mirror of https://github.com/samvallad33/vestige.git synced 2026-07-26 23:51:02 +02:00

Sam Valladares cfe8d03d36 feat(landing): full-viewport GPU node-engine hero + waitlist + dashboard auth removal

The /launch landing page (bleeding-edge waitlist hero):
- Full-viewport WebGL node engine (src/lib/hero/nodeEngine.ts): 40k GPU particles
  on a two-FBO GPUComputationRenderer running an 18s looping cinematic — particles
  stream in from the screen edges, slam together and EXPLODE at center, reform into
  a brain / graph constellation / neural lattice, dissolve back out, loop. Glowing
  additive particles + UnrealBloom, fills the whole screen edge to edge.
- Key GPGPU fix: custom DataTexture shape targets read black in GPUComputationRenderer
  (three.js #15882), so shape targets are computed PROCEDURALLY in-shader from the
  per-particle seed. Position integrates raw per-frame velocity (Codrops pattern),
  texel-center reference attribute.
- AmbientField (god-ray glow + parallax starfield), phantomBrain (deterministic
  seed-from-identity memory graph), LandingHero/neuralFlow (WebGPU Cinema-storm
  reuse + WebGL fallback, kept as alternates). Manifesto copy, seed input, email
  capture. No em-dashes. SSR off + prerender for the WebGL route.

Waitlist backend (Supabase, owned data):
- supabase/migrations/0001_waitlist.sql (RLS-locked table, dedup), Edge Function
  waitlist-join (CORS, validation, honeypot, dedup, Resend confirmation), setup doc.

Dashboard auth removed (reverts the launch-polish auth wall that 401'd the dashboard
against itself): mod.rs/state.rs/websocket.rs back to no-token, api.ts/websocket.ts/
+layout.svelte stripped of the token client. Pending-review UX + Memory PR gating kept.

Research/spec docs under docs/launch/. Frontend typechecks 0/0.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

2026-06-24 22:54:51 -05:00

6.3 KiB

Raw Blame History

Vestige bleeding-edge landing page — verified 2025/2026 research

Synthesized from a 6-lens web scour (web-tech, award landings, viral waitlists, why-devs-share, interactive heroes, shareable artifacts) + adversarial fact-check. Every item below was verified as genuinely current for late 2025 / 2026 and shippable on a static SvelteKit + GitHub Pages site (Svelte 5, Kit 2, adapter-static, Three.js r172, Tailwind v4, Vite 6).

The throughline (what all 6 lenses agree on)

The 2025/2026 winning waitlist for a skeptical dev audience is: a live in-page product embed as the hero + a manifesto + a SHAREABLE ARTIFACT made of the visitor's own data + real (not fake) scarcity. Vestige is uniquely positioned: its artifact (a cinematic render of a memory graph) is more shareable than any generic "#1,204 in line" OG card — and it's already built.

Bleeding-edge web tech (shippable, static)

Three.js WebGPURenderer + TSL (r171+), now 100% browsers — Safari 26 shipped WebGPU Sep 2025. First time you can ship WebGPU-first to everyone (with WebGL2 fallback from the SAME TSL source). Headline claim: "150,000 GPU-compute particles, one shader, every browser."
TSL compute particles 50k–350k @ framerate — net-new in 2025/2026; ~150x over CPU. Memory Cinema's 150k sits dead-center in the sweet spot.
TSL post-processing (r183 RenderPipeline) — runtime-swappable bloom/dotScreen; glow the dream geometry, hot-swap effects per scene.
Paper Shaders (zero-dep canvas shaders) — cheap mesh/grain gradient backdrops below the fold so the heavy WebGPU canvas is only the hero.
Real-time ASCII / dithering / halftone post-FX — a defining late-2025/2026 aesthetic (Codrops "Efecto" Jan 2026). Run the Agent Black Box through a dithered/CRT pass so an agent's "thoughts" read off a screen.
CSS scroll-driven animations (animation-timeline: scroll()/view()) — JS-free scrollytelling, Interop 2026 target. Drive all non-canvas reveals off this so the GPU budget stays for the hero.
View Transitions API (same-document Baseline Oct 2025) — morph hero → graph → dream → Black Box on scroll, app-like, off-main-thread.
Motion (motion.dev), MIT, WAAPI off-thread — 2.5–6x faster than GSAP, no Webflow license issue. Use for all DOM/UI motion so it never janks the canvas.

What makes award-winning dev landings 1-of-1 (2025/2026)

Product UI rendered INSIDE the WebGPU pass (Igloo Inc, @pmndrs/uikit) — labels/wordmark as SDF text in the same pass as particles, sharing grain/aberration/depth. DOM-over-canvas looks "pasted on"; this doesn't.
Scroll = camera = narrative (worldbuilding, not parallax) with offscreen render-pass culling for perf.
Cursor-reactive POST-processing warp (whole composited frame bends toward cursor; scroll velocity smears particles) — generic "spotlight follows cursor" is 2022.
Live spec-sheet hero / big real numbers, NOT logo soup (Evil Martians 100-page study) — the 2026 dev crowd distrusts "trusted by" walls. Lead with live HUD: "561 stars · 43 countries · 150,000 particles."
Kinetic/variable-font typography driven by scroll velocity (wordmark "tightens" as you dive into the graph).
Switchable one-canvas, three-lens hero — Vestige has exactly 3 surfaces: Dream (Cinema) / Graph (force-directed) / Black Box (agent replay). One canvas, three toggles.

Viral waitlist mechanics (2024–2026, verified)

The share artifact = the product's OUTPUT, not an OG counter card — Suno song / v0 gen / Lovable URL. For Vestige: a 6-second rendered loop of the VISITOR'S OWN seeded memory graph flying through Cinema's dream worlds.
Manus tradeable invite codes — "scarcity that creates content"; access itself is the artifact. Give each signup 2–3 unique founder codes that unlock a rendered Cinema world.
Robinhood skip-the-line position jump — now table-stakes SKELETON; layer the novel artifact on top. Tie milestones to Vestige-native unlocks (new dream geometry / longer render), NOT "free month."
REAL scarcity beats fake — Superhuman's own builder now warns against fake scarcity; deceptive.design lists it as a dark pattern. Frame the cap truthfully: "each Cinema render is real WebGPU work; seats open in batches as render capacity scales" (literally true).
Police leaderboard fraud from minute one — #1 self-inflicted failure with a taste-driven HN/X crowd. Require GitHub OAuth or email verification before a referral counts; dedupe by verified identity.
Manifesto-first, capture-second (Arc Browser pattern) — "Your agent forgets everything. Your memory should be yours, local, and beautiful." then scroll INTO the live demo.
Compounding milestone unlocks — each referral milestone spawns a NEW shareable (new dream world / longer render: "I just unlocked the gyroid world"), not one-shot share-to-join.

Personalized shareable artifact — how to build it (all client-side, fits local-first)

Raycast Wrapped 2025 pattern: fully client-side "Copy as Image" from LOCAL data — exactly Vestige's local-first positioning; a server OG pipeline would CONTRADICT it.
WebGPU/WebGL canvas → media via captureStream() + MediaRecorder (works for Three.js AND WebGPU) — MemoryCinema's sandbox already renders to a canvas host with camera-only fallback. Wire capture for a shareable clip.
canvas.toBlob → PNG for the still card; HTML-in-Canvas drawElement() (Chromium 147+) to composite a stat overlay onto the live canvas where supported.
Seed-from-identity deterministic art — on the waitlist (no real memory yet) seed a one-of-one "phantom brain" from the visitor's GitHub handle / email hash / typed memory. Unique before they do anything.
Build-time dynamic OG via Satori + resvg-js (or @ethercorps/sveltekit-og) so the shared LINK preview is a stunning Cinema frame, even though the page is static.

The two pieces that CANNOT be static (the only hosted bits)

Referral state + position counter — needs a tiny API/edge fn (GetWaitlist/LaunchList, or our own Supabase fn — already started).
Per-user OG/share-card if done server-side — but the Raycast pattern lets us do it CLIENT-side instead, staying local-first.

WebGPU caveat: default-on in Safari Sep 2025 / Firefox Jul 2025 — always ship the WebGL2/Canvas/pre-baked WebP fallback so share never silently breaks.

6.3 KiB Raw Blame History Unescape Escape