:: 10-Jun-2002 18:45 GMT (Monday) ::
For those who have noticed differences in the numbers, this is most likely
due to bugs that were present in the old stats code. We run an audit script
at the end of every daily run of the new system that does a sanity check on
the numbers represented by all the different parts of the system. It would be
difficult to overstate it’s value; at least a dozen bugs have been identified
thanks to this script as the new stats code evolved from it’s inception.
Because of this audit script, I have a very high degree of confidence in the
new stats code. We migrated the rc5 data from the old system into the new
system several months ago. Part of that migration was modifying the raw data as
best we could so that everything would actually pass the audit script. Since
that time, rc5 logs have been processed by the new system each night, as well
as the old system. This period of both systems running is what has allowed
people to do most of the before/after comparisons I’ve seen.
What this boils down to is that any rc5 stats processed by the new system
(ie: anything in the past month or two) can be assumed to be 100% correct and
accurate as compared to the logs provided by the master. Anything prior to that
time could still be off.
At some time in the future, I would like to start re-processing all of the rc5
logs from the beginning of the project. This should correct any other errors
that might exist in the rc5 stats.