# Renderer provenance

This renderer follows the measured source-span layout and aligned overhead
annotations from Fluxtapose's `reading-chapter-ui`, including its later highlight
and underline work. The recovered source was the Nix-cached package at
`/nix/store/2ka0jcdpyijmpjsrwqkap682l4xk88zi-source`, rather than the earlier
prose-probe prototype. Fluxtapose's `reading-app-2` references package revision
`4e479c1a74c13d7467071da60b98fc148128dd99`; exact cache revision is unconfirmed.

Recovered file SHA-256:

| File               | SHA-256                                                          |
| ------------------ | ---------------------------------------------------------------- |
| text-layout.ts     | e6c330f60b7f1c16235454d1ce9c7656ef2379e744322ac13826753831e13d72 |
| text-rendering.ts  | ce2ea4f15f4d3b86d4b8f72d6ac7221cb131c996d19ec24052d80e3235012f77 |
| text-processing.ts | 7d8a72aa2d8791ad37351d1c7c872419b288b5e0e51214983a29ff48c8465038 |

The original shared source/annotation flow is preserved: every annotation source
has one row per physical text line. Long words and translations request horizontal
padding around their source ranges and continue across subsequent lines. A forward
capacity constraint replaces the ten-pass spacing loop; collisions never create
additional annotation rows. Conflicting ranges from the same source share an
ordered horizontal run. Row heights use all glyphs/ranges, and mixed font sizes
share a baseline. Text keeps original
UTF-16 offsets, including leading spaces and grapheme clusters. Measurement is
injected for tests and uses a bounded cache in browsers. Canvas drawing and
selection have no Konva or memoizee dependency.

Legal wrap positions use the pinned `@cto.af/linebreak` 4.0.3 implementation of
Unicode 17 UAX #14, intersected with Intl grapheme boundaries. Intl word
segmentation remains responsible for word targeting and preferred word affinity.
Source text and overhead labels use the same line-breaking layer. Closing marks
stay with preceding text, opening brackets stay with following text, and
annotation padding cannot insert a break inside those punctuation spans. Oversized
words and numbers use emergency tailoring of the alphabetic/numeric rules while
retaining punctuation rules; viewports narrower than an inseparable span still
require clipping. The package's Yarn lock and generated Nix dependency closure
retain the library and its transitive dependencies for reproducible browser builds.