# Native review gates

Native TextLayoutResult records are emitted by the shipped Compose surfaces at 1920×1080, density 2.0. Weights below are actual layout requests, not a claim of identical Roboto/DejaVu glyph ink. Original-pixel stroke and ink measurements are in native-measurements.md.

| Gate | Native observation |
|---|---|
| Version heading | 22 sp; requested weight 800; bounds [1184, 80, 1336, 138] px |
| Focused source label | 15 sp; requested weight 700; bounds [1203.8, 235.97, 1528.16, 275.75] px |
| Stacked media facts | 38 px pitch; declared CSS pitch 37.2 px; key weight 700; line height 15.6 sp |
| Poster-caption separator | ARGB ['4cf4f4f6']; plain fact bdf4f4f6; alpha ratios [0.4021] (rounding of muted × 0.4) |
| Nav initial | ARGB ff0b0b10; shared dark onFocus default |

Header initials remain muted in the shared library header (LibraryScreen.kt). Title cast/person fallback initials follow the frozen .pinit rule: white at 82%, not the library-header rule. Title/person pages have no library-header avatar; the nav remains dark-initialled.

Line-height fixes use cached explicit TitleTextStyles only for title sheets, person copy and title recovery. Shared nav/library typography and platform Roboto are unchanged. Full launch paths remain explicitly deferred until the Home merge; exact title-entry suffixes are the approved current gate.

CSS bleed audit: .crow img produces enlarged rating chips and .crow .tx span changes readiness lines only inside Coming in request rows. None of these 34 title/person/options frames has a .crow ancestor. Title/filmography marks retain the frozen .rr one-em / 1.8-em rules; the Searching title state remains inline as in st-068. No global CSS cleanup or unobserved bleed was invented.

Borders: title controls keep canonical four-pixel rings outside a four-pixel gap; physical thresholded ring results retain JPEG differences. Saved/error borders are two pixels with true border-box padding. Season selection uses an inset two-pixel shadow, so it does not add border width to its label padding.

The real async focus regression delays the SDK facade reply through TitleDestination and WolfskinTitleViewModel, exercises inert loading Down, then actual Android Version/Tracks key navigation and hardware sheet Back. No direct state-ready shortcut substitutes for that transition.