Skip to content

Blog

Field notes from
the deep end

Short essays on root-cause analysis, engineering leadership, and building systems that survive the people who built them.

6 entriesTopics#api-clients#btrfs#error-handling#go#hypermedia#kubernetes#linux#methodology#nixos#observability#sse#storage#type-designRSS

  1. How a 7-year-old PCIe 3.0 M.2 SSD can be 57× faster than a brand-new PCIe 4.0 M.2 SSD in real-world use cases

    GMKtec EVO-X2, Ryzen AI Max+ 395, 128 GB RAM, 2 TB QLC NVMe — my experience as a developer on NixOS, Btrfs, Linux: kernel freezes, disks, and what I learned.

    23 min read#linux#btrfs#storage#nixos

  2. This projects page is a build artifact

    Every claim on /projects gets checked the boring way: numbers are computed at build time, install commands were executed in a clean sandbox before shipping.

    4 min readUpdated 2026-09-05#go#methodology

  3. I wanted a live UI. I built two Go libraries.

    Every Go SSE handler reinvents the same four pieces. I extracted them, then built go-datastar on top — where patches are values.

    4 min read#go#sse#hypermedia

  4. Every Go program has if err != nil. Then what?

    Go's most-written line answers nothing: retry? exit code? whose fault? I built an error protocol. Also: 52 suppressions for a linter that never existed.

    5 min read#go#error-handling#type-design

  5. My /health endpoint was a snitch.

    Redis sneezed, my /health endpoint said 503, and the kubelet pulled the trigger. Three-probe health checks with opinions, plus a live status page.

    5 min read#go#kubernetes#observability

  6. I needed a Go SDK for Wise. Nobody built one.

    In 2020 I used Wise's API directly. Years later, no Go SDK existed — so I built one. And discovered the API returns money as float64.

    4 min read#go#type-design#api-clients