Research Findings
Short write-ups from homelab and engineering work. Everything here is tested and new: each homelab finding was reproduced in a test setup before publishing, and checked that it isn't already written up somewhere online. The nutrition review cites a primary source for every number.
-
How to eat healthy: the full research behind the six rules
The full evidence review behind six plain eating rules: what each one rests on, how strong that evidence is, and what result would overturn it.
-
Reolink camera cpuUsed metric does not track 4K encoder load
A Reolink camera's reported CPU usage does not track its 4K video encoder, so it can read low while the encoder is saturated and the camera is dropping frames.
-
Loki bug in LogQL wildcard alternation: the pattern silently matches nothing
In Grafana Loki, a log-search database, a wildcard OR pattern silently loses its wildcards and matches almost nothing, so a rule written that way never fires.
-
go2rtc bug in producer reconnect: one hung source freezes its stream for every client
In go2rtc, an open-source camera-stream router, a source that starts and then hangs holds a lock forever, so every client of that stream freezes until restart.
-
auditd syscall rules on a KVM host: 12x CPU overhead while guests idle
On a Linux machine that hosts virtual machines, audit-logging every system call costs about 12x CPU, burning several cores and ~100W while the guests idle.
-
Mitsubishi IR remote °F to °C mapping is a lookup table, not a formula
A US Mitsubishi heat-pump remote converts Fahrenheit to Celsius by an irregular table, so any infrared code set built by interpolating sends wrong temperatures.
-
NVIDIA Blackwell NVENC bug: H.265 codec header changes on every restart
An NVIDIA Blackwell GPU's H.265 encoder sends different codec-header bytes after every restart, with identical settings, which breaks browser video playback.