Scarlet>Fire

Scarlet>Fire is a free app for streaming the Grateful Dead's live archive: more than 17,000 recordings on the Internet Archive, there because the band spent thirty years letting fans tape and trade their shows.

I'm a Deadhead who wanted a better way to listen, so I designed, built, and shipped it alone, starting in early 2026 and landing on the App Store that April. No ads, no subscriptions, no account required to press play. What has shaped it since is my own listening and the feedback of friends who use it.

It's one of the great cultural datasets on the internet, and the Archive organizes it the way an archivist would: by date, source lineage, and taper attribution. Every existing app inherits that taxonomy. A list of years, then a list of dates, then a list of cryptic source identifiers, and you're expected to already know you want the Miller transfer of the Betty board from 5/8/77. That works for the person who taped the show. It's a dead end for someone who just wants to hear a great Scarlet Begonias>Fire on the Mountain.

The existing apps are database browsers, utilitarian and proud of it. I wanted something different in kind: the Spotify experience brought to the archive. Not Spotify's catalog, which will never have these recordings, but its standard, where curation greets you at the door and playback just works and every surface feels like it cost something. The bar wasn't the other Archive apps. It was the app people already spend their listening life inside.

Every session starts one of two ways: you know what you want to hear, or you don't. Being good at both meant having an opinion about the catalog instead of just exposing it.

If you don't know, the app points you somewhere great. A three-tier rating system (3 stars legendary, 2 excellent, 1 notable) is built from community consensus, HeadyVersions, archive popularity, and a close reading of the Taping Compendium, rather than my taste. Ratings render the same way everywhere, so quality is readable at a glance.

The homepage leads with answers instead of questions: legendary shows, hand-built collections like the great multi-night runs, a Show of the Day drawn from the 365 most popular, a radio mode that shuffles top-rated performances. Less deciding, more listening.

Filtering grew into a real tag system: years grouped into the band's actual chapters (newcomers don't know 1972 and 1977 are different bands), recording format, venue type, instrumentation like pedal steel and acoustic sets. Every tag carries a live count, so a dead-end filter is impossible. Search meets people where their memory is, matching dates in any format (5/8/77, "May 8," 1977-05-08), venue names, release titles, and resolving "california" to CA shows.

The Deadhead's Taping Compendium is three out-of-print, phone-book-sized volumes of fan scholarship: reviews of nearly every show the band played, written by the people trading the tapes. I rebuilt it into the app. Parsing the scans meant recovering text the OCR mangled, repairing two-column layouts, and stripping thirty years of page furniture. The result is prose notes for more than 1,500 shows, so a show page can lead with what actually happened that night: the guest sit-ins, the equipment disasters, the second sets people still argue about.

Then the notes became data. A scoring pipeline weighs what each note asserts, docking praise that's about the tape rather than the playing, and awarding a tier only where the evidence supports one. Community consensus always wins; the Compendium fills in the thousands of shows consensus has no opinion about, and where the two disagree the date gets flagged for a human. The same pass pulled 4,000 song highlights backed by 7,600 verbatim quotes, each verified against its source.

A random Tuesday in 1984 used to be an unlabeled list of tracks. Now it comes with liner notes and, when the book makes the case, a rating.

Deadheads chase versions, not shows. The question isn't "what happened on 2/13/70," it's "what's the best Loser ever played?" So there's a full A-Z song index with performance counts. Tap a song and every performance in the catalog is there, sortable by rating and filterable by the same tags as everything else, so "best Sugaree, Brent years, soundboard only" is three taps. It inverts the Archive's data model, which only knows about shows, into the one fans actually use. No other app answers the question at all.

That knowledge the Archive assumes, the Miller transfer of the Betty board of 5/8/77, is real, and it matters enormously to the people deepest in the archive. So the app organizes it instead of hiding it. A parser reads the Archive's free-text source notes and tags every recording by format (soundboard, audience, matrix, FM) and by lineage, the Betty Boards and Charlie Miller transfers among them. The picker lists every tape of every show, not a shortlist.

A playback preference decides which tape wins by default. Pin a different recording on any show and the app remembers the exception, synced across devices. Casual listeners never see any of this; press play and the most popular recording plays. The taxonomy that made the Archive hostile to newcomers is still here, opt-in, serving the people it was always really for.

Dead shows aren't albums. They're three-hour arcs with sets, segues, and 25-minute jams, and generic players mangle them.

Gapless had to be done properly: the whole show goes to the native player queue instead of loading track by track, so Scarlet Begonias flows into Fire on the Mountain the way it did in the amphitheater. When a show ends the next chronological one starts, and every show page lists the next three tour stops, so you can follow 1977 night by night the way fans followed the bus. An optional setting skips the tracks that are just tuning.

The Archive has no artwork, which is why every competitor is a wall of text. This one has looping video behind the main player and the frosted mini player, and share cards get seeded generative art, so a posted link looks like something.

Lock screen and Control Center scrub live, with next and previous routed back through the app so shuffle and radio logic stay correct. That routing is also what made CarPlay work for free.

Dead shows are long, and the places you most want them, planes, canyons, the drive to the lot, are exactly where streaming dies. So whole shows download to the device: tap the button and a progress ring fills as the tracks land. Downloads run in the background, survive force-quits, default to Wi-Fi only, and require no account, because nothing in this app does. They live in the Saved tab, and the player prefers the local copy automatically.

The constraint that shaped the feature is invisible. Soundboards are streaming-only by longstanding arrangement between the band and the Archive, so the app blocks downloading them rather than treating a technical gap as permission. The deal that keeps 17,000 recordings free is worth more than a feature.

Shows are how the Archive thinks; fans keep their own shelves. Collections save the shows that matter to you, and playlists go a level deeper: track lists built from across the catalog, so the best Scarlet Begonias from '77 can sit next to the best one from '89. Once the app can rank every version of a song, keeping your winners somewhere is the obvious next move.

Playlists are first-class in the player, not an afterthought. A playlist plays gapless like everything else and gets its own shuffle mode with a labeled badge in the mini player, and saves apply optimistically, flipping before the server confirms. Everything has a deep link, and the link is a real invitation: it unfurls with a branded share card in the message thread, and tapping it opens the app directly to the thing being shared, not a web page asking you to go find it.

Dead fandom was never a solo activity. Tape trading was social infrastructure before the internet had a name for it, yet every Archive app treats listening like a private errand. So you can follow other fans, browse their public profiles, and watch a feed of what they've been listening to, favoriting, and playlisting. Popular community collections are browsable too, which turns other people's taste into another way in.

Profiles are opt-in. Signing up is one tap with Apple or Google, usernames check availability live, and a public-profile toggle decides whether anyone can see you at all. Profiles have deep links too, so "check out this guy's '73 collection" works as a text message.

Streaming from a nonprofit's free infrastructure means I don't control the backend, so I attacked latency everywhere the client could. DNS and TLS warm at launch. Opening a show page fires a HEAD request that warms the CDN for the first track, so tap-to-play is effectively instant. Radio tracks and detail pages prefetch ahead of need, deferred until after first interaction so they never compete with render. When playback fails, the app retries a fallback stream before it ever shows an error. None of this shows up in a screenshot, but it's most of why the app feels nice.

The visible half works the same way. Free apps tell you they're free in a hundred small ways: default haptics, system spinners, a login wall before you hear a note. Haptics here follow a hierarchy, strong for play and pause, medium for skips and saves, light for pills. The dark system runs on near-black surfaces, a signature red, and Familjen Grotesk over Inter. Show pages open on full-bleed artwork that fades into the tracklist as you scroll, loading gets skeleton shimmer, and empty and error states are unified.

Guest mode has a prominent "skip for now," so nobody hits a login wall before hearing music, and signing up is one tap, framed as backing up your favorites rather than a demand. Share cards preview before you send, and the sender's card matches the recipient's unfurl because both come from the same seed.

Scarlet>Fire is an Expo app on Supabase, which is what lets one codebase ship to iOS and the web. I set the style in Figma, a few screens, and built the rest with Claude Code, starting in early 2026 and landing on the App Store in April.

The archive is a backend nobody controls, so a lot of the engineering was defensive: recordings with missing tracks, sources with inconsistent metadata, streams that drop. Testing was listening. I used the app the way I wanted to use it, every day, and the friends who picked it up sent back what broke and what they wished it did. That loop set the roadmap more than any plan did.