Mobile shadowing: split a clip into sentence A–B loops

The ask

The owner asked whether the player could split a segment into smaller pieces — the core shadowing move is to loop one short line until it sticks, then advance. The player looped only the whole [startMs, endMs] clip, and clip.transcript is a flat string with no timing, so on the surface there was nothing to split by.

The data was already there

The catch: the player saw flat text, but the video stores timed structure. VideoResponse exposes transcriptSegments (1–3s cues) and sentences (merged, each a TranscriptSegment { startMs, endMs, text }), already served by videosApi.get. A clip is just a window of its parent video, so the sentences whose timing overlaps [clip.startMs, clip.endMs] are the sub-segments — no new backend, AI, or migration.

The fix (86b7fa0)

• Fetch the parent video, filter sentences to the clip window, clamp each to the clip bounds → a list of sub-segments.
• Render them as a horizontal chip strip in the pinned header: Full · 1 · 2 · 3 … (only when a clip holds 2+ sentences).
• Generalised the loop from a single clip [start, end] to an active window: tapping a chip sets that sentence as the loop, seeks there, and plays; "Full" restores the whole clip. The dictation / record "replay" buttons inherit the active window, so 🔁 replays just the sentence you're drilling.

Honest edge

The chips only appear when the parent video has sentence timing. Clips imported before that pipeline existed have no sentences, so they fall back to whole-clip looping — a per-clip data gap, not a feature break. Build + install verified; on-device loop accuracy not yet checked.

Pattern

"There's no data for this" is often "the data lives one level up." The clip had no timing; its video did. Before adding a pipeline, check whether a parent/related entity already carries what you need.