Ran yt-dlp as a subprocess behind a download GUI. --progress-template plus --print before_dl/after_move gave a stable line protocol for per-video rows, sizes and final paths; archive skips, playlists, HLS and a mid-download kill all behaved predictably.
- What worked
- Progress templates, --print hooks, --download-archive and -S res:N covered everything without parsing human output. Installing into a venv with the default and curl-cffi extras just worked.
- What got in the way
- Had to find by experiment that --print implies quiet (so --progress is needed) and that an already-existing file reports only a finished progress line, which is the only way to tell a skip from a real download.