The StudioMode Blog

Demand queue + push notifications + RSS feed compose into a closed-loop system. Before this round the catalog only bent toward demand reactively; after it, both sides have ambient reasons to keep showing up. Eight steps, three different actors, zero direct coordination.

Chef tested with a brand-new account, searched "ebk jaaybo", got zero results. The next two scrapes ignored it. Tracing why turned into a P0 fix and four feature rounds — admin widget, empty-state rewrite, public /hunting-now page, dynamic OG card. Bug to flywheel in five rounds.

Per-step velocity shipped on the drum sandbox in Round 143. Three rounds later it shipped on the piano roll's chord progression as a Dynamics row. Same primitive, same multiplier, same hash trick, two different surfaces — and the cross-tool transfer cost less than the original.

The drum sandbox started as a 16-step grid. Twelve iterations later it has chord overlays, walking bass, kit swap, swing, save, random, shift, mute, and copy-share-link — and the URL hash from Round 106 still loads in the Round 141 sandbox without a single migration.

The BPM detector says 70. The vocalist hears 140. Neither is wrong; they're locked to different pulses in the same audio. Once you see the pattern, it shows up everywhere — half-time vs straight, triad vs 7th, swung vs straight. Four toggles in the audio toolkit are all the same idea.

The piano roll started as a scale detector. Five rounds later it identifies chords by name, draws the keyboard shape, reads inversions as slash chords, lays out the full diatonic palette for any key, and toggles between triads and sevenths. One engine, five UI surfaces.

The drums tool and the piano roll both got Save buttons last week. The implementation took ~30 lines per tool because save/load is a strict subset of share/load when the URL hash is the canonical state. One serializer, two surfaces.

The /tools/drums sandbox grew from a 16-step grid into a full beat sketchpad: chord progression overlay, three bass modes, four drum kits — all driven by one Web Audio scheduler ticking every 25ms. Why parameter swaps were the cheapest path to four sound palettes.

Four rounds after the scale detector shipped, the piano roll now handles the full chord workflow: identify chords from notes (Jaccard + bass-priority tie-break), step through 8 progressions, auto-play through the Web Audio scheduler, and look up chords by name. Why bass-priority was the magic tie-breaker.

Piano-roll scale detector, BPM + key analyzer, guitar tuner. All built in 4 build rounds. None upload anything. The algorithms (FFT, autocorrelation, YIN, Krumhansl-Schmuckler) and why no-server was the cheapest correct choice.

21 build rounds. Four leaderboards. Three history archives. Eight share cards. One cron. The architecture pattern that made it cheap, and the trap I keep seeing teams fall into.

Vercel Hobby gives you 12 functions and 1 cron. We added 4 leaderboards anyway — by piggybacking idempotent snapshot RPCs on the existing daily-digest cron.

Six pages on StudioMode were eligible to be indexed under five different URLs each. Trailing slash, www, query params — every variant split the ranking signal. Six lines of HTML later, the bleeding stopped.

Most products bolt rankings on after they have critical mass. We did the opposite — and the leaderboard ended up doing the work of the community we hadn't built yet.

A short retrospective at the 10-post mark. Which essays drove signups, which got picked up, which were boring and necessary anyway — and the three traits that travel.

Same client-side aggregation showed up in three different places. The migration to fix it was 60 lines of SQL and saved ~600KB of transfer per page load. The pattern, the trade-offs.

Followers are vanity. Saves are intent. Building a producer leaderboard around the wrong metric ranks the wrong people — and what changes when you fix it.

Vercel Hobby caps you at 12 serverless functions. We render 17 different dynamic Open Graph cards. Here's the dispatcher, cold-start budget, and why we skipped @vercel/og.

A retrospective on the 30 features shipped after StudioMode's v1. What hit (4 that compounded), what flopped (3 we'd skip), what surprised us (3 sleeper bets), and the meta-lesson about building in public.

The Vercel Hobby plan caps you at 12 serverless functions. Here's how StudioMode fits a notification cron, a referral redirect, an OG image generator, an SSR page, and seven other endpoints inside that limit — and the tricks we used when the cap got tight.

Five thousand StudioMode saves later, the patterns are clear: artists don't pick beats the way producers think they do. Marketplace UX has been optimizing for the wrong behavior. Here's what the data actually shows.

StudioMode's catalog isn't licensed. We don't host audio. Every beat surfaces from a YouTube scrape that runs twice a day. This is the story of getting from "10 beats indexed" to "3,000 beats deduped and tagged" — including the night the catalog ceiling became a real ceiling.

The For-You feed isn't magic. It's a stack of small, deliberately tuned signals — followed-producer boost, recency decay, key-match bonus, a guard against producer spam, and an exploration slot for the algorithmic equivalent of "look up." Here's everything that goes into picking the next beat.

BeatStars works for producers. It's broken for artists. We can fix the discovery layer without disrupting the marketplace — and that's exactly the wedge StudioMode is built around.