Field Notes

Sheet
N-000
Set
Notes
Status
Issued

Notes from production engineering.

Write-ups from real work on real systems: what broke, how it was found, and what fixed it. The detail we would want if we were reading someone else's notes.

N-000NotesIssued

Field Notes index

  • N-101 · · 4 min read

    The reviewer was wrong about the client IP

    An independent AI review, a merged fix, passing tests, and a rate limiter that got weaker. A live probe with forged headers caught what review could not.

    security · review · infrastructure

  • N-102 · · 4 min read

    Row-level security is the first thing we check

    Why a Supabase app built with an AI builder can expose its database to a public key, what CVE-2025-48757 showed, and two queries that find it.

    security · supabase · prototype-to-production

  • N-103 · · 3 min read

    How we counted 20,900 tests

    The commands behind the track record on our home page, where the count is imperfect, and what a big test number does and does not prove.

    evidence · testing

  • N-104 · · 4 min read

    Seven diagrams in parallel, and the two defects only a person caught

    How our AI engineering system redrew every diagram on this site for phones in parallel, and the two defects only a person looking at screenshots caught.

    ai-engineering · process

  • N-105 · · 4 min read

    Our animated drawings burned a CPU core doing nothing

    Fourteen animated figures took 25 to 37% of a core, even off screen. How we measured it, the control that found the cause, and the player that took it to zero.

    performance · frontend · animation

  • N-106 · · 3 min read

    The page that shipped uncompressed

    Our home page went out as 122 KB of uncompressed HTML from launch until we measured the bytes on the wire. The fix, the tests, and the check that now runs after every deploy.

    performance · infrastructure

Want this kind of engineering on your system?

Tell us what you're building. We'll tell you honestly what we'd check first.

Next sheet · N‑101 The reviewer was wrong