Published counter was bumped up only when not migrating in, but it was
decremented when a migrating participant leaves without expectation to
resume. That could have resulted in negative counts.
Always change counters irrespective of migration or expected to resume
on leave. Control events send based on migration/resume.