Measuring 60+ support desks by hand, and publishing the correspondence

Not code. A measurement, which is its own kind of engineering, and the protocol is the whole thing. One identical question to every exchange, so results are comparable rather than a pile of anecdotes. Response time in whole days, with 0 meaning a same-day reply. Every reply archived in correspondence/, which is what turns a table of numbers from an assertion into something anyone can check.

The biases I did not remove are worth stating, because a clean table hides them. Time zones shift day-granularity by a full unit at the boundary. Channel matters, since a web form, an email address and a chat widget are different queues inside the same company. And a desk that recognises the question as research may answer differently, which nothing controls for. Day granularity was chosen because it is the resolution that survives all of that; hours would imply a precision the method does not have.

The data is also in a public Google Sheet. It fed bitcoin.org’s exchanges page. The result expires, support quality moves with headcount, which is why the date is in the title rather than a footnote.