The problem occurs where IVC reduction appears to be attempted on a property that has (possibly) been falsified.

These errors occur intermittently but on some models more consistently than on others. Running the same model multiple times the error might or might not occur. When the error does not occur, the property(ies) in question is(are) valid. When it does occur, the property(ies) is(are) unknown. It appears to be some sort of race condition.
Initial analysis indicates that the errors do rarely occur in Java VM version 1.8 but occur much more frequently in later JVM versions.
Application of the same Lustre file to multiple versions indicates the errors may occur in (at least) JKind versions 4.4.4, 4.5.0, and 4.5.2.
The it seems that setting pdr_max to zero either eliminates the bug or makes it so much less probably that it was not detected in a few dozen runs.
I am struggling to put together a simple model that illustrates the problem; changing anything appears to affect the likelihood of occurrence.
The problem occurs where IVC reduction appears to be attempted on a property that has (possibly) been falsified.
These errors occur intermittently but on some models more consistently than on others. Running the same model multiple times the error might or might not occur. When the error does not occur, the property(ies) in question is(are) valid. When it does occur, the property(ies) is(are) unknown. It appears to be some sort of race condition.
Initial analysis indicates that the errors do rarely occur in Java VM version 1.8 but occur much more frequently in later JVM versions.
Application of the same Lustre file to multiple versions indicates the errors may occur in (at least) JKind versions 4.4.4, 4.5.0, and 4.5.2.
The it seems that setting pdr_max to zero either eliminates the bug or makes it so much less probably that it was not detected in a few dozen runs.
I am struggling to put together a simple model that illustrates the problem; changing anything appears to affect the likelihood of occurrence.