This issue isn't about how midi works or about the way the Keystep pauses - it is an issue with how the Keystep resumes after it has been paused by the source that sends start/stop messages, when on the latest firmware (1.0.1.15)
I have the same issue exactly as described by mememe. This was not an issue in firmware 1.0.0.5
I updated the firmware to 1.0.1.15 yesterday and immediately had this issue.
For testing, I connected together the Keystep, an MPC2500 (master clock), an Access Virus C, and also a TR-8 so timing will be absolutely obvious even if only slightly out (though timing goes out mostly by 1 or more seconds with this issue so..)
On Firmware 1.0.0.5:
1. I have the Keystep switched to sequencer mode (and obviously will be sequencing the Virus)
2. I press 'PLAY' or 'PLAY START' on the MPC and the Keystep starts playing the sequence (triggering the Virus sounds), in time with the MPC and TR-8, from the start.
3. I press 'STOP' on the MPC and the Keystep pauses (as it has always done) and the TR-8 stops too.
4. I press 'PLAY START' (or hold 'STOP' and press 'PLAY START') on the MPC to start from the start (and not resume/continue as pressing the 'PLAY' button on the MPC would do.)
5. The Keystep plays exactly in time with the MPC and the TR-8, from the very start of the sequence ..every time. Works beautifully.
On Firmware 1.0.1.15:
Steps 1 to 4 as above
Step 5. The MPC and TR-8 sounds/beats play from the start and are in perfect time together (as is all my other gear when added to the chain). The Keystep (Virus) starts playing at a completely random time. Anything from .5 sec to maybe 5 seconds after pressing 'PLAY START'. Totally unusable.
I tried other gear in place of the Virus but same issue. The play button on the Keystep can be seen light up at the random time anyway so it's clear the issue is with the Keystep, not the instrument it's sequencing.
I rolled back to 1.0.0.5 and tested and the issue was gone. I updated to 1.0.1.15 again and the issue came back.
Finally I went back to 1.0.0.5 again and it is perfect again as expected so it's staying on that. I can't imagine this is by design so must be a bug in the newest firmware.
Pity, because the new firmware fixes a lot of other issues, though there were none that made it unusable like this bug it introduced.
Hopefully will be acknowledged and fixed soon.