Study-Specific R References

Some R packages enter the repository through a named scientific study rather than a general-purpose live parity registry. Their role is narrower: they may own a reference calculation, establish a structural mapping, or identify an open computational boundary. The Evidence Book preserves that role without turning package presence in an R script into a blanket compatibility claim.

Reference	Governed role	Current claim	Explicit limit
nlme	GLS optimization over covariance structures created with APE	checked PGLS coefficients, likelihoods, and model comparisons for named primate fixtures	no standalone live nlme registry
treeio	node lookup and tree–trait joining in the primate longevity study	verified node identifiers and joined tip count for the governed fixture	no general tree-object API equivalence claim
tidytree	tabular tree representation in the primate reference workflow	package version and reference provenance retained with the structural comparison	no broad verb-by-verb parity registry
ggtree	course-study visualization surface	source fragments and plot intent retained	figures are plot-only; no pixel or aesthetic parity verdict
MCMCglmm	posterior computation in the cichlid mixed-model studies	source-backed external computation boundary	no governed reproduction of the PCM4 or PCM5 posterior core

Record Origin Matrix

Origin	Execution identity	Permitted statement
live R registry	current R process, package version, registered case and fixture	the selected cases passed, failed, or skipped under their observation rules
checked reference fixture	frozen output identity, producing environment, input identity, and comparison code	the named fixture corresponds under its checked rules
structural study record	source fragment, dataset, tree, and structural observations	the named join, node, or table contract was observed
plot-only source	source code and intended figure semantics	the visualization provenance is known; analytical or pixel correspondence is not claimed
external computation boundary	source and model identity without comparable execution	the intended method is located, but its numerical claim is not_comparable

These origins are not interchangeable. Re-running an adjacent native method does not turn an external boundary into a live registry result.

APE Plus nlme

The PGLS reference is a composition, not a claim about either package in isolation. APE constructs Brownian or Pagel-lambda correlation structures and nlme::gls fits the corresponding regression. Bijux compares the scientific outputs that the checked fixture declares: coefficients, log-likelihood, information criteria, estimated lambda where applicable, and selected diagnostics.

That comparison is reproducible from repository-owned inputs and frozen reference outputs. It is not labelled nlme-live, and a successful checked fixture does not establish coverage for every gls correlation structure, optimizer, missing-data policy, or variance model.

treeio, tidytree, And ggtree

The primate longevity dossier separates structural data correspondence from rendered presentation:

• treeio supplies reference node lookup and tree–trait joins.
• tidytree supplies the reference tabular tree representation and recorded package context.
• ggtree fragments describe the intended visual analyses, but remain plot-only because no governed image-comparison contract exists.

For the governed primate fixture, the structural comparison records a joined tip count of 75 and exact reference node identifiers for Pan_paniscus, Hylobates_lar, and Node32. Those observations support the named study claim; they do not imply comprehensive compatibility with Bioconductor tree objects.

MCMCglmm

PCM4 and PCM5 use MCMCglmm for Gaussian and generalized phylogenetic mixed models, including random-effect covariance, posterior summaries, and chain diagnostics. Bijux has adjacent native Bayesian, comparative-regression, and external MrBayes/BEAST capabilities, but none is presented as a substitute for that model family.

The Evidence Book therefore retains the source programs, datasets, governed claim records, and not_comparable verdicts while identifying MCMCglmm as the unresolved computational owner. Closing that boundary would require an independent governed runner, aligned priors and parameterization, posterior summary rules, convergence criteria, and reproducible comparison artifacts.

Promotion Into A Study Verdict

Mentioning or loading an R package is provenance, not verification. A study-specific result becomes a parity claim only when the repository can name the input, executed implementation, versions, compared observations, acceptance rule, artifacts, and verdict. Until then, the package remains a documented dependency or open boundary.

Reference-package evidence enters an Evidence Book verdict only when a bundle binds it to a claim ID, exact inputs, source and runtime identities, checks, eligible observations, limitations, freshness dependencies, and one governed verdict. The bundle must account for failed, skipped, excluded, and absent outputs rather than copying only successful rows.

A live harness case can remain passed while the study verdict is not_comparable because the study uses a different model, dataset, or claim. A checked fixture can support matched_with_tolerance without implying a live R execution on the reviewer's machine. State both records independently.