Skip to contents

Changelog

Source: NEWS.md

serodynamics (development version)

Internal

  • Updated .github/copilot-instructions.md so the local rjags install instructions branch by platform, matching what R-CMD-check.yaml and test-coverage.yaml do since #308: source builds on macOS/Linux, the binary on Windows. The file previously showed type = "source" for every platform (#309).

  • Removed the .ai-config git submodule, its .gitmodules entry, and the scheduled Bump submodule workflow (#316). The ai-config Claude Code plugin registered in .claude/settings.json supplies the same corpus without a second copy to keep pinned.

  • Granted pull-requests: write permission to the review dispatch job in claude-code-review.yml (#307), allowing acknowledgment comments to post on pull requests without GraphQL permission errors.

  • Hardened JAGS download and installer execution across CI workflows (R-CMD-check.yaml and test-coverage.yaml), using curl.exe with retry flags, start /wait for synchronous Windows installer execution, and platform- tailored rjags installation (type = "source" on macOS/Linux, binary on Windows).

  • Regenerated NAMESPACE and DESCRIPTION under roxygen2 8.1.0 (#288). The documentation check installs whatever roxygen2 is current rather than a fixed version, so the 8.1.0 release started rewriting both files against the 8.0.0 they were generated with, and the check failed on every pull request – including ones that touch no R code at all. The NAMESPACE change is formatting only: where several symbols come from one package, 8.1.0 groups them into a single importFrom() call instead of writing a line for each. Parsing both versions gives the same 22 exports, 27 imports and one S3 method. DESCRIPTION moves to Config/roxygen2/version: 8.1.0 and drops the older RoxygenNote field, which 8.1.0 no longer writes. The version is left floating rather than fixed, so a later roxygen2 release will need the same treatment.

  • Restored @claude review as a way to request a review (#285). Disabling the agent bot moved review dispatch into claude-code-review.yml behind a comment starting with /review, on the reasoning that the mention form belonged to claude.yml – which had just been switched off. So @claude review stopped doing anything at all, and did so silently: both jobs skipped, nothing went red, and the person asking got no reply. All three review requests made in the twelve days that followed used the mention form and were ignored; /review was never typed once. Both spellings now work. The mention is matched with Morrison-Lab/gha’s own detect-review-request action rather than a pattern of our own, so an agent task (@claude, please fix the failing test) is still not mistaken for a review request, and a mention quoted in a code span or a quoted line does not trigger one.

  • Replaced the silence with a reply when the @claude agent is addressed. A mention that is not a review request now gets a short comment saying the agent is switched off and naming the triggers that do work, since a skipped workflow is indistinguishable from a broken bot – which is why the gap above went unnoticed for so long.

  • Disabled the @claude agent bot. .github/workflows/claude.yml’s reactive triggers are commented out and its job carries if: false, so no comment, issue, or review event invokes the agent, and neither does a manual dispatch (the reusable workflow runs unattended on workflow_dispatch by design). Reviews are the only Claude capability left, and they run on request only: comment /review on a pull request. That path is new here – claude-code-review.yml gained an issue_comment trigger and a dispatch-on-comment job, since the @claude review mention it previously relied on went away with the agent.

  • Pointed the ai-config Claude Code plugin marketplace at Morrison-Lab. The corpus moved to a new GitHub organization and renamed the marketplace declared in its own .claude-plugin/marketplace.json. A plugin reference resolves by that declared name, so .claude/settings.json’s ai-config@d-morrison matched nothing and aborted plugin installation in cloud and web sessions opened on this repo. The clone URL was never the problem, since git follows GitHub’s transfer redirect; only the name lookup failed. .gitmodules and CLAUDE.md’s live links now point at the new organization as well.

  • Removed the local dispatch-explicit-review job from .github/workflows/claude.yml. It existed to cover @claude, please review, a phrasing the reusable workflow’s own pattern missed (#277), but that pattern has since been broadened upstream in d-morrison/gha#341. With both patterns live, a plain @claude review dispatched two paid review runs, which could review different heads because the local job had no needs: claude (closes #277, closes #276).

  • Added a project-level Claude Code skill, reprexes (.claude/skills/reprexes), capturing a workflow for isolating a problem into a minimal reproducible example and iterating fixes on it before porting them back (#239).

  • Embedded d-morrison/ai-config as a .ai-config git submodule, with a scheduled Bump submodule workflow to keep the pin fresh, and registered its Claude Code plugin marketplace in .claude/settings.json so cloud/web sessions opened on this repo load its skills (closes #264).

  • Added a scheduled Clean up PR Previews workflow that prunes closed-PR gh-pages previews and compacts gh-pages history, so deleted render snapshots stop bloating the repo (closes #260).

  • Added a CLAUDE.md review-guideline item flagging roxygen doc copy-paste (use @inheritParams/@inheritDotParams/@inheritSection instead) and manual argument relaying (use ... passthrough instead) (closes #262).

New features

  • Documented the CRAN installation option in the “Getting Started” article, which previously showed only the development-version install from GitHub. The article still installs the development version by default, since the released version exposes an older set of function names (#301).
  • Added plot_residuals() to visualize residuals over time, faceted by antigen-isotype. run_serodynamics() stores the original input data (and the stratification variable name) as original_data/strat attributes; plot_residuals() computes fitted and residual values on demand from those attributes via calc_fit_mod(), which returns 2.5%/97.5% posterior quantiles for each residual on both the natural scale (residual_low, residual_high) and the log10 scale (log_residual, log_residual_low, log_residual_high), which plot_residuals() uses to draw a precision interval around each point. fitted_residuals is no longer added as an attribute. (#230).
  • Added an exponential decay option for antibody decay curves via decay_type. (#252)
  • Added plot_serocurve() for graphical visualization of population-level serodynamic curves using posterior samples of the predictive newperson parameter distribution (or optionally the population level hyperparameter distributions). Supports 95% credible interval ribbons, stratified curves with color or faceting, and multiple antigen-isotypes (#74).
  • Renamed user-facing functions for clarity (#241):
  • Including optional population parameters as attributes in run_mod output. (#141)

Bug fixes

  • Fixed the development-version link in README.Rmd and README.md. The link pointed at the GitHub homepage rather than at the package repository (#301).
  • calc_fit_mod()’s output now covers all observations across all strata (previously only the last stratum was retained) and always includes a Stratification column ("None" when unstratified). (#240)
  • plot_residuals()/calc_fit_mod() now forward the decay_type attribute stored by run_serodynamics() through to ab(). Previously the argument was silently dropped, so exponential-decay models were fitted with the power-decay formula; since exponential decay’s shape is fixed at 1, this made every post-peak fitted value collapse to exactly 1. (#230)

Developer-facing changes

  • Cut down on run_serodynamics() tests to lower run time/load. Went from 5 separate run.jags chunks down to 3. (#253)
  • Documented in CLAUDE.md, .github/copilot-instructions.md, and a note in .lintr.R that dplyr::*_join() calls must specify the relationship argument (for example relationship = "many-to-one"), so an unexpected many-to-many match errors out instead of silently duplicating rows.
  • The test suite now sets options(lifecycle_verbosity = "error") (via tests/testthat/setup.R), so tidyverse lifecycle deprecations - including soft deprecations such as using the .data pronoun in a tidy-selection context - fail the tests instead of passing silently.
  • Updated the internals of calc_fit_mod() to use tidy-selection (all_of() and bare column-name strings) instead of the .data pronoun in select(), .by, and pivot_wider() contexts, removing a soft deprecation surfaced by the stricter test option above. No change to behavior or output.
  • The Claude Code Review workflow now skips (rather than fails) when a bot triggered the run, so a commit pushed by @claude or the Copilot agent no longer produces a red review check.
  • The Claude Code Review workflow now posts a fresh review comment per run and collapses the superseded ones as OUTDATED, so each push surfaces as new PR activity while older reviews fold up out of the way. @claude task comments are left untouched.
  • Added CLAUDE.md and expanded the Code Style Guidelines in .github/copilot-instructions.md to direct reviewers (human and AI) to flag unnecessarily convoluted or non-idiomatic code - in particular data-masking used in tidy-selection contexts and if/else branching that only varies which columns are selected, renamed, or joined.
  • Clarified Code Style Guidelines in .github/copilot-instructions.md: the UCD-SeRG Lab Manual takes precedence over the tidyverse style guide where they conflict, and functions should end with an explicit return() call per the lab manual / Google R Style Guide. This closes a gap where @claude reviews were flagging explicit returns as non-conforming.

serodynamics 0.1.0

CRAN release: 2026-06-02

This is the first CRAN release of serodynamics, a package for Bayesian hierarchical modeling of antibody kinetics from longitudinal serological data. It serves as the upstream companion to the serocalculator package.

New features

  • Reorganized pkgdown documentation with new “Getting Started” guide demonstrating main API workflow, organized articles into “Get started” and “Developer Notes” sections (#73).
  • Replacing old nepal_sees_jags_output data object with new run_mod() output (#102)
  • Including fitted_residuals values as data frame attribute in run_mod() output. (#101)
  • Adding class assignment to run_mod() output (#76)
  • Making prep_priors() allow for modifiable inputs in run_mod() (#78)
  • Exported run_mod() function (#22)
  • Added attributes as a return to the run_mod() function (#24)
  • Changes to run_mod() output:
    • Taking out include_subs as an input option, default will include all individuals with_post
    • all subjects now optionally included in curve_params output component, as specified by argument include_subs
  • A new run_mod() function that runs jags with option of stratification included. (#14)
  • Diagnostic plot_jags_Rhat() function to produce R-hat dotplots with stratification (#67)
  • Added plot_summ() function for summarizing estimates in a table (#74)
  • Diagnostic plot_jags_trace() function to create a trace plot with stratifications (#64)
  • Diagnostic plot_jags_effect() function to produce effective sample size plots with stratification (#66)
  • Diagnostic plot_jags_dens() function to produce density plots with stratification (#27)
  • Added plot_predicted_curve() with support for faceting by multiple IDs (#68)
  • Fixingnepal_sees SEES data and added jags_post for SEES (#63)
  • Added nepal_sees SEES data set data folder and documentation (#41)
  • Added postprocess_jags_output() to API (#33)
  • Added initsfunction() to API (#37)
  • Added participant IDs as names to nsmpl element of prep_data() output (#34)
  • Made “newperson” optional in prep_data() (#73)
  • Added initsfunction() to API (#37)
  • Added as_case_data() to API (#31)
  • as_case_data() now creates column visit_num (#47, #50)
  • Added prep_priors() to API (#30)
  • Added autoplot() method for case_data objects (#28)
  • Added examples for sim_pop_data(), autoplot.case_data() (#18)
  • Changed package name to serodynamics. (#19, #20)

Bug fixes

Developer-facing changes

  • Added platform-aware snapshots and darwin-specific variants for macOS platform differences (#73).
  • Updated Copilot instructions to encourage code decomposition and avoid copy-pasting substantial code chunks.
  • Expanded .github/copilot-instructions.md with additional guidance on evidence-based claims, Quarto markdown/cross-reference conventions, R style practices, and phrase-level line-break formatting for source text.
  • Added R 4.5+ snapshot variants to handle the changed attribute ordering in as_case_data(), ensuring test suite compatibility with R 4.5 and later (#109).
  • Added dev container configuration for persistent, cached development environment that includes R, JAGS, and all dependencies preinstalled, making Copilot Workspace sessions much faster.
  • Added .github/workflows/copilot-setup-steps.yml GitHub Actions workflow to automate environment setup for GitHub Copilot coding agent, preinstalling R, JAGS, and all dependencies.
  • Switched ggmcmc dependency from GitHub dev version to CRAN v1.5.1.2 (#135)
  • vectorized ab() function (#116)
  • Added lintr::undesirable_function_linter() to .lintr.R (#81)
  • Reformatted .lintr as R file (following https://github.com/r-lib/lintr/issues/2844#issuecomment-2776725389) (#81)
  • Set shortcut pipe to be base pipe (#80)
  • Added snapshot test for run_mod()
  • Clarified prep_data() internals using dplyr (#34)
  • Removed “.R” suffix from jags model files to prevent them from getting linted as R files (#34)
  • Added dobson.Rmd minimal vignette (#36)
  • Overall cleaning to get checks working (#28)
  • Added units tests for prep_data(), sim_case_data() (#18)
  • Added various GitHub Actions (#10, #15, #18)

serodynamics 0.0.0

Started development.