Where side projects become Success Stories

Reworked Lumen's command palette: fuzzy search, recent commands pinned to top, inline argument hints. Two reviewers say it's fast, two say it's cluttered. Tie-breaker, please. What's in it Fuzzy match across commands + recent files Right-aligned keybinding hints Inline preview of what the command will do Is the inline preview helpful or noise? And does showing recents-by-default beat a blank slate? Genuinely torn - screenshots attached.

Coregraph turns a repo into an interactive dependency graph. It's lovely at 3k nodes and falls over a cliff around 8k. SVG is clearly not the answer past a point. Where I'm at SVG: dies ~3k nodes (too many DOM elements) Canvas 2D: smooth to ~15k but hit-testing is painful WebGL: scary but probably the real answer The two hard parts are (1) a force-directed layout that doesn't take 30s, and (2) picking/hover at 50k nodes. // naive hit test = O(n) per mousemove = death nodes.find(n => distance(n, cursor) < n.radius) Planning WebGL + instanced rendering for draw, and a quadtree for hit-testing. Before I sink two weeks into it - is that the right architecture, or is there a library that already nails this scale?

Building the auth module for Pocketship and stuck on a product line: do I bake in a full opinionated auth (sessions, refresh, biometrics, social) or ship thin adapters and let people bring Clerk/Supabase/Firebase? Bake-it-in = magical first 10 minutes, painful when someone needs something custom. Adapters = flexible but the "weekend MVP" promise dies in config. Starter kits live or die on the first hour. But they get abandoned on the first wall. Where would you draw the line?

With Pocketship I keep getting the same support question: how do I share a design system across React Native and a marketing website so they don't drift apart after the first redesign? I've seen three camps: Design tokens only - share colors/spacing/type as JSON, rebuild components per platform Shared component package - one RN/web component lib, lots of platform ifs Just copy-paste and accept the drift The dream is: designer changes a token in Figma, it flows to native + web with zero manual work. For teams who've lived with this for more than a year - what actually survived contact with reality? Style Dictionary? Tamagui? Something custom?

Building realtime multiplayer for Driftboard (infinite canvas, shapes, sticky notes). I keep going back and forth between CRDTs and operational transform and I'd love real production experience, not blog-post theory. My constraints Offline-first - people sketch on planes and sync later Boards can get big (10k+ elements) Small team, can't babysit a custom OT server forever CRDTs (Yjs) give me offline + conflict-free merges almost for free, but the memory footprint on long-lived boards scares me. OT is leaner on the wire but the server logic is genuinely hard to get right. Specific questions: For those running Yjs in prod - how do you keep the doc from growing unbounded? GC, snapshots, something else? Anyone regret choosing OT and wish they'd gone CRDT (or vice versa)? Not looking for "it depends" - tell me what bit you. 🙏

Tempted to use RSC for the document/comments shell around Driftboard's canvas. The canvas itself is obviously client-side, but everything around it (sidebars, comments, presence) could be server-driven. My worry is latency: every interaction that touches server state pays a round trip, and an editor lives or dies on feeling instant. Do you split "shell = RSC, hot path = client" and keep them strictly separate? How do you avoid waterfalls when the client island needs server data? Is RSC genuinely worth it here, or am I about to spend a month fighting the framework for a worse result than plain client + a good cache?

Honest product question for Sentinel (error + perf monitoring). Every competitor does per-seat or per-event pricing, and my early users keep telling me both feel hostile for small teams. The tension Per-event punishes you exactly when you have an incident (more errors = bigger bill during your worst day). Per-seat discourages inviting teammates, which kills the collaboration that makes the tool sticky. Options I'm weighing: Flat tiers by retention window, unlimited seats Usage-based but with a hard cap you set (never get surprise-billed) Generous free self-host + paid cloud convenience I'd rather make less money than have founders dread opening the dashboard. For people who've shipped a paid dev tool: what pricing model did you regret, and what converted best without feeling extractive?

Lumen is getting a plugin system. I want plugins to extend the terminal (commands, panels, themes) without being able to read arbitrary files or exfiltrate env vars. Considering: WASM sandbox with a capability-based API surface A separate process per plugin + IPC QuickJS/V8 isolates with no host bindings by default // plugins only get what we hand them const api = { registerCommand, readClipboard }; // no fs, no net runPlugin(code, api); For anyone who's shipped an extension ecosystem (editors, browsers): what's the realistic threat model, and where did you regret being too permissive early?

People hate waiting for a big analytical query in Querly. I want to stream something useful before the full result lands. Ideas Stream rows as they arrive (cursor) and progressively fill the table Show an approximate count / sample first, then refine Run a fast LIMIT 100 preview while the full query runs in the background const cursor = db.queryStream(sql); for await (const row of cursor) { table.append(row); // paint as we go } What feels best to users (not just engineers)? Progressive fill, or a fast preview that might change under them? Worried the "numbers changing" effect erodes trust more than a spinner does.

Our RN e2e suite is ~12% flaky and it's training the team to ignore red builds, which defeats the entire point. Suspected causes Animations not settled before assertions Network races (real backend in CI) Timers and debounced inputs I'm tempted to quarantine flaky tests automatically and retry once, but that feels like sweeping it under the rug. How do you keep e2e honest without it becoming a coin flip?