0003: Six findings became principles, the rest stayed findings

Decision: the card holds six principles: three about counting, whose grades are measured or convention, and three about reporting, all convention, each enforced by a program or by the standard's shape check. Findings the review graded without a rule a program can check stay in the review.

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

Context

  • The review has seven claims, each graded (review front page). The standard asks a principle to carry a grade and a check.
  • The measures program of the first case already checks a count by running; the repository's rule is that what a program can check is checked by a program.
  • The family's cards list principles with a Check line; a principle with no registered check fails principle-coverage.

Decision

  • Became principles, counting:
    • print the caution beside the count (measured, descriptive, for the perils; case-counts);
    • distinct work over raw commits (measured, descriptive, and measured in one case; case-distinct);
    • a hand count is marked (convention; case-hand-counts).
  • Became principles, reporting: a participant's report is an experience report (case-genre), the reader's checklist is answered part by part (shape-conformance) and no model analyses the data (case-analysis); all three are convention.
  • measures.py is the program behind the counting principles; the new checks read what it prints, and its unit test stays beside it.
  • Stayed findings:
    • studies of coding agents measure a task, not a day (measured per study; the gap is of this search);
    • company write-ups report sessions and cost (owner's claim);
    • medicine's case checklists (by analogy) and postmortem practice (convention);
    • no test shows that a checklist makes a case more trusted (untested).

Options that lost

Option Why it lost
One principle per claim of the review the claims about agent studies and company posts have no check a program can run, so they would name none
Only the two measured findings the other four rules are already enforced by a program or by the shape check; leaving them out leaves them without a grade or a reason
Copying the shape into shapes/examples/ the two cases are the examples and carry the Shape line; a copy would give a fact two places

Consequences

  • Good: every principle names a registered check, and CI runs the tests of the new ones.
  • Bad: four of six grades are convention, which is what the review found; the card does not claim more.
  • Bad: case-analysis and case-genre see a statement, not the practice. The maintainer confirms the statement.
  • To keep true: a new case follows the shape and carries the program, test, results and hand counts the checks look for; a later case may link an earlier case's program and genre.

See also

Decisions; the card; the review.