If a run is not completed, compliance computation goes mad
If a run is not complete (i.e, we don't have the "end run" report), then the compliance computation for the node goes made and report everything missing and/or unexpected.
The technical problem seems to be that the non complete run should be ignored and is taken into account. As we don't have the end report, the run is seen as not having a configuration ID. So Rudder thinks it's a run from an agent before 3.0, look what could be the expected configuration for it (a very old one, the first with a configuration ID), and see that nothing matches: either the configuration expired a lot of time ago, or only a few or none of the expected reports matche the ones received.