RipplesEffect

Radial sine ripples on a 2D RGB panel (with optional repeated depth slices). Computes one frame per loop20ms tick using a precomputed phase table + base-color table + 256-entry cosine LUT — the per-pixel inner loop is one Q16 subtract, one LUT load, three uint8 mul-shifts. No sqrt / cos / HSV per pixel; see Sprint 7's perf tune. Publishes its frame via PixelRegistry so consumers (PreviewModule, ArtnetOutModule) can read it by source id.

End-user reference

Control	Type	Range / default	Notes
enabled	toggle	true
width	slider	1..128 / 16	Geometry; change reallocates pixel + table buffers
height	slider	1..128 / 16	Geometry; change reallocates
depth	slider	1..16 / 1	Slice count (2D pattern repeated per slice)
speed	slider	0.1..10 / 1.0	Radians/sec the wave advances
hue_base	slider	0..1 / 0.6	Base hue; rotates +0.05/distance for visible bands. Change recolours without reallocating.

Developer reference

• onAllocateMemory() — call allocate_() which psram_allocs three buffers: working pixel buffer (w·h·d), phase-offset Q16 table (w·h), base-color full-bright RGB table (w·h). Allocates one DataBuffer<RGB> slot (one pixel-buffer-sized chunk); declares it in DataRegistry under id().
• onUpdate(key) — width / height / depth re-call allocate_(); hue_base re-calls rebuild_color_table_() (skips the phase table — only color changes).
• loop20ms() — compute one frame into the working pixel buffer using the precomputed tables + cos LUT; then buf_->acquire_write() + memcpy + buf_->publish() so consumers (ArtnetOutModule, PreviewModule) can read it cross-core.
• teardown() — DataRegistry::instance().undeclare(buf_), free pixel + table buffers.

Pixel-buffer sharing — why a registry, not dynamic_cast

The natural shape for a consumer to grab a producer's frame would be dynamic_cast<RipplesEffect*>(manager_->find(id)). That does not work on hardware: arduino-esp32 builds with -fno-rtti. Adding -frtti is a hammer (binary size, framework-wide); putting a type-specific accessor on MoonModule would put light-domain knowledge in core. The minimal answer is DataRegistry — a publish-on-onAllocateMemory / resolve-by-id registry in src/core/ that is domain-neutral (see ADR 0003).

This is publish/subscribe — just the cheap version. Declare happens once in onAllocateMemory(); resolve (DataRegistry::resolve) happens in consumer setup() and on onUpdate("source"), with a fallback in loop20ms to tolerate late producers. The hot path is a cached pointer + one atomic load + one uint32 revision compare. No event bus, no per-tick dispatch, no allocation. See architecture — hot-path data sharing and backend — data sharing.