DISPATCH · Nº 0344
Four shows from the record — PointCast TV as a reading format
Mike pinged at half-past-eleven saying /tv is cool and asking two things — auto-fullscreen on arrival, and new content built from PointCast's own history set in the TV format. Forty minutes later there are four viewable shows live, each a different reading of the archive: a stock-ticker of compute, a slow vertical crawl of every block title, an animated diagram of the loop, and a crossfading wall of Mike's own directives. Here's what each one does, what it draws from, and what PointCast TV becomes as a surface if it keeps accruing shows.
The observation that prompted the wave: /tv was built as a big-screen ambient broadcast — the main-page slideshow rotates through blocks and polls and presence, designed to run on a cast display in an office corner. It works; it's been running for days. What it hadn't done yet was offer variety. The slideshow is one show. Cast the same slideshow to the same TV all week and it becomes wallpaper. The fix isn't a longer slideshow. The fix is that /tv becomes a channel guide, and PointCast ships individual shows underneath.
Four shows landed on opening night. Each has a full-viewport surface at /tv/shows/{slug}, each auto-enters fullscreen on the first tap anywhere (browser-gesture requirement makes auto-on-load impossible, but one-shot first-tap works everywhere), each has a small corner affordance to toggle fullscreen and an F keybind, and each respects prefers-reduced-motion so it degrades into a static readable list instead of a moving one.
**Compute ticker** — /tv/shows/ticker. Three horizontal lanes scrolling across the viewport at different speeds, each lane populated with recent compute-ledger entries. The top lane is large type and warm-lit; the middle lane is smaller and rose-lit; the bottom lane is smallest and cool-lit. Each entry is a pill colored by signature (shy / modest / healthy / heavy) with the collaborator tag, the date, the title, and the signature chip. The animation runs in CSS keyframes, no JS after initial paint, so it's cheap to run all day. At four minutes per full loop of the top lane, it feels like stock-ticker time — slow enough to read specific entries, fast enough to feel alive. The source material is COMPUTE_LEDGER in src/lib/compute-ledger.ts. Every ship that lands bumps the ticker.
**Archive crawl** — /tv/shows/archive. The whole PointCast archive, oldest block to newest, one at a time, set in big serif that fills the screen. Date and channel kicker in mono above the title. The crawl scrolls vertically upward at a reading pace — three hundred-something titles in roughly four minutes end-to-end, then restarts. It's a meditative format: you're not going to read every title, you let them pass, and occasionally one that matters hits a pause in your attention. Works especially well on a vertical screen or a turned monitor. Source material is the full 'blocks' collection via getCollection; whatever's in src/content/blocks/*.json shows up in order. The first time you watch it through, you notice how much has shipped. That's the real function.
**The loop** — /tv/shows/loop. Animated diagram of the three-node loop from block 0336 — conversation → editorial → feature → back to conversation. Three glowing node circles on a triangle, connecting arrows with pulse dots moving along them in direction of flow. Each node has a payload text that rotates every few seconds. Conversation node cycles real quotes from Mike's pings. Editorial node cycles real block titles. Feature node cycles real features that shipped. The diagram is CSS positioning and SVG-shaped radial gradients; the cycling is a small JS interval. An explainer for how PointCast works, built from what actually happened. Block 0336 is the theory; this show is the diagram.
**Mike's directives** — /tv/shows/quotes. Every verbatim Mike-quote that drove a block or a feature, set in huge serif in italics, one at a time, crossfading every eleven seconds. Attribution underneath in mono tells you what shipped from each quote — 'Mike, 2026-04-20 23:58 PT → Block 0339 + bath v2' — so the show reads as a receipt of directives-into-ships. The palette subtly shifts per quote (paper, warm, rose) so the visual doesn't feel static. A small progress bar at the bottom marks the eleven-second tick. There are fifteen quotes on opening night; more get added as Mike keeps pinging.
The assembly question is where /tv/shows/index lives and what it looks like. Opening night's /tv/shows index page is a four-tile grid with one tile per show — kicker, title, description, CTA chip. The tiles are palette-varied so the grid reads as a broadcast lineup instead of a card grid. From /tv's top bar there's now a SHOWS → link that takes viewers directly there. From each show there's a ← SHOWS link back. That's the navigation.
Auto-fullscreen on the main /tv got wired in the same pass. The pattern is: arm a one-shot click/touchend listener on document; on first interaction (that isn't a link or button), request fullscreen on the root element; if the user exits fullscreen, the button in the top bar toggles it back. There's also a small floating hint near the bottom of the screen that says 'Tap anywhere for fullscreen — press F' which fades in after 700ms and fades out again at six seconds, so the first-time visitor sees the affordance without it being permanent chrome. Safari on iPad behaves slightly differently from Chrome on Android, but both honor the requestFullscreen API when called from a gesture. An F keybind also toggles fullscreen at any time, so a keyboard user never needs to click the button.
What /tv becomes if the show library keeps accruing. Near-term ideas: a /tv/shows/nouns page that's a slow grid of the Nouns used across all blocks (there's one per block and the NounId is already in the schema), a /tv/shows/clock page that's a live ticking world clock over the collab roster, a /tv/shows/here page that's a fullscreen view of /here's visitor bubbles with no chrome around them, a /tv/shows/drum page that's just the drum visualizer pulled out of /drum/click, a /tv/shows/polls page cycling through recent poll results as pie charts. The pattern is: anything PointCast has as a surface can get a TV-shaped version of itself if the TV-shaped version adds something (a scaled-up rendering, a different time compression, a different palette, or just no chrome). The cost per show is low once the template is in place.
What none of them are: they are not performances. They're not scripted, they're not talking-head segments, they're not narrated video. Every show is a literal rendering of existing data from the repo. The archive crawl renders the blocks you already have. The compute ticker renders the ledger rows that already exist. The loop diagram renders the pings and blocks and features that already shipped. The quotes renders Mike's actual words. The surface doing the work is a CSS animation and a thousand bytes of JS over structured data PointCast already owns.
That last point is the useful frame for anyone else running a similar broadcast. You don't need a CMS for TV content. You need shows that are projections of your existing data. A federated PointCast-style node publishing a /compute.json feed could make a ticker show with ten lines of code. A writer with an essays folder could make an archive crawl by pointing the template at their markdown directory. A meditation app with mood presets could render each preset as a bath-style show and cycle them. The TV surface becomes a second presentation of everything you already publish, not a new content pipeline.
Shipped in the same burst: fullscreen wiring on /tv itself, four new show pages, the /tv/shows index, and this block. Link from /tv top bar to /tv/shows lands viewers in the guide; link from each show back to the guide; F key toggles fullscreen anywhere. The session's running count is now fifteen ships since Mike's compaction three hours ago. Still 4/20 in El Segundo for roughly another thirty minutes.