trustangle × NHJ · design language · version comparison

Three versions of the same page.

The register as the council ratified it, the facet system that replaced it without authority, and the route-2 correction. Same content in all three, so the only variable is the design language.

scroll inside any pane · open ↗ for full screen
Ratifiedv1A-on-B, 2026-07-16 open ↗
Unratifiedv2facet system, 07-19 open ↗
Route 2v3grid built, cut demoted open ↗

What actually differs

Measured on the page, not asserted. "B" was defined by three components: a visible hairline grid, mono section labels, and the logo's angle as the only geometric operator.

Measurev1 ratifiedv2 facetv3 route 2
Visible hairline grid — B component 1003
Structured mono section labels — B component 2003
Cut applications on the page — B component 31251
Cut shape classes in the stylesheet342
.ta-varied auto-variation enginepresentremoved
RTL mirror coverage15/15 (was 2/15)complete
Colour tokenised — binding req. #4yesyesyes

A correction worth reading before you judge these. The council's finding was that the facet system removed B's hairline grid. Checking the files says otherwise: the grid scores zero in v1 too. It was never built — not in the facet era, and not in the version the council ratified 6–0.

So v2 did not delete the grid. It filled a vacuum where the grid should have been. That makes v3 an implementation rather than a restoration, and it means the ratified register was never fully built in the first place — which is a larger finding than the one that prompted this comparison.

What v3 changes, and why

DecisionReasoning
Grid drawn as a background gradientCosts no DOM, never interferes with content flow, and survives RTL without a mirrored copy. It switches off under 820px, where a 12-column rule stops being legible rather than becoming noise.
Section labels get an index and a ruleB's second named component. An index plus a name on a hairline reads as a system; a floating uppercase word reads as decoration.
One cut, one size, used once per pageThe governing rule inverts: the grid is constant, the cut is rare. Regularity is the argument the register is making. If two adjacent elements both carry a cut, one is wrong.
Auto-variation removed entirely.ta-varied was the mechanism of "nothing is uniform" — the rule the council found unratified. Removing it is the substance of route 2, not a side effect.
Column spans use logical propertiesArabic sets the grid natively rather than being mirrored afterwards — binding requirement #1, the one two members changed position to adopt.
Tokens, type scale and light grounds keptThe genuinely good parts of the facet work. None of them were what the council objected to.