A music box with no motor. Drag the wheel. Slow down, stop between notes, or rock one peg back and forth. Nothing turns unless you do.
Switch tines with a second finger while turning. Lift a tine to leave space.
Try just C3: rock across its top peg, pause, then answer with a full turn. Switch other tines back in when you want company.
Keyboard: focus the wheel, hold ← / →. Hold Shift for a careful creep. Press 1–4 to switch tines without letting go of the arrows. The nudge buttons turn 1/32 of a revolution. Sound is optional; matching tines flash on contact.
Retiming bench · move the peg rings
Shift one voice's pegs around the wheel to change where it answers the others. Each step is 1/16 turn; the tines stay put. Editing is silent and affects the next turn, not notes already ringing. The sound and nudge controls stay within reach while you edit; try a few steps forward, then back.
Hear two ring relationships, forwards and backwards · optional online recordings, with steps to try them yourself.
Try one clockwise step on G3, then turn again. Same eleven pegs, different neighbours. Original positions are one reset away. Changes last only until reload. Experimental: musical benefit is not yet established.
Workshop note: the first note after a pause
A pause can soften the first returning note: the browser may report only two positions, with no indication of when your hand began moving between them. A controlled browser test found a 13.1 dB difference in the attack envelope between a short movement sample and the same displacement across a pause.
Simply shortening that interval would also make an indistinguishable, deliberately slow gesture louder. That repair was rejected; timing and contact behavior are unchanged. This is a measured sampling limitation, not a listening verdict. Musical feel and real-device touch remain untested.
Workshop note: a tiny musical traffic jam
Several contacts in one pointer event are played immediately, so attacks can bunch. A browser comparison reached the same C3–A4–G3 contacts with spaced and batched input. An experimental 12/24 ms stagger separated the batched attacks without delaying single contacts—but left notes scheduled after release. Cancelling them discarded two of the three contacts in the tested quick-release trace.
I have not adopted that trade: the wheel keeps immediate contact timing, not an invented little arpeggio. That timing pass is finished; the retiming bench above is a new experiment, not a claim of musical polish. These were scheduling checks, not listening tests; musical benefit and physical-device feel remain untested.