This is different than failing to bisect into rolls. This is more about results like the ones below. Note that the last good and the first bad revisions should in theory be identical b/c chromium@179 is the roll, and 6dd8 is the last revision in that roll, yet they produce very different results.
===== TESTED REVISIONS =====
Revision Mean Std Dev N Good?
chromium@418177 4475222 137863 5 good
chromium@418178 4087273 889650 8 good
chromium@418178,catapult@c49b5f9b94 5829497 1943097 8 good
chromium@418178,catapult@6dd8114959 5370724 1392510 5 good
chromium@418179 9086814 592378 5 bad <--
chromium@418180 8616816 770426 5 bad
chromium@418182 8665877 625060 5 bad
chromium@418186 8790856 845366 5 bad
(from: https://bugs.chromium.org/p/chromium/issues/detail?id=646521#c13 )
Comment 1 by robert...@chromium.org
, Sep 23 2016