Diataxis is a framework, not a result: no controlled study compares readers on separated and mixed pages.
Small repositories may be served better by one README holding several forms.
Against checklists as quality
The JOSS criteria test presence, not usefulness: a page can satisfy all six and still fail a reader.
Prana 2019 describes what READMEs contain, not which contents help.
Against strict comment rules
Comment quality is a cluster of attributes (Rani 2023), not one measurable quantity; weights between them are a judgment.
The register differs by language (imperative in PEP 257, descriptive in the Google guide), so no single rule holds.
Generated documentation sites (rule 8 of Lee 2018) trade zero redundancy for tooling; rejecting them is defensible, and so is adopting them.
Ousterhout's test relies on judgment about what the code "cannot" say, which two reviewers can read differently.
Practice against the principles
One mode per document: Diataxis critics argue the cost of strict separation.
Viktorin 2022: one participant asks for guidance on pages that mix forms, Batchelder asks for guidance on applying the four perspectives, and Wichmann warns developers may just end up ignoring a heavy method.
Hacker News 2024: critics call strict use rigid and overkill for small projects, and say some explanation inside a guide is helpful.
Some commenters report poor links between split pages and duplicated content that goes stale; one calls repetition not a big problem.
A README mixes modes by design: GitHub, Make a README and standard-readme each place several modes in one file. Read as an index, it does not break the principle.
Define before use: Google leaves well-known terms undefined. The exemption turns on the audience, so it is a reader-style choice, not a contradiction.
Diataxis tutorials: minimal explanation can put a term before its definition unless the step links to it.
Adoption is self-reported: the Diataxis site and Canonical 2021, written by its author, give no metrics, so adoption does not show the separation helps readers.
Length: "too long is better than too short" (Make a README) opposes "shorter is always better" (Microsoft); neither is measured.