• Added deterministic engine-specific tests and a Roux numerical-posterior regression test.

  • Repaired summary compatibility for posterior objects without a model tag, corrected Roux grid-option placement, and updated psPosterior structure expectations.

• Validated through the full fitPS documentation, test, check, vignette, build, and installation workflow. ## fitPS 1.0.7.016

• Corrected the Roux vignette description to identify the response as the number of different glass sources found on each surveyed pair of shoes.

• Added posterior interpretation of the zero/one-inflation parameter, including the distinction between concentration near zero and formal posterior probability of a no-inflation model.

• Added the Efron bootstrap/Bayesian connection to the bootstrap discussion while preserving the distinct inferential interpretations.

• Added future-development context for Rubin’s Bayesian bootstrap, centred on weighted-likelihood support and a possible psBayesianBootstrap object.

• Added long-term architecture context for distribution-pluggable fitPS models and evaluation of the distributional package as a possible distribution representation layer.

• Validated through the full fitPS documentation, test, check, vignette, build, and installation workflow. ## fitPS 1.0.7.015

• Added a dedicated Bayesian and bootstrap probability vignette using the real Roux et al. footwear survey supplied with fitPS.

• Documented plug-in, bootstrap-mean, and posterior-mean P probabilities and their distinct frequentist and Bayesian interpretations.

• Documented the numerical, MCMC, Laplace, and importance posterior engines, including priors, importance weighting, and Laplace simulation for derived probabilities.

• Added Roux-based README examples for posteriorProbs(), bootstrapProbs(), fitted(), plotting, and bootstrapFit().

• Expanded Bayesian probability help and fitZIDist() cross-references to the public uncertainty APIs.

• Added an end-to-end real-data regression test covering MLE, numerical Bayesian, bootstrap, fitted probabilities, interval bounds, and legacy probfun() behaviour.

• Added the Stage 5.11 final regression audit documenting completed interfaces and optional future cleanup only.

• Validated through roxygen, strict offline tests, strict package check, vignette build, source build, and installation. ## fitPS 1.0.7.014

• Harmonised predict.psFit() across plug-in, posterior-mean, and bootstrap-mean probability definitions while preserving plug-in predictions as the default.

• Added equal-tailed credible prediction intervals for stored posterior summaries and percentile confidence intervals for stored bootstrap summaries.

• Required posterior and bootstrap interval levels to match the level stored in the corresponding uncertainty object, avoiding unsupported implicit recomputation.

• Updated print.psFit() to route fitted probabilities through fitted(), fixing the legacy long-output path that could ignore zero inflation.

• Added concise posterior and bootstrap printing with optional nterms selection and bootstrap-aware psFit summaries.

• Clarified that probfun() evaluates plug-in probabilities and directs users to posteriorProbs() or bootstrapProbs() for uncertainty-distribution means.

• Added deterministic offline regression tests for prediction semantics, intervals, term selection, printing, and summary presentation.

• Validated through roxygen, strict offline tests, and strict package check. ## fitPS 1.0.7.012

• Added bootstrapFit() as the public entry point for attaching a psBootstrap distribution to an MLE fit.

• Added bootstrapProbs() methods for psFit and psBootstrap objects with the same P/S term selection rules as posteriorProbs().

• Added fitted(…, type = “bootstrapMean”) while preserving plug-in fitted probabilities as the default.

• Added plot.psBootstrap() for bootstrap mean probabilities and percentile confidence intervals.

• Kept Bayesian posteriorMean fitted values and all existing frequentist defaults unchanged.

• Added deterministic offline API, indexing, fitted-value, plotting, and validation tests.

• Validated through roxygen, strict offline tests, and strict package check. ## fitPS 1.0.7.011

• Added an internal psBootstrap S3 object for frequentist bootstrap parameter and P/S probability distributions.

• Added a reusable nonparametric bootstrap replicate engine with seeded reproducibility and failed-fit diagnostics.

• Added bootstrap mean, standard deviation, and percentile confidence intervals for parameter and probability replicates.

• Reused the existing zero-inflated probability transformation and added the matching standard-zeta P/S transformation.

• Kept bootCI() return structures backward compatible while routing its replicate generation through the shared engine.

• Added deterministic offline tests for zeta and zero-inflated bootstrap summaries, reproducibility, failures, and legacy bootFit() shapes.

• Validated through roxygen, strict offline tests, and strict package check. ## fitPS 1.0.7.009

• Added the Stage 5.7 planning audit for bootstrap distributions of fitted P and S probabilities.

• Recommended a psBootstrap S3 object that parallels psPosterior while preserving the distinct frequentist interpretation.

• Documented reuse of retained bootstrap parameter replicates, zizProbabilities(), percentile probability intervals, failure diagnostics, and seeded serial reproducibility.

• Preserved bootCI() return values and ordinary plug-in fitted probabilities as compatibility constraints for the implementation stages.

• Planned Stages 5.8-5.11 for bootstrap objects, public bootstrap probability APIs, harmonised prediction, and final documentation.

fitPS 1.0.7.008

• Added plot.psPosterior() for posterior mean P or S probabilities with equal-tailed credible intervals.

• Supported plotting all stored terms, leading terms, or explicitly selected P and S indices through posteriorProbs().

• Kept plot.psFit() and plotPosterior() behaviour unchanged for backward compatibility.

• Added deterministic offline plotting tests for P and S posterior objects and interval suppression.

• Validated through roxygen, strict offline tests, and strict package check. ## fitPS 1.0.7.007

• Exported posteriorProbs() for Bayesian psFit and psPosterior objects.

• Added consistent selection of leading or explicitly indexed P and S posterior probability summaries.

• Extended fitted.psFit() with type = “posteriorMean” while preserving plug-in probabilities as the default.

• Kept posterior probability access unavailable for frequentist fits with clear errors.

• Added deterministic offline tests for S3 dispatch, subset selection, fitted-value semantics, and compatibility.

• Validated through roxygen, strict offline tests, and strict package check. ## fitPS 1.0.7.006

• Added the psPosterior S3 class as the coherent posterior component of Bayesian psFit objects.

• Stored parameter summaries, posterior probability summaries, engine-specific representations, interval level, and diagnostics under fit$posterior.

• Preserved chain, posteriorGrid, weightedSamples, laplace, importance, and posteriorProbs fields for backward compatibility.

• Added print, summary, and fitted methods for psPosterior objects and delegated Bayesian psFit summaries to the posterior object.

• Corrected the Stage 5.3 tests by comparing probability estimates without incidental names and by using valid compact S-survey data.

• Validated through roxygen, strict offline tests, and strict package check. ## fitPS 1.0.7.004

• Repaired the existing parallel import documentation in bootCI so every @importFrom tag occupies one physical line.

• Removed the roxygen warning reported during Stage 5.2 documentation regeneration without changing package behaviour.

• Scanned the remaining R sources and found no other wrapped @importFrom continuation lines.

• Validated through roxygen, strict offline tests, and strict package check. ## fitPS 1.0.7.003

• Added one shared internal transformation from zero-inflated zeta parameters to requested P or S probabilities.

• Used the package standard zeta shape parameterisation and supported vectorised posterior parameter values.

• Added deterministic tests for P and S formulas, names, validation, scalar recycling, and probability-mass truncation.

• Removed and ignored generated Rplots.pdf test artifacts.

• Corrected stage-runner source-package path capture so build console output cannot be mistaken for a filename.

• Validated through roxygen, strict offline tests, and strict package check. ## fitPS 1.0.7.002

• Repaired the zero-inflated zeta Metropolis-Hastings sampler so each component update evaluates only the current state and its proposed component.

• Implemented the beta-prior and independence-proposal cancellation explicitly for pi updates and used the selected psPrior for shape updates.

• Preserved legacy a/b shape-prior bounds when no explicit prior is supplied and added seeded reproducibility support.

• Added non-uniform Beta(2, 5) tests comparing MCMC posterior means and variances with deterministic numerical integration.

• Updated the stage runner delivery to use a single outer bundle with an exact sibling change-set and optional ChatGPT bundle creation.

• Validated through roxygen, strict offline tests, and strict package check. ## fitPS 1.0.6.013

• Rationalise the Bayesian fitting API around method = “bayes” and bayesOptions$posteriorMethod.

• Add deprecated legacy aliases for numerical integration, MCMC, Laplace, and importance posterior methods.

• Preserve compatibility by translating legacy Bayesian method values into canonical bayesOptions.

• Add tests for Bayesian method alias translation and default option handling.

• Validated by the Stage 4.6.1.1 full package runner.

fitPS 1.0.6.011

• Added an internal importance-sampling helper for zero-inflated zeta Bayesian posteriors using the Laplace approximation as a proposal.
• Returned weighted samples, posterior means, covariance estimates, and diagnostics including effective sample size and maximum normalized weight.
• Added deterministic offline tests for weighted samples, seeded reproducibility, and fitZIDist dispatch through bayesOptions$posteriorMethod = “importance”.
• Added ignore/build-ignore coverage for generated built-package path files so package checks do not inspect runner artifacts.
• Ran the full package validation workflow for this package-impacting stage.

fitPS 1.0.6.010

• Repaired the Stage 4.4 numerical posterior grid file by restoring the missing bracket in the returned pi value.
• Kept the Stage 4.4 Laplace approximation work unchanged while making the installed R code parsable for roxygen documentation.
• Preserved ignore/build-ignore coverage for generated built-package path files so package checks do not inspect runner artifacts.
• Ran the full package validation workflow for this package-impacting repair stage.

fitPS 1.0.6.006

• Added Bayesian options plumbing so method = “bayes” can select posterior engines through bayesOptions$posteriorMethod.
• Stripped inherited names from scalar transform inputs and outputs before constructing working-scale vectors.
• Routed fitDist() numerical integration through method = “bayes” with posteriorMethod = “numerical” while retaining method = “integrate” as a legacy alias.
• Kept zero-inflated zeta MCMC available through posteriorMethod = “mcmc” until the numerical Stage 4 posterior is implemented.
• Kept deterministic offline test coverage unchanged while repairing the helper implementation.

fitPS 1.0.6.001

• Added the Stage 4.1 design audit for deterministic and semi-deterministic Bayesian posterior approximations for zero-inflated zeta fits.
• Recorded the recommended path for transformed posterior helpers, grid quadrature, Laplace diagnostics, importance sampling diagnostics, and plotting integration.
• Preserved existing Bayesian MCMC behavior and user-facing defaults; this lightweight documentation stage did not run package validation.

fitPS 1.0.5.008

• Stabilised the generated stage runner so controlled paths are staged one at a time.
• Prevented absent optional directories from masking git add failures and leaving new files untracked.
• Preserved the controlled commit scope while keeping the workflow compatible with package roots that do not contain every optional directory.
• Kept the no-ChatGPT-bundle workflow and completed-stage archive cleanup behaviour.
• Validated by the strict fitPS stage workflow.

fitPS 1.0.5.007

• Added plotPosterior() for posterior-density plots from Bayesian psFit objects.
• Supported zeta MCMC chains, zeta numerical posterior density functions, and zero-inflated MCMC chains for shape and pi.
• Kept existing plot.psFit() fitted-probability plots unchanged for backward compatibility.
• Added deterministic offline tests using small fake psFit objects rather than long MCMC runs.
• Validated by the strict fitPS stage workflow.

fitPS 1.0.5.003

• Added Stage 3.2 workflow documentation for the Windows Ghostscript R CMD check NOTE under dev/.
• Kept strict package validation while allowing only the known missing-Ghostscript PDF-size reduction NOTE in the stage runner.
• Documented why the Ghostscript NOTE is a local check-environment issue rather than a package failure.
• Preserved the Stage 3 1.0.5.xxx build convention, including one consumed build number per attempt.
• Validated by the strict fitPS stage workflow.

fitPS 1.0.5.002

• Added the Stage 3.1 Bayesian posterior plotting audit under dev/.
• Recommended a separate plotPosterior() API for posterior parameter-density plots while preserving plot.psFit() for fitted probabilities.
• Documented support requirements for MCMC-backed zeta and zero-inflated zeta fits, integrated zeta posterior densities, credible intervals, deterministic tests, and shape-parameterisation safeguards.
• Reset the Stage 3 development series to the 1.0.5.xxx build convention.
• Validated by the strict fitPS stage workflow.

fitPS 1.0.4.018

• Migrated the user-facing vignette sources from Sweave .Rnw files to R Markdown .Rmd files.
• Removed generated vignette build artifacts from the staged source tree so they can be regenerated by the package toolchain.
• Added rmarkdown to Suggests for the R Markdown vignette builder.
• Preserved the Stage 2 zeta shape wording while making the vignettes easier to maintain.

fitPS 1.0.4.017

• Removed user-facing discussion of VGAM’s shifted zeta parameterisation from help and vignette source text.
• Kept documentation focused on the fitPS shape parameter, where shape > 1.
• Corrected spelling, grammar, and punctuation issues in the vignette source using NZ English conventions.
• Regenerated help with devtools::document() during the strict stage workflow.

fitPS 1.0.4.016

• Cleaned roxygen comments and README wording for the Stage 2 zeta shape parameterisation documentation.
• Fixed spelling, punctuation, and grammar in the fitDist(), fitZIDist(), and predict.psFit() help source.
• Kept documentation on NZ English wording, including parameterisation and optimisation.
• Left generated help files to be refreshed by devtools::document() during the stage workflow.

fitPS 1.0.4.015

• Updated documentation so shape consistently refers to the standard zeta parameter alpha with shape > 1.

• Removed obsolete VGAM-shifted wording from fitDist() and fitZIDist() help text.

• Documented that the default Bayesian prior is placed on log(shape - 1), giving support only on valid standard-shape values.

• Added README guidance for users who compare fitPS results with VGAM zeta functions.

• Validated by the strict fitPS stage workflow. ## fitPS 1.0.4.014

• Refactored zeta and zero-inflated zeta workflows so fitPS shape means the standard zeta parameter alpha with shape > 1.

• Converted to VGAM’s shifted parameter only at VGAM boundaries using shape - 1.

• Added active regression tests for fitting, probability functions, prediction, fitted values, and random generation under the standard-shape convention.

• Validated by the strict fitPS stage workflow.

fitPS 1.0.4.012

• Added Stage 2.1.1 corrective notes for the zeta shape parameterisation workflow.
• Removed active target-behaviour tests from tests/testthat because they intentionally fail before the Stage 2.2 refactor.
• Preserved the standard-shape expectations in dev/ so Stage 2.2 can reinstate them with the implementation repair.
• Validated by the strict fitPS stage workflow.

fitPS 1.0.4.010

• Added README.md with installation, data-format, fitting, prediction, comparison, and development-check examples.
• Updated .gitignore to exclude the fitPS built-package path marker produced by stage runners.
• Kept the stage focused on documentation and local-build hygiene without changing package APIs.
• Validated by the strict stage 1.4 package workflow before commit.

fitPS 1.0.4.009

• Added NEWS.md as durable package release notes rather than a stage transcript.
• Recorded the stage 1 stabilization series, including baseline tests, build hygiene, start-value handling, and zero-inflated prediction fixes.
• Kept entries version-numbered and newest-first for future fitPS maintenance.
• Validated by the strict stage 1.3 package workflow before commit.

fitPS 1.0.4 development series

• Added baseline testthat infrastructure for core psData and psFit workflows.
• Added deterministic tests for data construction, CSV import, model fitting, prediction, and logLik.psFit().
• Repaired stage check issues by documenting the fitDist() prior argument and adding required stats imports.
• Cleaned package-build inputs by excluding local development files, generated archives, stage runners, scripts, and generated vignette artifacts from R package builds.
• Repaired start-value handling for fitDist() and fitZIDist() while preserving backward-compatible shape aliases.
• Repaired zero-inflated zeta prediction logic so fitted predictions use the model object’s pi component.
• Validated through the staged fitPS workflow before completed stage commits.