Evidence and grades

Every component and style carries a claim about why it might help, and that claim is graded, not asserted. switchback show <id> prints a component’s grounding: the claim itself, its grade, its sources, and the conditions under which it helps or backfires.

The grades

Grade	Meaning
A	Robust — meta-analytic or replicated experimental evidence, directly on point
B	Supported — good evidence, though with some heterogeneity or a smaller effect than the headline number suggests
C	Suggestive — mixed, small, or indirect evidence
D	Practice — an established technique people use, with no direct study behind this exact use, or evidence that has not yet been read in full

A component’s displayed grade is the weakest grade among the claims it actually depends on, not the strongest one it could quote. Only one item in the whole catalogue currently reaches A: the commit page, which rests on implementation intentions — one of the most replicated findings in this literature.

Where D is not a warning sign

A D grade means one of a few different things, and they are not the same:

• Craft, not failure. Timeboxing, the proofreading-mark convention, and a handful of other moves are established professional practice with no controlled study behind this specific use. They are kept and labelled, not removed, because “untested” is not the same as “shown not to work.”
• Theory only. A mechanism that is well understood in principle (why placing items on a two-axis grid helps comparison, say) but has not been isolated in a study of its own.
• Unread. A small number of claims are graded D only because their source could not be read in full during grading, not because the source disagrees. Component pages and switchback show print these as a plain D, like any other; the “(unread)” note that marks them lives in the CLI package’s research/grading-report.md. They can move if someone reads the source properly — they are not evidence the page fails.

Nothing graded D is hidden or dressed up as stronger than it is. If a claim’s grade moved down during an adversarial review pass, the page’s own text was reworded to match, rather than left overstating what is now known.

How the grading was done

Each claim traces to a specific construct in the research literature (for example, “epistemic action” behind moving pieces during a card sort), with its sources cited by DOI, ISBN or URL, verified against the source itself rather than taken on trust from a summary. Claims went through an adversarial review pass afterward, specifically looking for reasons a grade should be lower than first assessed — several were downgraded as a result, and the reasoning for each is on record. The full accounting, including every claim examined, is in the CLI package’s research/grading-report.md, alongside the sourcing files for each research area.