Postmortems, engineering deep-dives, and the small decisions behind a monitoring tool people can actually rely on.
A single trigger wired to the wrong table meant every new signup would have failed — and every digest email would have silently gone nowhere. Here's how I found it before opening the doors.
Paused monitors still cost something to keep around. Removing them automatically is easy; removing them without ever surprising someone is the actual design problem. Here's the 25 → 85 → 90 day path we landed on.
Your users don't file a ticket when your site goes down — they just leave, and you find out from a support email or a revenue graph, hours after the fact. Here's what actually catching it early looks like.
A monitor that pages you for a blip that fixed itself in ten seconds isn't protecting you — it's training you to ignore it. Here's where false alerts actually come from, and how to make the real ones stand out.
PageSpeed flagged 630ms of render-blocking CSS on the homepage. Next.js documents a one-line fix for exactly this. I shipped it, measured zero change, and traced through next/dist to find out why: the fix only exists for a router I'm not using.
Three reskinned layouts, a live preview while you customize, and a 24-hour timeline chart that was quietly dropping data — twice, until we found the actual cause.
A personal, owner-only page to stop bouncing between the Hetzner, Supabase, and Resend consoles by hand — and it immediately surfaced a stack of real database performance issues I didn't know I had.
Before renting a real server, I built a load-testing rig to find Impulseye's actual ceiling. It reported a ceiling that didn't exist — the test harness itself had two bugs, and both of them looked exactly like a capacity problem.
Discord, Slack, Teams, Telegram, Google Chat, PagerDuty, and Pushover each format an alert differently — and each one will silently reject you if you don't respect its limits. Here's what I wish I'd known on day one.
Nine channels, six alert types, three monitor kinds, reminders, flapping — that's a lot of ways for a message to come out malformed. So I built a tiny script whose entire job is to try all of them.
Three fix attempts, two of them wrong, before I found the actual culprit: a property that behaves completely differently on iOS Safari than everywhere else I tested it.
Any feature that lets a user tell your server to make an HTTP request to a URL of their choosing is a server-side request forgery surface — even if it looks like an innocent settings field. Here's how I hardened mine.