The bug depends on operating system level behaviors that may be different between Windows versions, and combinations of device drivers, and need human-machine interactions over physical MIDI devices.

We could have an automated test for this specific case if we write a special device driver to fake physical device. But still it can not cover similar platform dependent problems, and it's practical to install such drivers onto all bots.

This time, the MidiManager for Windows was re-written from scratch. That's the reason why I mistakenly introduced such platform dependent regression that happens in edge cases (I actually ran manual tests, but could not catch this). But such large modification rarely happens. So, writing device driver won't be a reasonable idea.

I guess you are seeing many device specific regressions, and could have a good idea.
E.g. Chrome Apps has many APIs that requires special devices to test (https://developer.chrome.com/apps/api_index). Also Web Bluetooth and Web USB would have similar risks. Any idea is appreciated.

Tentatively assign you to hear your thoughts.