925515 - Remove filter on recorded search engine identifiers

Status: RESOLVED FIXED

(Firefox Health Report Graveyard :: Data Collection, defect)

Description

[Mike Connor [:mconnor]]

This is the bug to flag the (potential) payload changes.

Now that we have FHR data coming in, we've discovered that the number of users falling into the "other" bucket is much larger than expected, both in share of search volume as well as users with unknown search engines.  Data we've obtained from studies (Search Tabs) and partners has confirmed that hijacking of search defaults is more widespread than previously believed, and more importantly, users struggle to undo these actions.  Most users simple resort to directly visiting search engines instead of using the built-in UI, which is not an optimal experience.  Anecdotally, we have reports that due to sideloading and other techniques, users are actually unaware of the provenance of these search engines, and assume they came from Mozilla.

To help us detect and flag this type of hijacking, I propose that we should stop filtering search providers into an "other" bucket, and directly identify them in all cases.  This would allow us to identify potentially-hijacked users and provide them with an effective remedy within the FHR feature.

The potentially for disclosing personal  information is pretty low here, certainly no worse than add-ons, and would require someone to intentionally disclose that information in the identifier for the search plugin.  The initial discussion indicated this was an extremely unlikely case, and an acceptable risk vs. the user benefit.

Comment 1

[Richard Newman [:rnewman]]

Patch to follow shortly…

        "org.mozilla.searches.counts": {
          "_v": 3,
          "eBay.searchbar": 1,
          "other-IMDB (akas).searchbar": 1,
          "other-Mycroft Project.searchbar": 1,
          "twitter.searchbar": 1
        },

Comment 2

[Richard Newman [:rnewman]]

Note that I plan a follow-up to this that alters how searches are threaded through, which will avoid some 'other's.

Comment 3

[Richard Newman [:rnewman]]

Attachment: Part 1: correct behavior for installed engines. v1

I plan a part 2 that addresses some of the comments in this bug, and allows an engine added at runtime to be recorded with a non 'other' ID.

Comment 4

[Richard Newman [:rnewman]]

Attachment: Part 2: extract identifier dynamically. v1

This should perform better (no _initialize), and will also catch runtime engine additions.

Comment 5

[Richard Newman [:rnewman]]

Fixed in 28. See dependency bugs.

Comment 6

[Mike Connor [:mconnor]]

In the interest of getting data (and resolutions) to our users sooner rather than later, I'd like to see this uplifted as far as we can, at least to 27, possibly to 26 if it's super safe.