Convert captions between WebVTT and TTML/IMSC while identifying what is lossy

Steps

1. Note the structural mismatch: TTML supports nested <div>/<p>/<span> hierarchies with per-level language/styling, while WebVTT has no <div> equivalent, so TTML's block structure must be flattened into cues.
2. Map positioning: WebVTT's viewport-relative, horizontal-only line/position/size/align cue settings must be reconciled against TTML regions (origin/extent/writingMode), which can exceed or sit outside the viewport and support vertical writing — the W3C TTML-WebVTT mapping document flags this as the hardest part of conversion.
3. Map styling: build a lookup between TTML style attributes (tts:color, tts:fontFamily, tts:textAlign, etc., resolved via TTML's inheritance model) and WebVTT's supported CSS subset plus b/i/u span tags; unsupported attributes are dropped.
4. Convert timing: TTML supports multiple time bases (media, smpte, clock) with frame-rate/drop-mode parameters, while WebVTT supports only a single hh:mm:ss.fff expression, so SMPTE/frame timecodes must be resolved using the TTML file's frameRate (and dropMode if applicable) first.
5. For OTT delivery, target IMSC1 (a constrained TTML profile) rather than generic TTML, since consumer devices generally support only a narrow feature subset.