This article seems fairly scant on details to be honest. I understand how they came to the conclusion, but there's still quite a lot of conjecture here.
I feel like the main problem is the inability to properly debug your problem without paying while the plan essentially necessitates a degree of self-diagnostic when things go wrong. I also serve an xml feed via CF and also use it as a caching service in front of my Backblaze B2 bucket, and on the free tier by virtue of the files I'm serving far fewer unique users (less than half of the author's last 30 days) but a far greater amount of data (111GB). I'm not certain if the author should jump from what's clearly a vague and uncommitted conjecture from the Cloudflare tech about where the 503 error comes from straight to "they are definitely throttling me", since I've had the opposite experience and have seen 503s in all sorts of situations unrelated to throttling, but with a tiny sample size and no adequate diagnostic tooling it's also impossible for me or anyone to say that he's not being throttled, except CF themselves. I realize that it's their business model but it seems counter-intuitive that the service they sell seems partly "better ability for you to fix your problems yourself", sort of the opposite of say, managed/unmanaged servers. It'd make more sense, from a customer's point of view, that if I'm paying for a service, there should be service in both senses of the word - the product and the servicing of issues if possible, but not making diagnostics more transparent makes little sense because it prevents people who want to use the service from continuing to use the service, presumably part of CF's goal.
