See recent messages in mozilla.dev.tree-management with the subject "Talos Regression :( Trace Malloc Leaks increase 14.5% on MacOSX 10.5.2 Firefox" > Regression Trace Malloc Leaks increase 14.5% on MacOSX 10.5.2 Firefox > ------------------------------------------------------------------------ > Previous: avg 1815027.333 stddev 138039.291 of 30 runs up to revision 8d9465a318f5 > New : avg 2078330.000 stddev 3397.793 of 5 runs since revision ba49fff91d82 > Change : +263302.667 (14.5% / z=1.907) > Graph : http://mzl.la/fiaiGF > > Changeset range: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=8d9465a318f5&tochange=ba49fff91d82 > Looking at the graph at http://mzl.la/fiaiGF this should quite clearly have been blamed on changeset 2736d8740c66 instead.

I've seen this many times before with changesets I know moved the numbers not getting the blame.

The algorithm used to detect regressions doesn't cope well with relatively stable data, it's better suited to noisy data. It's looking at sliding windows of data and raises a warning when the current window has shifted too much from the historical data. In this case, it took a few revisions before the mean value shifted enough to raise a warning. I'll see if I can tweak the parameters for this type of test to find the correct revision more often.

With the patches from the dependent bugs, I'm now getting much more accurate blame for all of the Talos data I've tested. For details, see bug 858735, bug 858756, and bug 858877.

Found in triage.

Does this system still exist? Does this bug?

Still exists and current detection is adequate. However, detection is moving to perfherder in the near term. So this is FIXED/WFM/WONTFIX.