In the load balancer analysis, I took the presence of load balanced paths containing the given LB into account. In particular the count of UDP LBs was reduced in the UDP LB not TCP LB case, where a common destination non LB TCP path containing a clump existed. This modified value was used to recalculate two statistics: TCP LB and UDP LB count over UDP LB count* and UDP LB and not TCP LB count over UDP LB count*.
In the LB path analysis the percentage of UDP TCP differences that could be explained by clumps was recalculated if there were cases of a TCP path containing a clump but also containing a corresponding UDP LB node that is not load balancing for TCP. The result reduced from 99% to 98%.
Case 9 from the LB path analysis was investigated by looking at warts files. This involves TCP LB paths which are not also UDP LB paths where corresp. UDP path does not contain a clump. Here we see a TCP path that has a load balancer and a UDP path where the node does not load balance. It was decided that improved data collection is required to validate these small numbers in this category. Steps are underway to confirm this type of result at the time of probing the Internet. It would appear that part of this is to give scamper MDA a mode that is better than 99% confidence for use on special occasions.