07 — Levels and variants
Summary
The v0.33.61 levels-and-variants brief operationalised. Levels vary depth for one reader; variants vary the rendering of one level; they are orthogonal, and a programme that conflates them cannot tell whether a poor result needs different content or different wording. The variant rule is mechanically enforceable here because the library gives facts identity: a fact-set diff across variants must be empty. Everybody starts at level one — the advanced user holds the largest grant and the strongest prior, the reactance combination — and the five scenarios are ordered by grant size, landing in this tool as library examples. The three sets organise the dashboard, with the exercised set's import gated on an open decision because its source is a session transcript. The stopping rule answers the depth question by dissolving it: each level must be a complete answer, a landing rather than a stair.
Key concepts
- The grid — five levels by three variants, generated rather than written
- The three sets — the third row — held and never used — is the product
- The stopping rule — shown only level three, the reader says what they would do — or the level failed
- Reserved words — altitude stays stakeholder; persona stays the generator
Key ideas
- Test the visual variants first: a visual eliminated a framing effect significant in text alone, and this tool ends in a picture.
- At a few dozen readers this is qualitative research — watch, ask for say-back, count nothing.
- Synthetic readers clear the levels; humans judge the variants; that order is also the cheap one.