Deep reads through real codebases.
Short notes on the clever hacks and design decisions hiding inside the systems we use every day. New one roughly every week.
TigerBeetle Would Rather Crash Than Guess
Two asserts buried in a financial database. One panics if your hardware lies about time. The other panics if you allocate memory after startup. Together, they're the whole philosophy.
How OpenClaw Builds an AI Brain That Remembers, Sees, and Fails Over
Most AI agents break the moment an API goes down. OpenClaw doesn't. A week inside its codebase to understand the retry loops, permission pipelines, hybrid memory, and model failover that keep it running.
The Propulsion Principle: How Gas Town Makes AI Agents Stop Being Polite
Steve Yegge built an orchestrator for running 20-30 AI coding agents at once. 157k lines of Go, 100% written by Claude. The hack that keeps it running: literally type into the agent's terminal.
How Postgres Prevents Both Deadlock AND Livelock in Upserts
A one-line rule buried in PostgreSQL's INSERT ... ON CONFLICT code prevents two of the nastiest concurrency bugs at once: the transaction with the higher XID backs out.
How nginx Tells Time Without Asking
A lock-free time cache lets millions of nginx requests read the current time without waiting for each other. 64 pre-allocated slots, a memory barrier, and a bet that your thread won't sleep for a full minute.
The Hidden Message in SQLite's Random Number Generator
Four magic numbers in SQLite's source code spell out a secret phrase. It's not an easter egg - it's cryptography's way of saying "trust me, there's no backdoor."
The Bool That Holds Go's Ecosystem Hostage
A single boolean in Go's source code, if flipped from false to true, would instantly break thousands of packages. The Go team put it there on purpose.