On the Publish tab, Audio stems exports per-track submixes alongside the master, which is what a post house asks for first.
Options: WAV or FLAC, 48 or 96 kHz, 16 or 24-bit, and whether to bake in your mix gains. The default is pre-fader at unity gain, which is the usual delivery — it lets the recipient rebalance freely.
Ducking is deliberately not applied to stems. A dialogue duck printed into the music stem cannot be removed later, which defeats the entire point of delivering stems.
Without FFmpeg the export produces the exact per-stem plan and the FFmpeg commands that would render it, and reports rendered: false. It does not write silent placeholder WAVs — a bundle of empty files would be discovered at the worst possible moment, in someone else's session.