Is the data center alive?

On the morning of August 13, 2026, a power outage at a Phoenix, Arizona datacenter took this entire platform off the air — along with every other website in the building, and the hosting company's own homepage. No cyberattack, no disaster movie. The power went out. That incident is the first entry in the ledger above, and the reason this page exists.

Then it got more instructive. A few hours in, the domain name itself stopped resolving — a second, independent failure. The nameservers were alive and answering queries, but they had stopped serving our zone: no records, no authority flag, which every recursive resolver on the internet turns into SERVFAIL. For a while, everythingisperfect.org did not exist as far as the internet was concerned. At that point it wouldn't have mattered if the server had come back — nobody could have found it.

That is the lesson worth carrying, and it's more specific than "the cloud went down." Our compute, our website, our vendor relationship, and our DNS were all the same company. One bad morning in one building in Phoenix reached every one of them. Redundancy you buy from a single vendor is not redundancy; it's the same bet, written twice. Worth noting without malice: we checked, and the hosting company's own domain is served by a multi-provider nameserver set. Customer domains sit on the single cluster. The people who run the building know exactly what this failure mode is worth.

This platform spends a lot of its energy documenting the data-center buildout — the ~16 gigawatts proposed around our region, the ~289 gigawatts proposed nationally, the water, the power, the tax deals. The industry's pitch leans on an aura of permanence: mission-critical, hardened, redundant, always-on. Some of that is real engineering. But a building full of computers is still a building — it loses power, it floods, it overheats, it gets sold, its operator changes terms, its owner goes dark. When your community's records, archives, and conversations live in exactly one such building, they inherit every one of those failure modes at once.

We're not exempt, and we won't pretend to be. This server is a rented slice of a commercial datacenter in Phoenix, 400 miles from the communities it serves. The ledger above is us measuring our own landlord.

What this outage actually produced

Not a lesson — a second address. On the same day the datacenter and the DNS both failed, the project registered localnode.is through Iceland's national registry and stood up a mirror on an independent host in Reykjavík. It carries the ledger, the module map, and every part of this site that is just files rather than a database. It does not share a company, a country, a registry, a nameserver, or a power grid with the server that went dark.

Being precise about what that does and doesn't fix, because the difference matters: it means the project is reachable when Phoenix is not. It does not mean everythingisperfect.org would survive a repeat — that name is still served entirely by the vendor whose nameservers stopped answering, and fixing it needs a second DNS provider on the original domain. One of those two problems is now solved. The other is still open, and this ledger will keep saying so.

The plan: a network of local nodes

The measurement has a purpose. This platform's long-game is the local node framework: small, cheap, physically-present machines — the same family as our arcade-cabinet project — living in the communities they serve: a shop counter, a library corner, a garage shelf. Each node carries a copy of its community's archive and can serve it locally, even when the wider internet or a distant datacenter is having a bad day. The Yellow Pages module already sketches the off-grid end of this (Meshtastic mesh, no ISP at all); the nodes are the middle layer.

Whether that's romanticism or engineering is an empirical question, and this ledger is the instrument. We are collecting, in public:

• Availability — the datacenter's actual measured uptime, against the marketing nines, and eventually against the measured uptime of our own nodes.
• Time-to-recover — how long each outage lasts and what fixes it: for a rented server, waiting on someone else's electricians; for a local node, walking over and pressing a button.
• Cost — rented hosting per month vs. one-time hardware + a residential connection, per community served.
• Failure independence — one Phoenix breaker took down every "site" in the building at once; ten nodes in ten buildings don't share a breaker.

When the node network exists, its nodes join this same ledger as peers of the datacenter, and this page becomes the scoreboard of centralized versus community infrastructure. Until then it does the humbler job: keeping an honest public record every time the cloud turns out to be somebody else's building.