Engineering notes
Short write-ups of real incidents from our own governed releases: what broke, why it broke, the fix that shipped, and what generalizes to your stack. Nothing here is aggregated or invented. Each note cites the published release it describes, and the list grows only when something real happens.
2026-08-05
Exit 0 with every vendor unreachable
Our launch tooling reported success while reading nothing at all. Why the empty plan file it wrote was more dangerous than the wrong exit code, and the contract that fixed both.
2026-08-05
A smoke test's first run caught a real bug
The weekly vendor smoke rejected its own GitHub Actions token. The bug was in how we asked whether a credential is valid, and the fix rests on a distinction worth stealing: validity and identity are different questions.
2026-08-04
The repository whose checkouts were born dirty
A merge-time digest probe failed on its first CI run. The cause: one file's committed bytes disagreed with the repo's own line-ending declaration, so every fresh clone started out modified before anyone touched it.
Why publish failures? Because this product's whole claim is that evidence beats assertion. A vendor that can show its own gate saying no, and its own smoke test catching its own bug, is making a different kind of promise than one that can't.