tanner

• calculate_sds(..., type = "tv") no longer has a discontinuity at Prader bead boundaries: a raw ml reading is now mapped onto a continuous bead-index scale before interpolating, so both the low and high percentiles in the mid-P formula move together and the SDS varies smoothly with ml instead of jumping right at each bead threshold (e.g. 11.99ml vs 12.00ml).
• calculate_sds(..., type = "tv") now correctly returns NA for an NA age or stage, instead of a finite but wrong value. The continuous bead-index interpolation above used approx()'s default na.rm = TRUE, which silently drops NA (x, y) pairs from the interpolation table before interpolating -- so an NA age (which makes every real bead percentile NA) left only the two artificial boundary values (P(TV>=0)=1, P(TV>=Inf)=0) in the table, and interpolated between just those two instead of propagating the NA.
• interpolate_trajectory()'s returned stage component is now the same length as x/y/patient. It was previously built from the cumulative length of the output-so-far across all segments instead of the current segment's own length, making it longer than the other three components whenever a patient had more than one stage segment.

Documentation

• Documented that calculate_sds()'s return value is always the same length as age, and that stage must be the same length as age (not recycled) -- a length-1 stage, or any other length mismatch, is unsupported and can silently truncate, warn, produce NAs, or error depending on type.
• Added vignettes/tanner.Rmd, covering a fuller plot_stadia() walkthrough across its type/plotline/colors/overlay arguments, how raw tv (ml) readings are handled differently by calculate_sds() (continuous interpolation) versus plot_stadia() (snap to the nearest Prader bead curve), and the mid-P value SDS methodology of van Buuren & Ooms (2009) -- including why it is a discrete/ordinal method distinct from LMS/BCCG continuous-measurement SDS (e.g. centile::y2z()). Originally written as a Quarto vignette (tanner.qmd, %\VignetteEngine{quarto::html}); converted to classic knitr::rmarkdown (VignetteBuilder: knitr) after CI revealed an unresolved upstream Windows-only bug in the Quarto vignette engine (see "Continuous integration" below) that would have blocked CRAN submission, since CRAN's own build machines check Windows.
• Added a "Package structure" section to the README distinguishing the supported general-purpose library API (plot_stadia*, interpolate_trajectory, find_stage_segments, calculate_sds, tnologo) from the rApache webapp request handlers (plotter, plotterpro, upload_tryCatch_pro) kept only for the existing deployment. plotter(), plotterpro() and upload_tryCatch_pro() are now also grouped under a @family webapp-internals roxygen tag, so they cross-reference each other in ?tanner and on the pkgdown site.
• Added a pkgdown site (_pkgdown.yml), with a reference index split into "Plotting & analysis", "Reference data", and "Webapp internals" sections, and URL/BugReports fields in DESCRIPTION.
• Added full citations for van Buuren S (2014), "Growth charts of human development," *Statistical Methods in Medical Research*, 23(4), 346-368, in nl1997's source documentation.

Testing

• Added testthat unit tests for calculate_sds(), interpolate_trajectory(), find_stage_segments(), and smoke tests for the plot_stadia* drawing functions (which draw to a graphics device and return invisible NULL, so are tested by asserting they run without error). Extracted the tv ml-to-nearest-bead snapping logic out of plot_stadia_data() into an internal snap_tv_to_nearest_bead() helper so it can be unit tested directly.

Internal refactoring

• calculate_sds() and snap_tv_to_nearest_bead() each independently parsed the ml threshold from "T"-prefixed tv reference column names (e.g. "T12" -> 12); extracted the shared parsing step into an internal tv_ml_thresholds() helper. Pure refactor, no behavior change -- the bracket-and-interpolate vs. snap-to-nearest logic that follows at each call site is untouched.

Continuous integration

• Added .github/workflows/R-CMD-check.yaml (a standard R CMD check matrix across macOS/Windows/Ubuntu) and .github/workflows/pkgdown.yaml (builds and deploys the pkgdown site to GitHub Pages on push to master). windows-latest was briefly excluded from the check matrix after CI revealed that the quarto::html vignette engine hits an unresolved upstream Windows-only bug during R CMD build (the Quarto subprocess can't see the temporary library R CMD build just installed the package into; r-lib/rcmdcheck#236); re-added once the vignette was converted to classic knitr::rmarkdown (see "Documentation" above), which doesn't hit this bug since it renders in-process rather than via an external Quarto subprocess.
• _pkgdown.yml now sets template: bootstrap: 5 (Quarto-rendered articles require it; the default Bootstrap 3 made the pkgdown site build fail before it ever reached the deploy step) and llm-docs: false (pkgdown's llms.txt generation, not needed here and fragile against the runner's pandoc version).

Package loading

• library(tanner) no longer creates a plotdumpdir directory (or writes to disk at all) as a side effect of loading the package. The directory is now created lazily, on the first actual write inside plotter()/plotterpro()/upload_tryCatch_pro(); plotdumpdir remains exported and mutable.
• plotdumpdir now defaults to a "plotfiles" subdirectory of tempdir() instead of the hardcoded /tmp/plotfiles, and upload_tryCatch_pro()'s uploaded-file and .RData handoff files move there too (previously hardcoded to /tmp independently of plotdumpdir). This is a behavior change for the rApache deployment: tempdir() must resolve to a writable location that stays stable across the upload and plot requests for a given user session (true for a single long-running R process, which is how rApache normally hosts an R interpreter per worker). Unix-only /tmp was also a portability problem for R CMD check/CI on Windows and macOS runners, even though the webapp handlers themselves only ever ran on Linux.

Breaking changes

• Renamed three exported functions to consistent snake_case, matching plot_stadia()'s existing naming style:
• calculateSDS() -> calculate_sds()
• findXY() -> interpolate_trajectory()
• findSegments() -> find_stage_segments()

There are no backwards-compatible aliases for the old names; update any

code calling them directly.

• Renamed the two bundled reference datasets, following the <population><year> naming convention used by the nlreferences and clopus packages:
• pub.ref -> nl1997
• pub.ref.lines -> nl1997_lines