Pre-1.0 polish: Delphi caveats apart from the key, anova in 80 columns, the site menu, and Scheibe's 2002 edition

[JUhalt]

The four small items left for 1.0 after 0.9.0. The development version bump to 0.9.0.9000 rides along. No computed value changes. This is not a release, and 1.0 is not planned here.

The maintainer's guiding point for this work: contentvalidR is a teaching tool for instructors to use with students, and a tool for researchers in their own serious work. The output should teach without burying the result.

1. The Delphi printout, in two layers

The Delphi printout ran about 100 lines for five statements, and two method paragraphs printed every time.

• Always printed: two short caveats that change how a result should be read.

  • Read kappa as a trend beside the share of experts who kept their rating, not against a cut-off, because kappa falls as a panel converges (Holey et al., 2007).
  • With fewer than about 40 experts, the kappa intervals cover less than their stated 95% (Klar et al., 2002).

• The teaching layer, printed with the key under "How the stability statistic works":

  • what the weights do, and the intraclass-correlation equivalence (Fleiss & Cohen, 1973);
  • why no verbal labels are shown;
  • Klar et al.'s coverage figures.

  Students see it by default; a researcher can hide it with options(contentvalidR.show_key = FALSE) and keep every caveat.

• Unchanged: the critique printed for an alternative stability method (lambda, the chi-square tests, the 15% rule) still always prints, as the method-options policy requires.

The example goes from 100 lines to 92, or 58 with the key hidden. A new test checks that the caveats survive hiding the key, that the teaching layer doesn't, and that an alternative's critique is never hidden. The Delphi vignette's sentence about the key now describes what it holds.

2. anova_content() fits in 80 columns

It was about 96 columns wide and wrapped into two blocks. The competitor column is dropped: htd() and the rating workflow already show it, and it stays in strongest_competitor, which the note now names. The headings are shorter ("F test", "met"), and the note explains "met". The example is 79 columns, and a test checks the width.

3. The website

I checked the rebuilt site in a browser:

• the home page and its 11 guide links;
• the figure's alt text;
• a help page's formatted example output;
• the reference index, including its Deprecated section.

All pages resolve (HTTP 200). Two fixes:

• Articles menu: the menu listed the Start-here pages and hid the four workflow guides under "More articles…". Every guide now sits under a heading, in the order a course would use them: Start here, Workflow guides, Planning and reporting, After content review.
• Stale label: the auxiliary helpers' reference section no longer calls itself "not flagship v0.1.0 workflows".

4. Scheibe et al., cited from the edition we read

APA needs a page range for a book chapter. Citing sources give pp. 262–287 for the 1975 printing, but I couldn't check that against a copy. The package's percent-change rule was verified against the editors' 2002 web edition, so, at the maintainer's choice, the reference now names that edition:

Scheibe, M., Skutsch, M., & Schofer, J. (2002). Experiments in Delphi methodology. In H. A. Linstone & M. Turoff (Eds.), The Delphi method: Techniques and applications (pp. 257–281). https://www.foresight.pl/assets/downloads/publications/Turoff_Linstone.pdf (Original work published 1975)

In-text citations read "Scheibe et al. (1975/2002)", as APA 7 asks for a republished work. This covers the help page, the Delphi vignette, the README's References, inst/REFERENCES.bib, the printed method label, and the citation a handoff carries for stability = "percent_change". That is a prose field; schema v1 lets prose change, and nomologR's reader never matches on citation text.

Also: the partnership, stated

The ROADMAP's v1.0.0 section now states the maintainer's intent. contentvalidR and nomologR are partners, kept on the same page about handoffs before and after 1.0. solomonR is a separate project: it reads no handoffs and takes no part in the joint 1.0. The v0.7.0 schema item records the solomonR check as withdrawn (solomonR#32 was deleted on 2026-09-25); the released NEWS is left as history.

Verification

• All 2,950 tests pass.
• tools/release-gate.R on this PR's head (05c5c11), run in a detached worktree: build, smoke, and check all pass. R CMD check: 0 errors, 0 warnings, and only the two expected notes (New submission; math rendering skipped without V8).

🤖 Generated with Claude Code