It depends on what you're measuring. The aggregates are recorded by build_id, submission_date, OS, OS version, channel, major version, e10s enabled setting, application name, and architecture. They can be returned aggregated against any of those dimensions except the first two. So the error depends on the values distributed amongst those dimensions... which is a lot of things to check.
Generally speaking we don't add a use counter for anything that is used a lot. Even people who see these features being used don't see them used a lot. The only other candidate for largest change would be GetPreventDefault (https://mzl.la/2mQqYMR), in the sub-2% category. Most use counters are less than 0.1% usage. So on the count of "baseline sense for how much I should care"... if you have put in a Use Counter and thought usage was too high to remove that feature, look again. Re-evaluate with the new, correct, data. If you have considered using Use Counters but heard they couldn't be trusted, they can. Now. If neither of those things apply, this is just a note that Use Counters exist and can be useful. Caring is optional :) :chutten On Wed, Jan 17, 2018 at 12:51 PM, Steve Fink <sf...@mozilla.com> wrote: > On 1/17/18 7:57 AM, Chris Hutten-Czapski wrote: > >> Hello, >> >> Use Counters[0] as reported by the Telemetry Aggregator (via the HTTPS >> API, and the aggregates dashboards on telemetry.mozilla.org) have been >> over-reporting usage since bug 1204994[1] (about the middle of September, >> 2015). They are now fixed [2], and in the course of fixing it, :gfritzsche >> prepared a nifty view [3] of them that performs the fix client-side. >> > > Can you give a sense for the size of the error in observed counts? (Is > this a 15% should-have-been 10% type of thing, or almost always a 15.001% > should-have-been 15% type of thing?) > > Could you order all of the counts by percentage error and send out a list > of the top 10 or so? (eg in your blog post, you had PROPERTY_FILL_DOCUMENT > going from 10.00% -> 9.46%, for an error of 0.54%. If that 0.54% is in the > top 10, report the measure with its old and new percentages.) Really, it > ought to be the list of measures, if any, that crossed over some "decision > threshold", but I'm assuming that for the most part we don't have specific > thresholds. > > I'm just trying to get a baseline sense for how much I should care. :-) > > > Thanks, > > Steve > > > _______________________________________________ > dev-platform mailing list > dev-platform@lists.mozilla.org > https://lists.mozilla.org/listinfo/dev-platform > _______________________________________________ dev-platform mailing list dev-platform@lists.mozilla.org https://lists.mozilla.org/listinfo/dev-platform
