For the moment, user feedback indicates that the Slow Add-on Notification appears to flag add-ons randomly. We need to make sure that this is not the case.

Yeah, I'm seeing e.g. DOM Inspector being flagged, which doesn't make a lot of sense.

Is there a way to log/dump the decisions that cause the warning to show up somehow - e.g. to the browser console to get an better idea why the notification shows up in unexpected situations?

Note that a new version of the AddonWatcher (and everything behind the scenes) has just landed in the very latest Nightly, which changes the triggers. There are currently two triggers. The first condition is an add-on causing the main thread to spend at least 2^7 consecutive milliseconds blocking the main thread by either owning the CPU and/or waiting for a system call – this is not influenced [anymore] by the machine going to sleep or by other applications eating CPU, but this may be influenced by a heavy disk load. This trigger is not reliable under Mac OS X yet (bug 1142457), and possibly not on Windows either (bug 1149897), but I'm working on fixing both issues. You can follow this locally in about:telemetry (and soon on the Telemetry dashboard) using Keyed Histogram MISBEHAVING_ADDONS_JANK_LEVEL. The second is an add-on blocking the parent process for more than 1s every 15s through a cross-process call (CPOWs). Due to bug 1150432, this is currently 1ms instead of 1s, I'll try and fix this today. You can follow this locally in about:telemetry (and soon on the Telemetry dashboard) using Keyed Histogram MISBEHAVING_ADDONS_CPOW_TIME_MS.

Oh, and I forgot to mention about:performance, which shows the same kind of data.

All the blockers are resolved.