Per-part clock divide + timing offset (polyrhythm / phasing)

[philoking]

Summary

Give each part its own clock divide and timing offset so the four lines don't all march on the same 16th grid. Today every part advances together at STEPS_PER_BAR = 16 (js/generator.js), driven globally by js/scheduler.js. Let the Pad run at 1/1, the bass at 1/4, the arp at 1/8-triplet, each optionally shifted by a 16th — and the lines drift into evolving, non-repeating polyrhythm.

Why it fits

Phasing and cross-rhythm are the most modular-native textures there are — exactly what you want under a long-evolving patch. Because the parts are already strictly monophonic and independently generated, staggering their rates is cheap musically and gives a lot of motion for very little UI.

Proposed scope

Per-part rate + offset (js/generator.js, js/scheduler.js)

• Add per-part clockDiv (e.g. 1/1, 1/2, 1/4, 1/8, 1/16, plus triplet 1/4T, 1/8T) and offset (0, 1/16, 1/8, 3/16) to each entry in state.parts.
• The scheduler already advances time incrementally; have each part decide whether this global step is one of its trigger steps given its divide, and apply its offset when stamping the note time.
• Keep mono enforcement per part (drop coincident starts, clip before the next note / bar line) exactly as it works today.

The phasing trick (length ≠ length)

• Allow a part's note-source length and its rhythmic length to differ, so the note pool rotates against the pulse and never lines up the same way twice. (Dovetails with the step/rhythm editor idea, but works on the generated streams too.)

UI (js/app.js)

• Two selects in each part's ⚙ panel: Clock divide and Offset. Show the resulting feel (e.g. a "notes : beats" readout) so the polyrhythm is legible.

Where

js/generator.js (per-part step gating + offset), js/scheduler.js (time stamping), js/app.js (⚙ controls + state).

Done when

• Each part can be set to a different clock divide and a 16th offset.
• Triplet divides work against the straight grid.
• Settings persist in presets / JSON.
• Mono behaviour and clean gate retrigger are preserved per part.

Open questions

• Should swing apply per divide, or stay global for now?
• Cap on combined divides to avoid pathological density at high tempo?