While I don't agree it's a Spendesk problem, it's definitely a hard position to take that Spendesk shouldn't fix it themselves. When a large (I'd guess, at mini

While I don't agree it's a Spendesk problem, it's definitely a hard position to take that Spendesk shouldn't fix it themselves. When a large (I'd guess, at minimum, double digit) percentage of your virtual card traffic goes through Amazon specifically, having your product trigger Amazon's fraud detection system, however broken it is, isn't a great look, and borders on being a bug. At the end of the day, the customer doesn't really care whether what Spendesk does is perfectly reasonable - they only care if it works. We have some other examples of this in software: * There was an article here on HN a couple of days ago, regarding how Windows 95 detected SimCity and treated memory allocations differently if it was running. (https://arstechnica.com/gadgets/2022/10/windows-95-went-the-extra-mile-to-ensure-compatibility-of-simcity-other-games/ https://arstechnica.com/gadgets/2022/10/windows-95-went-the-...) * Excel still treats 1900 as a leap year, which it was not, for greater compatibility with Lotus 1-2-3. https://learn.microsoft.com/en-us/office/troubleshoot/excel/wrongly-assumes-1900-is-leap-year https://learn.microsoft.com/en-us/office/troubleshoot/excel/... * The Linux kernel has an extremely strict policy of not breaking userland - to the point of keeping buggy behavior in place (I can't find a reference for this one, but I believe it one notorious case involved PulseAudio and one of Linus's earlier "moments").