Sorry those lines should not have been printed. I had modified the invariant checker to keep going but the change was incorrect and those lines print uninitialized data.
I checked further and found that the invariant checker produces a different order of records and maybe that's why the reduced list size is different. For example, here the serial schedule file has 2690 records, but the size seen in invariant checker is 2684 only.
Serial schedule: read 2690 records from the file and observed 3043 transition in the trace. Serial redux size = 2684
I added some code to dump the schedule in invariant_checker and see there is a difference in the order of the records.
For example
21,22c21,22
< 0001200 125859 13340877902486153 1 1
< 0001240 125856 13340877902486153 0 1
---
> 0001200 125856 13340877902486153 0 1
> 0001240 125859 13340877902486153 1 1
Then I noticed that the compare function used in raw2trace was different from the one used in invariant_checker. I updated raw2trace to use the same function. This fixed the invariant failure, but did not change the original scheduling issue.