Codex review round 2 on #1471:
- Encoding detection now scores the Windows-1251 candidate's decoded text
instead of any byte ratio: genuinely Cyrillic files decode into
pure-Cyrillic words, while misread Latin text produces words mixing
Cyrillic and ASCII letters ("était" -> "йtait") — so accent-dense CP1252
dialogue ("Été") stays Latin and short CP1251 dialogue stays Cyrillic.
- The packaged Linux frame-copy smoke now drives addEmbeddedMpvSubtitle
with a fixture .srt against the real helper (asserting the track appears
in the session snapshot) and round-trips the subtitle delay/style IPC;
the native file dialog itself cannot be automated and is documented as
manually verified.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>