Build log — Music Trafficker disco ball

Date: 2026-07-09

Update 4 (same day): discussion popup + members

Installed Wix Blog + Wix Members Area apps (apps-installer API), seeded one published post "What do you think of AI music?" (id 7ed6fe11-3420-4585-9a20-56b61cf03626).
New /api/thread route: GET lists comments (public, newest first, author names via @wix/members), POST creates a comment (member-gated, 401 for anonymous). Comments key on the post id — this fresh Blog V3 site has no referenceId (that field only exists on legacy-migrated posts); verified by creating/listing/deleting a probe comment via REST.
UI: "What do you think of AI music?" button under the ball → translucent popup (composer when logged in, login CTA otherwise); log in/out chip top-right using the built-in /api/auth/* routes. Disco ball behavior untouched.
Verified: popup + empty thread + anon-post rejection live (desktop & 390×844); login link reaches the Wix-hosted login/signup dialog. Not verified end-to-end: posting as a logged-in member — Wix signup requires solving a reCAPTCHA, which an automated agent can't (and shouldn't) do. The comment write path itself was verified at the API level; if member posting misbehaves, check Blog comment moderation settings in the dashboard first.

User feedback: fast storyboard playback looked sped-up (frames are ~2 s apart in the source), slow playback looked laggy. Fix is two-tier:

Real motion: 18/30 videos have YouTube an_webp hover previews (~3 s of consecutive frames @ ~10 fps). Proxied via /api/motion-preview, decoded client-side with ImageDecoder (Chromium; falls back below elsewhere) into 160×90 atlases, played at natural speed.
Crossfade: remaining videos (and non-Chromium browsers) show storyboard frames that hold ~1.3–2.2 s and dissolve into each other via a small shader — no hard cuts. Also fixed a stale-CDN issue: an early release let /api/videos be cached for 1 h; the client now uses a versioned URL (?rev=2).

Animated previews: every window now plays its video as a muted 160×90 fast-forward loop, built from YouTube storyboard sprite sheets (one 5×5 sheet per video, animated by UV offsets — no decoding, works in all browsers). Watch pages turned out to be bot-walled from the Wix datacenter egress (consent/captcha stub, verified with a temporary debug endpoint), so the signed sheet URLs are resolved at build time from this machine into src/server/storyboard-snapshot.json (scripts/resolve-storyboards.mjs) with live resolution kept as a fallback for post-deploy videos. Sheets are proxied via /api/storyboard-image (upstream has no CORS headers). Verified live: 30/30 videos animated.
Auto-update: feed cache dropped to 15 min; the client polls /api/videos and hot-swaps added/removed videos into the ball without remounting the scene (swap deferred while a video is playing/flying). New videos appear at most ~20 min after publish — but animate only after the next snapshot refresh + deploy (static thumbnail until then). If YouTube storyboard signatures ever expire, affected tiles silently fall back to static thumbnails.
Fix (user report): mirror facets could poke through video windows — the radius jitter reached the window plane height. Jitter now clamps every facet below the window backings.

Wix Headless site + Astro project scaffolded with npm create @wix/new@latest (business name "Music Trafficker", blank template). Site id 4df484c7-bb1c-4d2f-990f-0ddf48f7b1ea. The scaffold's own git-commit step failed (no global git identity on this machine) — expected; fixed with a repo-local git config.
Wix agent skills installed into .agents/skills/ via npx skills add wix/skills.
3D deps: three@0.162, @react-three/fiber@8.18, @react-three/drei@9.122 (React 18 pins), @react-three/postprocessing@2.16.3 + postprocessing@6.35.2 (pinned — newer versions require three ≥ 0.168).
YouTube feed: src/pages/api/videos.ts server route; parses the channel's /videos tab (ytInitialData, full videos only, durations included), RSS + /shorts/ probe as fallback, 1 h in-memory cache. Channel id resolved once: UC5D9N79IcFdrPGoymBThZZA. 30 full videos found; Shorts (e.g. the "Simulated Universe" series) excluded by source.
Scene: src/components/disco/ — instanced mirror ball with ~82 thumbnail windows, orbiting spotlights, lightformer env-map, sparkles, bloom/vignette, time-based tile→fullscreen flight animation, YouTube iframe overlay, Escape/✕ to close, drag-to-orbit.
Deployed with wix build + wix release → https://music-traf-ab6ea66c-vytenisu.wix-site-host.com

Window planes invisible: the dark backing boxes' front faces sat 0.01 world units in front of the window planes (occluding them), which masked a second, opposite orientation bug during diagnosis (three.js Object3D.lookAt for non-cameras points +Z at the target, so tangent planes need a rotateY(π) flip). Fixed both.
GL_INVALID_VALUE: glTexSubImage2D warnings when the flyer upgraded to the hi-res thumbnail: swapping a larger image into an already-uploaded texture resizes immutable GPU storage. Fixed by giving the flyer a fresh THREE.Texture per flight/upgrade and disposing the old one.
Redundant allowFullScreen attribute warning on the iframe — removed (the allow list already grants fullscreen).

Ball spins with thumbnails visible, ~60 fps feel, desktop (1280×800) and mobile (390×844) — whole ball fits on both (camera distance adapts to aspect).
Click on a tile → tile flies smoothly to a letterboxed fullscreen rect → YouTube iframe fades in → video plays from 0:00 with sound (verified via progressing video frames and the player's [music] caption; autoplay-with-sound worked in the test browser).
✕ / Escape closes: iframe removed (audio stops), tile flies back, ball resumes spinning.
Console: no errors from site code. Two external notes below.

frog.wix.com beacon (Wix's own BI telemetry, plain-HTTP) is refused on this office network → one ERR_CONNECTION_REFUSED console error that is not from site code and is environment-specific.
releasePointerCapture errors seen during verification only — an artifact of the synthetic pointer events used to drive the canvas from DevTools; real mouse/touch input doesn't produce them.
Autoplay with sound on real mobile devices may still be blocked by stricter browser policies (the user then just taps the play button in the player). Desktop Chrome played with sound.
YouTube page-markup parsing could break if YouTube changes ytInitialData/lockupViewModel; the RSS fallback keeps the site functional (latest 15 videos) if that happens.