Reader sent Geerling's CM4 PTP write-up, which shows two CM4s syncing to 10-15ns.
Our headline overclaimed ('impossible on a Pi 4') and our mechanism was wrong (we
blamed the bcmgenet MAC for not exposing a PHC).
The real story is better: the CM4 (BCM54210PE) and the 4B (BCM54213PE) report the
IDENTICAL PHY ID 0x600d84a2, so the kernel cannot tell them apart by asking the
chip. bcm-phy-ptp.c disambiguates by MDIO bus address instead — addr 0 = CM4,
proceed; addr 1 = 4B, return NULL. Ours is addr 1.
It is not a kernel config gap and not our RT patch: CONFIG_BCM_NET_PHYPTP=y and
CONFIG_NETWORK_PHY_TIMESTAMPING=y are both set. The driver loads, looks at the
board, and declines.
Correction is visible on the page rather than quietly patched. Credit to Geerling.
Starlight renders the logo as an <img>, so CSS can't reach inside it. The single
cream-on-near-black SVG left a cream wordmark on a cream page. Ship a light
variant with warm ink and the darker brass the light theme already uses.
The landing page's headline results table was rendering as a paragraph of raw
pipe characters. Astro applies remark-gfm to .md but it wasn't reaching the MDX
pipeline, so every .md page rendered tables fine and only index.mdx was broken —
which is exactly why it went unnoticed. mdx({remarkPlugins:[remarkGfm]}) fixes it.
cuckoo.warehack.ing stays as a permanent redirect; it was live briefly.
Rebuilt with KBUILD_BUILD_USER/HOST/TIMESTAMP pinned. Without them the kernel
bakes the builder's user@hostname and wall-clock build time into /proc/version,
which publishes operator identity in a public artifact and makes the build
unreproducible. Pinned, anyone can rebuild and diff the bytes — the only trust
story available for an unsigned community kernel.
The installer refuses to run on anything but a Pi 4, refuses a mismatched
image/modules pair (vermagic failure means no /dev/pps0 on a kernel that
otherwise boots fine — silent and miserable to debug), verifies SHA256SUMS,
never touches kernel8.img, and leaves rollback as deleting one config.txt line.
Boot-tested on the Pi: Stratum 1, 273 ns offset, PPS handler confirmed NOT
threaded. Binaries ship as release assets, not in git.
Nine cuckoos on an idle click is a lot. There's no singleclick event and click
always fires before dblclick, so the single call is deferred by one double-click
interval (260ms) and cancelled if a second click lands.
The clock strikes the current hour. Two things in it are mechanism rather than
decoration:
- The pendulum TICKS. steps(2) over 2s = two discrete positions, one per second.
That is what an escapement does — it chops continuous energy into countable
ticks. A swept pendulum would contradict the entire site.
- The call is SYNTHESIZED, not sampled. A real cuckoo clock is two wooden pipes
a minor third apart (6:5, so 700 Hz -> 583 Hz), pushed by a bellows. That's
twenty lines of Web Audio: triangle pipes with a slight pitch sag as the
bellows empties, plus a puff of bandpassed noise for the breath. The sound is
the mechanism rebuilt, and the hero ships zero bytes of audio.
The talk is regenerated for a public audience — the finding, not the
procurement case. Transcript ships with it: audio is the alternative to reading,
never a replacement for the text. preload=none so the 800 KB only downloads for
readers who actually press play.
What the Raspberry Pi time-server guides get wrong on a Pi 4, with the
measurements. The headline artifact is a four-line pps-gpio patch: PREEMPT_RT
force-threads IRQ handlers, and pps-gpio takes its timestamp inside its handler,
so the realtime kernel puts a scheduler between the electrical edge and the
clock. IRQF_NO_THREAD takes RMS offset from 2468 ns to 199 ns.
- kernel/ the patch
- dashboard/ live status page (position hidden by default)
- docs-site/ the write-up (Astro/Starlight, brass, no tutorial section)
