One fix, found by reading the code rather than from a report.

An operator pause on a pre-generation run undid itself

/vdt pregen pause and /vdt pregen stop both worked, reported success, and were then quietly reversed.

The per-tick driver tracks two things: whether a run is paused, and whether the plugin paused it itself. Anything that blocks generation - a player logging in, an MSPT spike, a dip in free heap - wrote both. So the first blocker to come along relabelled an operator's pause as the plugin's own, and the release that follows lifted it as soon as that blocker cleared. The run carried on generating terrain nobody had asked it to resume, and /vdt pregen status showed the blocker as the reason instead of the command that actually paused it.

The failure path is the worse half. When a chunk request is refused, the run stops rather than cancels, specifically so the request is not retried in a loop and the progress survives until someone looks at the cause. That containment ran through the same operator pause, so it did not hold: the run came back on its own and asked for the same failing chunk again.

An operator pause is now settled before any blocker is looked at.

Upgrading

No config changes and no new keys. Replace the jar and restart.