Skip to content

SPECTACLE.md

Latest commit

 

History

History
70 lines (56 loc) · 4.03 KB

SPECTACLE.md

File metadata and controls

70 lines (56 loc) · 4.03 KB

Mycelia — the visual-spectacle proof point

SUPERSEDED 2026-06-10 by MYCELIA.md. This was the multi-species-physarum version; Carter (a mycologist) redirected to a real adaptive mycelial ecosystem (transport solver + foraging + zone lines + life cycle). The physarum work here (S1–S3) was the plumbing rehearsal and stands as a simpler sibling. See MYCELIA.md for the live plan.

Chosen 2026-06-10 (Carter: "Massive visual spectacle") as the web-supremacy proof point, after the rust-gpu compiler-contribution path was retired. Max-effort build.

The thesis (honest framing)

Not "JS can't render this" — WebGPU underlies both. The real, defensible claim:

A living, million-agent multi-species simulation whose entire kernel is ordinary, unit-tested Rust — running on your GPU via rust-gpu, and verifiable bit-identical on the CPU. Gorgeous, interactive, shareable.

The differentiator is Rust's correctness + testability + one-codebase-CPU-and-GPU meeting WebGPU's power. The spectacle makes that visceral. Lead with the experience; the Rust story is the substance underneath.

What it is

Multi-species Physarum: several agent colonies sharing a high-res trail field. Each senses its OWN channel (attraction) and is REPELLED by the others, carving competing territories with luminous boundaries. Millions of agents, 60 fps, palette + bloom rendering, mouse interaction, presets. The boundaries between colonies are where the beauty lives.

Plan

  • S1. Kernel (shared/src/mycelia.rs, 6 CPU tests green). Multi-species (N_SPECIES=3, planar trail channels), Jones agent rule generalized to signed (repulsive) signals, mouse force in-kernel, deterministic spawn/update/diffuse. The tested-Rust foundation = the proof.
  • S2. GPU entries (shaders/src/lib.rs): mycelia_spawn_cs (init agent buffer), mycelia_update_cs (one thread/agent: update_agent → non-atomic deposit into own channel of the trail-write buffer), mycelia_diffuse_cs (one thread/cell/channel: diffuse_at, ping-pong), mycelia_render_cs OR a fragment pass: trail channels → RGB → IQ palette + bloom + tonemap (reuse gpu-shader-lib::color). Verify a single-agent + diffuse step GPU-vs-CPU (bit-exact diffuse; agents statistical) like the existing sim.
  • S3. Web page (web/mycelia.html): full-bleed canvas at devicePixelRatio; scale agents to whatever holds 60 fps on the 5070 Ti (target millions); per-frame passes update → diffuse(×k) → render; mouse → params (attract/repel); a tasteful control set (species count, sensor angle/dist, speed, decay, self/other weights, palette) and morphing PRESETS; FPS + agent-count readout; screenshot/share. Gorgeous first, knobs second. wasm-CPU fallback at small N for the same-code story.
  • S4. Polish + verify + deploy: headless-Chrome ?auto beacons (frame renders, reactivity, agent count); tune the look until it's genuinely mesmerizing (Carter's eyes are the gate); deploy to docs/, cross-link from index. README section with the honest thesis + a hero capture.
  • S5. [RED] Package + announce: decide own-repo extraction (Carter picks a name — "mycelia" is the working name) + Pages site; any outward announce needs an Approvals line.

Reuse

  • Kernel pattern + GPU-verify + headless harness: the existing Physarum sim (shared/src/physarum.rs, web/sim.html, shaders/ entries physarum_*).
  • Palettes/tonemap/color: gpu-shader-lib (git dep, already a cross-repo consumer via oscilla).
  • Rendering reference: the gallery (web/gallery.html) fullscreen-fragment pattern.

Honest risks

  • "Another physarum demo" — beat it with scale (millions), multi-species boundaries, and rendering polish. The look has to be exceptional, not just functional.
  • Non-atomic deposits race at scale (documented; visually fine). If banding appears, switch the deposit to u32 atomic fixed-point.
  • Carter's gate is his eyes — budget for look iteration, not just "it runs."