Nishi Evidence

Every product we build, its measured state against the best tool in its class, the LADDER of work to state of the art, and dated runs of visible proof (screenshots, comparisons, recordings). One structure, no loose pages: a claim without a run here does not count.

Nishi Browser

Sovereign Windows-native browser (NishiLang first byte up) benchmarked against Chrome/Edge/Firefox

Live: /experiential — The census page the browser dogfoods (its own readability study target)

Latest: 2026-07-16 — THE TOP-20 SUITE (standing, one command): 13/20 render; REAL parity wikipedia 90.6 + hackernews 88.5 + stackoverflow 81.7; caught TWO bugs same-day (canvas bg-propagation; and the METRIC ITSELF — blank google scored 97.3% = emptiness-gaming -> ink-weight rung); gaps named as rungs

Sovereign 3D Renderer

NishiLang 3D raster organ (API+MCP) with browser mesh viewer

Live: /render3d — Live renderer gallery (page-verified); meshview3d browser viewer 25/25

Experiential Testing

The see/drive/judge stack (screenshots, recordings, visual diff, page verify, UX studies) vs Playwright-class SOTA — 19 censused axes, each wired organ+API+MCP+workstream+RACI as it lands

Live: /experiential — The live census (EMITTED by nx_s21_census — cannot drift from disk truth)

Latest: 2026-07-16 — KEYSTONE v1 LIVE: script drove the Nishi browser (click -> navigated the live ladder page, recorded); census GAP 14 -> 11, PARITY 8