A voice requirement records in the browser and ships PCM to the relay
as one binary frame (voice.js, same mount/stop contract as gesture);
its value becomes {key, digest, duration_ms}. Requirement.value
presets a field and, on a dynamic page, takes the URL segment - so
value: "{key}" tells the voice field where to record. A url resource
source takes the segment too. Resource items with an https `audio`
field render an <audio> player. The gesture widget reports picks to
the relay for ranking.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
