Package review checklist 2.1.0: scope, AI disclosure, test quality, executable examples, Hub data packages; fix stale link and duplicates

[lwaldron]

While mapping every bullet of the review checklist onto an agent-assisted review helper we're building (Sean Davis, Marcel Ramos and I; more on that when it's ready to show), I ended up comparing v2.0.0 line by line against the current guide. Most of it maps cleanly. This PR fixes the places where it doesn't, plus a few things the guide asks for that the checklist never mentions. Nothing here changes policy; it brings the checklist in line with chapters that already exist, and with what reviewers already ask for in practice.

Added

• General: an explicit scope bullet (in scope for Bioconductor, interoperable, not a near-duplicate, not on CRAN). It's the decision that matters most in a review and the only one with no checkbox.
• New section AI-generated and copied code: disclosure in the issue and attribution in-code, per the chapter 19 policy. Kept separate from the license question, which stays under Third-party code.
• Unit tests: split into presence, quality (tests that assert known values and edge cases rather than just "runs without error"), and no network / no broad skip_on_bioc(). Tautological tests are the failure mode we're seeing most in agent-written submissions.
• Man pages: examples must actually execute on the builder. BiocCheck already errors below 80% and treats \dontrun and \donttest as no example; @examplesIf interactive() has the same effect and nothing catches it. Also one line saying internal functions need @noRd, not more documentation.
• Package data: a Hub-based data packages subsection (metadata.csv, make-data.R, make-metadata.R, resources resolvable before review), and the license-of-downloaded-resource bullet made concrete.
• Vignette: no unconditional downloads in vignettes or tests, so the build system stays within its time limit.
• R code: constructor, validity, show() per exported class, and results that print something useful.

Changed

• The web-resources link pointed at the retired bioconductor.org/developers/how-to/web-query/; now the appendix on contributions.bioconductor.org, with tools::R_user_dir() named alongside BiocFileCache.
• Version: says 0.99.z for new submissions.
• Sweave note mentions Quarto and its SystemRequirements entry.
• Removed the two "No warnings or errors in R CMD check / BiocCheck" bullets under R code; they duplicate General.
• The coding-and-syntax sub-bullets that BiocCheck checks mechanically are tagged "(checked by BiocCheck)", so a reviewer's 30 to 90 minutes go to the judgement items.

Not in this PR, for discussion: a two-tier layout or severity tag per bullet, so fnd, ORCID and README are not at the same level as license and Remotes. That's a bigger reorganisation and I'd rather hear whether people want it first.

Version bumped to 2.1.0.

Drafted with Claude Code (Claude Fable 5.1); decisions and edits are mine.

[Copilot]

Copilot review overview

🟡 Changes recommended

Unresolved checklist alignment and documentation issues remain.

Get a fresh assessment by requesting another Copilot review.

Review effort: Lite
Findings: 2 · 2

Open (4)

• Consolidate R_user_dir and BiocFileCache requirements · New
• Align disclosure and provenance requirements with policy · New
• Require exact initial version 0.99.0 · New
• Limit BiocCheck claim to mechanical system-call detection · New

What changed in this PR

Updates the Bioconductor package review checklist to version 2.1.0 and align it with current guidance.

Changes:

• Adds scope, AI disclosure, test quality, executable examples, Hub data, and class criteria.
• Clarifies versioning, caching, Quarto, vignettes, and BiocCheck guidance.
• Updates links and removes selected duplicate checks.

File	Description
docs/​package-review-checklist.md	Expands and revises package review criteria.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.