You give the sequencer a list of soundbites. Each soundbite is just a time range on a sequence in your Premiere project — like 14.5s → 23.7s on "Interview Sam". No filenames, no source timecodes. The JSON literally doesn't know what media is under those times.

When you load that JSON into the Premiere panel and pick Original Media mode, the panel doesn't try to find the source media. It asks Premiere directly: "hey, walk every track of the Interview sequence and hand me back every clip object." Premiere already knows everything about every clip — its source file, its source timecode, where it sits on the timeline, what audio's linked to it. It hands all that back.

For each soundbite range, the panel finds the clips on every track that overlap that range, then does the math:

• A clip sits at 12.0s → 30.0s on the master timeline.
• Its source media starts at, say, 02:14:08:00 (in the original file's own timecode).
• Your soundbite starts at 14.5s on the master — that's 2.5s into the clip.
• So the slice from the source media is 02:14:10:12 → 02:14:19:18.

The panel sets those in/out points on the source clip and inserts it into a fresh sequence — on the same video and audio tracks it lived on originally. B-roll on V2 stays on V2. Dual-system audio on A2 stays on A2. You get a real, editable, full-resolution sequence made of the actual source clips, not a nest.

Why this matters: the JSON has no media references, so it can't leak any source paths or filenames. The translation only resolves inside your Premiere project, where the original media lives. Send the JSON to someone without your project — they get nothing. That's intentional.