Mad Tea Party was down for 24+ hrs. A datacenter problem. I’m not arguing a professional hosting service should have zero downtime. Failure itself is expected. Reliability is what happens after it. A full day of downtime is one full business day lost, and a good hosting service should be judged not only on physical performance, but on clear communication during a sudden failure and reasonable effort to reduce it.

What I bought, and what was sold to me, was a remote hosting plan for my blog. That is a fact. I doubt I have much of a compensation claim, not after whatever fine print and asterisks sit somewhere in the bill. This blog predates Cloudflare, and the list of things a hosting plan solved for me was long and real: reliable power, reliable network, my home IP kept private, commercial bandwidth, and a physical machine I don’t need to own and maintain. For two decades, a hosting service was the only practical answer — just another subscription to manage.

My grief started well before this growing frequency of outages. My plan would occasionally be throttled to roughly half of what I was paying for, and I found out after the fact rather than when the throttling kicked in. The diagnostics tools had nothing to offer. Based on what I could gather (the logs and the schedules), I concluded my backup job was the likely culprit. Likely. That was as close as I was ever going to get. Worse were the stretches where the shared box was simply unstable, and there was nothing on my side to check at all.

The outage itself followed the same shape. Notice arrived long after I already knew. The status updates ran optimistic enough that they kept being revised. And when it ended, nothing was said about the day itself: no credit, no post-mortem. The only thing on my dashboard was a note that support lines were busy due to the recent outage. Perfectly legal, I assume. But the criteria was never about failure itself. It was how they managed a failure, a risk. Handling it case by case, one ticket at a time, is not managing it. It was enough of a hint for me to notice the lack of system I was subscribing to.

Meanwhile the list had been quietly shrinking: origin IP, DDoS absorption, CDN, cached pages in case of a network outage on my end. One provider now covers nearly all of it, and covers it on a free tier. The registrar was always a separate bill, and it moved separately, for a plainer reason: the same provider sells domains at cost, and what used to be add-ons is now just standard.

What kept me put was never the features. It was the work: WordPress on a Linux box, the architecture around it, the operations I would now do myself. Self-hosting was always an option, and people took it — with compromises. That’s the part that changed. Not that the compromises went away, but that I could pick which ones I was willing to live with. Writing up a custom plan and having it reviewed is taxing work on its own. These AI services helped me through all of it, and then executed it on my behalf.