0008: Which findings became principles

Decision: five findings of the review became principles, each graded as the review graded it, and the others stayed findings.

Status: accepted, 2026-11-03. Shape: decision record.

Context

  • The standard asks an extension for a card of principles, each with a grade and a check.
  • The review graded few of its findings above convention. Its position names four conditions under which a topic can be the loop's interface, and the repository already enforces three rules by program: topic, lean-link and comparator.

Decision

  • Principles: the four conditions, and the rule lean-link enforces, which makes five:
  • One shape and one check: the claim and claim-shape, because the three claims already shared a form no program read.
  • Stayed findings: every other finding, because none earned supported with caveat for a rule about writing:
    • the benchmark accuracies and error rates describe models, not a rule for a writer;
    • the effects on thinking and on learning have no randomised study;
    • the error messages for people have mixed results;
    • the costs of the audit were not measured.
  • Each principle names an existing check, since principle-coverage fails a principle with none. Where a program decides only the form, the principle says that the meaning is judgment.

Consequences

  • Good: each rule has a grade, a source and a program that runs in CI.
  • Bad: guidance at the claim's grain and the person's splitting are checked by form only; whether they help the person is untested.
  • Bad: a finding that stays a finding is easy to forget; the review lists them.
  • To keep it true: a principle's grade moves only with a cited source, and a new finding that earns one becomes a principle by a new record.

See also

Decisions; the card; 0007, whose two audit columns the second principle holds.