UI-Debt Map
The design quality of every Nishi surface, measured — not by taste but by liar-killed byte-counts: design tokens, theme-awareness, responsive breakpoints, type scale, and polish. The score is the evidence.
| Surface | Score | Bar | Grade | tok | theme | resp | type | polish |
|---|---|---|---|---|---|---|---|---|
| /uidebt · this page | 12/12 | POLISHED | 3 | 2 | 3 | 2 | 2 | |
| /compare/browser | 11/12 | POLISHED | 3 | 2 | 2 | 2 | 2 | |
| /roi | 6/12 | FUNCTIONAL | 3 | 0 | 2 | 1 | 0 | |
| /governance | 1/12 | DEV-GRADE | 0 | 0 | 0 | 1 | 0 | |
| / · front door | 1/12 | DEV-GRADE | 0 | 0 | 0 | 1 | 0 | |
| /pm | 0/12 | DEV-GRADE | 0 | 0 | 0 | 0 | 0 | |
| /standup | 0/12 | DEV-GRADE | 0 | 0 | 0 | 0 | 0 | |
| /sheriff | 0/12 | DEV-GRADE | 0 | 0 | 0 | 0 | 0 |
The finding
Five of eight surfaces — including the front door and the governance board — are DEV-GRADE: no design tokens, no theme, no responsive breakpoints, no polish. Two are polished: the hand-crafted compare page, and this page, which is emitted through nx_ui_kit + the sovereign design system nishi_ds. The backend is state-of-the-art; the surfaces a person touches are not — now measured, not asserted.
How this is scored
nx_ui_audit <url> fetches the live page and counts real design signals in its source — var(--) token uses, prefers-color-scheme, @media breakpoints, clamp() fluid type, and transition / focus polish. Every sub-score is a measured byte-count, so it cannot be faked. The fix is not per-page hand-styling — it is routing each surface's emitter through the shared design system, so the whole fleet inherits quality. Dogfood law: this page scores POLISHED on its own judge, or it does not ship.