Visually distinguish between merged points and unmerged points about areweslimyet

[Nephyrin]

The problem with this is that soon there wont be any non-coalesced points - we are going to start running multiple tests on each builds to combat noise, so any given timestamp will have multiple tests. Perhaps once zoomed in enough we could put small numbers under each point to indicate how many tests they represent.

So the issue in bug 842756 is that the trend is only visible from a further zoom level, whereas its too noisy zoomed in to differentiate - if you turn on ?nocondense and zoom to similar levels you can still make out the trend, it's not like the coalescing is introducing phantom effects. Are we worried that coalescing is misleading somehow?

[jlebar]

Perhaps once zoomed in enough we could put small numbers under each point to indicate how many tests they
represent.

If that's not too visually noisy, that would work.

Are we worried that coalescing is misleading somehow?

I think it's just that you have to interpret coalesced data differently than non-coalesced data. The fact that we use medians in the coalesced data means that we have the potential to add a certain type of noise which exaggerates differences; I think that's what we're seeing here.

If we used box-and-whisker plots, this would be better, I guess. :)