ADR-0003 Four named themes on one axis
#Context
The system must serve interfaces in controlled interiors and in the field — direct sun and dark-adapted night. The obvious model is a 2×2: polarity (light/dark) × environment (interior/field). The obvious model is wrong.
#Decision
Four resolved, named themes on a single axis: interior-light, interior-dark, field-day, field-night. Every semantic token resolves in all four, from Phase 1, with no partial themes and no deferred environments. Attention-carrying components are designed in field-night first: shape, position, weight, and pattern are the primary meaning encodings, hue is reinforcement. A component that cannot express its meaning without hue is marked unavailable in field-night, and reaching for that exemption is a design smell interrogated before it is granted.
#Rejected options
The 2×2 (polarity × environment). Cleaner conceptually and it generates the same four cells. Rejected because two of the four cells it implies are incoherent: sunlight legibility requires maximum contrast, so a dark field-day theme cannot exist, and dark adaptation requires suppressed emission, so a light field-night theme would be actively harmful. A matrix with forbidden cells is a list wearing a matrix costume; the honest model is the list.
Light and dark now, field themes later. The standard scoping move, and it ships Phase 1 faster. Rejected because retrofitting environments is the most reliable way to rot a token architecture — every token authored without the hostile constraint accumulates assumptions the field themes then have to fight. Building for the hostile cases first produces a better system for the easy ones.
Hue-fallback as degradation. Draft 1's framing: components encode in hue and declare a non-hue fallback for field-night. Rejected and inverted: designing the constrained case as an afterthought produces afterthought quality exactly where legibility matters most. Field-first makes the strong encoding the primary design and turns hue into a bonus.
#Consequences
Phase 1 cost roughly doubles versus a light/dark-only start; accepted deliberately. The theme-completeness CI check (token diff across all four themes) exists from Phase 1. ADR-006 (field-night palette strategy) and ADR-007 (density's relationship to the theme axis) inherit this frame.