From an empty newsletter to the reference engineers reach for mid-incident.
The Sev-1 Database started in October 2025 as a newsletter with zero subscribers and one stubborn rule: never publish a fix we hadn't reproduced on real PostgreSQL. Eight months later, that rule has a following.
(0 → today since Oct 2025)
along for the journey
indexed & searchable
& verified (and counting)
Most database content is written from memory and a confident tone. We got tired of pasting a Stack Overflow answer into a production incident at 3 a.m. and hoping it was still true. So we built the opposite: a reference where every fix is run on a real cluster, captured, dated, and explained before it ships.
How it started
October 2025: one post, zero subscribers, and a Postgres incident that took four hours longer than it should have because the "fix" everyone quoted didn't reproduce. The newsletter began as a way to keep receipts — to write down what actually happened when you ran the command, on which version, with the literal output attached.
People showed up because that honesty is rare. The list grew. The format hardened: reproduce it, verify it, explain the why from the source, then publish. Nothing untested goes out under this name.
The journey, in milestones
Zero subscribers, one rule Day one
The first issue ships. The rule is set on day one: if it isn't reproduced on real PostgreSQL, it doesn't get published. Subscriber count: 0.
The Error Reference takes shape
Readers start sending the errors that ruined their week. We begin indexing them — SQLSTATE, message, cause, and a reproduced fix — and the searchable database is born.
Cookbook, Parameters & Glossary land
Production runbooks, a GUC reference with real defaults across versions 14–18, and a plain-English glossary. The newsletter becomes a place you keep open in a second tab.
Interview Prep & Learning Pathways
The same reproduce-or-die standard, pointed at the questions that decide senior offers — EXPLAIN plans, MVCC, vacuum internals, walked through from the source.
1,050+ subscribers · 2,500+ on LinkedIn Now
Over a thousand engineers read us in their inbox; more than 2,500 follow the journey on LinkedIn. Incident Pro now funds the deep reproductions — the fix stays free, the depth pays for the lab.
Why this exists
Because the fix should always be free. When you're on the hook at 3 a.m., you shouldn't hit a paywall to find out which command stops the bleeding. The error pages, the glossary, and the core of every parameter reference are free, forever. What you pay for is depth — the reproduced danger cases, the tuning formulas, the interview-grade internals — and that revenue is what keeps the lab running so the free part stays honest.
If we didn't reproduce it, we don't publish it.
Every fix on this site was run on a real PostgreSQL cluster — across versions 14 through
18 — with the literal output captured and the date recorded. No paraphrased docs, no
"should work," no AI-confident guesses. When we haven't verified something yet, we say so out loud.
Haider Z.
I write The Sev-1 Database because I needed it to exist. Years of on-call taught me that the gap between "what the docs say" and "what the cluster does at 3 a.m." is where incidents live. So I reproduce that gap, on purpose, and write down exactly what happens — so the next engineer doesn't lose the four hours I did.