[ https://issues.jboss.org/browse/DROOLS-1326?page=com.atlassian.jira.plugi... ] Geoffrey De Smet commented on DROOLS-1326: ------------------------------------------ That went into a different path, and a apparently this is a rare bug, so it didn't reproduce there. The first solver should have a CH stepCountLimit termination of 403 steps. The second solver should run the rest of the CH entirely and have an LS stepCountLimit termination of 1233 steps. The third solver should just run. At each point confirm that the score is the same as in the original log. There's still an edge case that that doesn't reproduce it - if the the last step of the first solver's CH saw only half the moves and picked a less optimal one - but let's worry about that if we see that. Let's talk on Monday if you can use more info. -- This message was sent by Atlassian JIRA (v6.4.11#64026)