The docs are a claim
Their own documented example, rejected by their own server.
I stopped reading and wrote an endpoint that did nothing but test the two things I was about to build on. Both assumptions failed. An hour of testing replaced a week of building the wrong thing.
A throwaway endpoint is cheaper than a wrong architecture.
WHAT I TESTED
An endpoint that did nothing else
Two assumptions were load-bearing and neither was checked: that I could stream named events, and that a cache would sit somewhere I could reach. So I built the smallest thing that could answer both and threw it away afterward.
WHAT CAME BACK
Rejected, and by its own documentation
The documented form for named streaming events does not parse. Not deprecated, not undocumented — published, copied verbatim, refused. The cache I had assumed is not on this plan at all, and nothing in the plan comparison says so.
WHY IT HOLDS
It was wrong in both directions
A vendor overstating what works is the expected failure. This one also understated it: the same test found a cheaper path I would not have looked for. Documentation describes intent. Only the instance tells you what it does.