inferguidR

Guided inference, power analysis and diagnostic accuracy in R.

inferguidR connects explicit statistical questions with validated analysis specifications, typed results and reproducible exports. It includes a method advisor, power and sample-size calculations, classification and cutoff tools, and an optional Shiny workbench.

Development software. Implemented methods differ in their validation status. A successful calculation is not proof of agreement with every external reference or suitability for a particular study. The package is not a CRAN release.

See the validation guide for evidence levels, reproducible public checks and known limitations, and the improvement roadmap for the numerical, product and release acceptance gates.

Installation

R 4.1.0 or later is required. Install the development version from GitHub:

install.packages("remotes")
remotes::install_github("CTTIR/inferguidR", build_vignettes = TRUE)

Building the tutorials needs knitr, rmarkdown and Pandoc. Set build_vignettes = FALSE if those are unavailable; the tutorial sources remain readable in this repository. Core calculations do not require a reference application, browser, database or external service.

For the optional workbench, install shiny and ggplot2. Some advisor-generated analysis scripts use additional packages named in their output; inspect the script and its requirements before running it.

Quick start

This hypothetical two-group design assumes independent normal observations, equal variances and a prespecified standardised difference. Its inputs illustrate the interface; they are not a recommended study design.

library(inferguidR)
spec <- analysis_spec(
  "t_means_two_independent", "a_priori",
  inputs = list(alpha = 0.05, power = 0.80, d = 0.5),
  options = list(tails = "two")
)
result <- solve_design(spec)
result$status
result$solved
result$achieved
result$validation
stopifnot(result$status$code == "ok")

Always inspect status, warnings, assumptions and validation before using a number. For this example, n_total is the total across both groups; the group_sizes field records the allocation. Unsupported, infeasible or unresolved calculations carry explicit statuses rather than a usable design estimate.

Save and reproduce

json <- ig_to_json(spec)
restored <- ig_from_json(json)
stopifnot(identical(ig_hash(spec), ig_hash(restored)))
file <- tempfile(fileext = ".json")
writeLines(json, file)
packageVersion("inferguidR")

Keep the specification, result, package revision and sessionInfo() together. Hashes identify content; they do not establish scientific correctness.

Function guide

Task	Entry points
Inspect available methods	list_methods(), ig_method()
Specify and calculate a design	analysis_spec(), solve_design(), calculate_power()
Compare design assumptions	sweep_design(), plot_design_sweep()
Evaluate diagnostic or classifier performance	classification_spec(), evaluate_cutoff(), classification_metrics()
Explore operating points	cutoff_sweep(), select_cutoff()
Work through an analysis question	advisor_session(), advisor_question(), advisor_answer()
Save specifications and results	ig_to_json(), ig_from_json(), ig_hash()
Explore interactively	launch_app()

Tutorials and workbench

• Guided analysis and reproducible power calculations.
• Classification metrics and explicit operating points.
• Understanding validation status.

After installing with vignettes, use vignette("getting-started", package = "inferguidR") or vignette("classification", package = "inferguidR").

Launch the workbench in an interactive R session:

# install.packages(c("shiny", "ggplot2"))
inferguidR::launch_app()

Troubleshooting and scope

An unknown method, an invalid probability or an inapplicable option is an input error: inspect ?analysis_spec and the selected method definition. A non-ok result is a calculation outcome to inspect, not a value to coerce into a number. Diagnostic examples are synthetic; selecting and evaluating a cutoff on the same observations does not establish out-of-sample performance.

Authors, citation and licence

Raban Heller. Use citation("inferguidR") for installed-version metadata and record the exact repository revision for a development analysis. The package is licensed under MIT. Report reproducible problems with synthetic inputs through GitHub issues.

Use of LLM tools

Where applicable, LLM tools were used for narrowly defined, non-authorial tasks: copyediting, prose refinement, Markdown/LaTeX formatting, boilerplate scaffolding (such as CI configurations and build scripts), and code refactoring. Tools included Chat AI, the KISSKI (GWDG) service, and self-hosted Mistral Small (24B, Apache-2.0) served through a local vLLM architecture on an Ubuntu Linux workstation with Intel processors and NVIDIA GPUs. The self-hosted model ran locally, with no data sent to third parties.

All scientific claims, methodological choices, analyses, interpretations and conclusions remain the authors' own. Where applicable, LLM-generated material must be reviewed and revised before inclusion, and references must be verified against their DOI, arXiv ID or ISBN.

Local software setup: vLLM with Python/PyTorch, plus R and RStudio.

Branding

Editable R artwork and regeneration instructions are in tools/branding.