I designed and printed a spacer for 6 IDE power cables. The cluster now lives on the shelves above my desk.

I designed a fastener to hold the front plate flush since there are no sockets or plugs to create inward pressure.

I started questioning my benchmarks because results were becoming fantastic. In response, I built a new binary patch corpus. The process took on the order of 1000 CPU hours but produced about 80k high-quality binary patches.
https://github.com/barrbrain/bsdiff-analysis

After another deep dive, I can configure ingress with local IP addresses and hostnames. MetalLB assigns addresses from a pool set aside by my WiFi router. Hostnames are advertised by avahi-daemon. Ingress is routed by contour-envoy.

I finally got https://k0sproject.io running on this system. It was blocked on emptyDir mounts backed by NFS and resolved by backing with tmpfs.

Hardware and software are both sufficiently complete to build Chromium quietly.
https://youtu.be/CV_ZNRSWjEI

Finally, I reassembled and correctly configured networking after returning the top board for warranty. Cooling is quieter but still reliable after adding dust filters to the design, even when running 40 cores at maximum power.

Oops, I mirrored the last 2 panels. Now, we can access the power and reset buttons. I broke up the emptiness of the middle by etching some logotype.

The print adhesive resolved the curling and related quality issues. It was still necessary to use a brim, which took quite some time to clean off all the edges and cut-outs. The panels hold each other in place without glue, but I will add some later for durability.

Printed in the correct orientation, but the same corner of the print bed curled again. I resolved to wait for the adhesive delivery before attempting any more prints. Meanwhile, this is close to the vision.