Must Know

A handful of ideas that hold the rest of this atlas up.

The ideas

A real system is the bar, not a toy. Every branch produces something that builds, runs, and does what production code does — at smaller scale, not lower rigor.
Concurrency is a property of the project, not a chapter. It gets harder exactly where the server and the database engine make it hard, not in the abstract.
Durability means you tested the crash, not that you wrote fsync(). A database engine that has never been killed mid-write hasn't proven anything.
Parsing shows up everywhere. The shell, HTTP, Git's object format, and SQL are all "turn text into a structure" — the regex-engine branch teaches the technique once, on purpose.
The spine project is versioned, not monolithic. v0 → v3, same discipline as Atlas I's OS milestones — a working v0 beats an imagined v3.
Atlas I is a prerequisite, not a reference you skim mid-project. If a syscall or a memory concept feels unfamiliar here, that's a sign to go back, not push through.