Against systems - Kennedy's case vs our token machine

Digested from Erik D. Kennedy, Why Beginning Designers Don't Need to Learn Grids, Type Scales, or Color Theory. This is the one article that argues against the shape of our system - captured honestly.

His argument

Grids - built for fixed paper, broken on elastic screens (a 12-column grid on a 320px phone yields ~26px columns, below touch targets; widening the screen shouldn't widen every element proportionally). Instead: alignment. Content dictates its own width; on mobile, three rulers suffice (16px left, 16px right, center).
Type scales / modular scales / golden ratio - mathematically satisfying, practically hollow: a ratio that works on desktop yields absurd headlines on mobile, and "if you try this on a design that looks crappy, it will… still look crappy." Instead: corral attention - make differences that should read as different clearly different (10 vs 12pt is invisible; 50 vs 52pt more so), and stay consistent within a project.
Color theory - wheel-derived palettes don't build interfaces. Instead: one base color varied by saturation/brightness (see color.md).
Personal style - don't chase it; make good things and style follows.

The diagnosis underneath: pick-a-number fatigue. Design is a thousand number choices, and systems sell relief from choosing. His verdict: you can outsource the number-picking, or you can learn the rationales. Outsourcing gives brittle rule-following; rationales give judgment.

Delta vs our system

CONFIRMS - Alignment over grids - our baseline-alignment rule ("every element answers: what am I aligned to?") is precisely his replacement for grids. We never adopted a column grid; we adopted alignment. He endorses our choice. - Color: variation over palette theory - full agreement (see color.md). - Consistency within a project - his fallback position is exactly what our tokens enforce.

ADDS - The rationale test. For each token in our system we should be able to say why - not "because 1.25" but "because a step must be visibly a step." The system doc mostly can; where it can't, the token is dogma by his definition. - Content dictates width - a useful check against forcing every block into the same container when the content wants otherwise.

TENSION - the honest one - Kennedy rejects modular type scales; our system IS a modular scale (1.25 × 5 steps). His critique lands on humans using a scale as a substitute for judgment, learned once and applied blindly across breakpoints. Our situation differs in kind: the "designer" here is a generator producing many pages across many sessions, and the scale is a consistency contract enforced by machine (token-audit.py), with the judgment layer kept separate (the 10-point Autocontrol, the squint/5-second test). We outsource number-picking deliberately so that judgment is spent on hierarchy and meaning, not on re-deriving 20px. - But his warning has teeth in two places: (1) responsive behavior - a ratio tuned for desktop reading may need a flatter scale on mobile; our system doc is silent on breakpoints; (2) audit-passing ≠ good - a page can trace every value to a token and still have no hierarchy; the diagnosis that birthed our system said exactly this ("twenty font sizes and no hierarchy" was cured by the scale, but the scale alone doesn't create the hierarchy). - Resolution to propose to Frank: keep the scale as law for consistency, adopt Kennedy's "corral attention" as the criterion the scale serves - differences must read as differences (our hierarchy-gaps-≥1-step rule already encodes this; name it as the rationale, not just the rule).