Caption Foundry is for documentary editors delivering interview captions to Adobe Premiere Pro. The recurring job starts whenever a transcript changes: inspect rough SRT cues, split unreadable lines, repair timing collisions, preserve the words, and hand Premiere a clean WebVTT file.
The usual workaround is a text editor plus repeated imports. Editors split lines by eye, retime overlaps manually, reload the file, and discover the next parser or readability failure downstream. Caption Foundry turns that clerical loop into an explicit repair pass.
Load an SRT, set maximum characters per line, minimum cue duration, and minimum inter-cue gap, then apply the rules. Each control independently mutates the canonical interview fixture and reports quantified evidence. Source and repaired cues remain aligned on one synchronized transcript transport, so timing changes can be auditioned in context before export.
Project marker: caption-foundry-2026-08-06.
The visible repaired cue list and the exported WebVTT come from the same transform. The committed validator checks the WebVTT header and repair metadata, sequential cue identifiers, positive durations, the reviewed 250 ms gaps and 1.8 second minimum duration, no more than two lines per cue, and a 24-character line limit.
Live verification opens the deployed app, runs the canonical rough-interview workflow, writes fresh WebVTT bytes, and reruns that validator against the result. A screenshot cannot fake the handoff. The browser has to produce the actual caption file Premiere receives.