What Screeny's chest-counter runtime taught me about automating an opaque browser game: when the useful state only exists as pixels, the fastest reliable system is a hybrid of calibrated coordinates and small visual probes.
Remote NixOS deploys get dramatically faster when the target host fetches public dependencies from its own substituters and you only push the private bits. The key is `--use-substitutes` and the trusted-user detail most guides skip.
NUR flips the AUR model inside out. Instead of one giant package repo, every maintainer keeps their own repository and NUR indexes the lot. That gives you composability, selective trust, and a much wider ecosystem than most Nix users realize.
The practical Codex app workflow I settled into: one long-lived Triage thread, one focused Fix issue-XXX thread per task, and separate Research threads only when they actually earn their keep.
Turning nftables rule counters into Prometheus time series with a custom exporter and Grafana dashboards, using rule comments as stable human-readable labels.
A tour of Hrafnsyn, a Phoenix LiveView app that merges ADS-B aircraft and AIS vessel data into one operational map, packages cleanly with Nix, and now runs as a real deployed service.
Turning a Raspberry Pi 4 and a u-blox ZED-F9P into a stratum 1 LAN time server with chrony, PPS, Prometheus, and Grafana, including the ntpd detour that wasted a perfectly good evening.
For headless Raspberry Pi bring-up on NixOS, image-time secret injection is often more reliable than first-boot decryption on the device. Here's the provisioning flow, the brittle design we tried first, and why bootstrap state belongs under /var/lib rather than /etc.
Monitoring two Eaton UPS units across FreeBSD and Arch Linux ARM with NUT, DRuggeri's nut_exporter, Prometheus, and Grafana.
Get Gardena smart system data out of the phone app and into Prometheus and Grafana with a Rust exporter, a NixOS module, and per-valve water-use estimates that are honest about what the hardware can and cannot measure.