Firefox/SHIELD

The SHIELD server hosts "recipes" that are fetched by Firefox once per day. These recipes help User Advocacy both gather feedback from users so that they can better advocate for their needs, and help them offer choices to users for things like fixing common problems with known solutions in Firefox.

Status

• Project Owner: Gregg Lind [:gregglind]
• Product Manager: Matt Grimes [:Matt_G]
• Development: Michael Cooper [:mythmon], Michael Kelly [:Osmose], Support Engineering
• Status: Planning
• Scheduled for: 2016 q1 and q2

History

• June 2014 - mconner files a tracker bug for enhancing FHR to proactively detect and solve problems for users.
• December 2014 - bsmedberg files bug 1111016 for implementing the smallest-possible set of changes necessary to support Heartbeat (a prompt shown to users asking them to rate Firefox) and Self-Repair (a minimal version of the problem-solving functionality, described in detail in dependent bugs).
• Early 2015
  • UITour is enhanced to support Heartbeat prompts, sending user responses to input.mozilla.org.
  • Self-Repair is implemented as a set of static files hosted on AWS that Firefox retrieves.
• Late 2015 - Self Repair is renamed to Shield since the project covers more than just repairing Firefox. The name covers the server that deploys configuration files (recipes) and the in-product component that executes the recipe.
  • Example 1: The Shield server deploys a recipe to a client indicating that the user should be included in the daily Heartbeat sample. The in-product component responds by offering the Heartbeat prompt.
  • Example 2: The Shield server matches a “preference repair recipe” to the user configuration, and returns it. Firefox runs the recipe steps to fix the broken preference.

Terminology

Action

A semantic thing that can be done by the browser, like "Display a survey" or "Highlight this menu item". Actions may be paired with executable code for self-repair, but other services may read the actions and implement them differently.

Recipe

A list of actions to execute to achieve something, paired with eligibility rules determining who they get sent to. For example, a recipe may check to see if a preference is set to a known-bad value, and offer to reset it for the user.

Eligibility Rule

A rule describing who a recipe should be sent to. For example, a survey may only be relevant to users in India, so the eligibility rules for that survey recipe will limit it to users located in India only. Other filters available include locale, sampling rate, and start/end dates.

Bundle

A bundle is a collection of recipes that are sent to Firefox. Bundles are built based on eligibility rules.

Phase 1

Summary

Phase 1 will consist of writing a server to replace the current statically-generated self-repair scripts. The server will be capable of sampling, limiting recipes to a certain number of users, timed releases of recipes, localized recipes, and geolocation.

Requirements

1. Serve bundles of recipes to all Firefox instances, once per day, filtered by eligibility rules.
2. Present webpage for creating and managing recipes and the rules determining who sees them.
3. Control rollout of recipes with composable rules:
   • Distribute recipe to approximately X users, then disable it.
   • Distribute recipe starting on X date/time and/or ending on Y date/time.
   • Distribute recipe to X% of users.
   • Distribute recipe to users in a particular country.
   • Distribute recipe to users based on the locale of Firefox (Ex: ru, de, en-GB).
   • Distribute recipe to users on a specific release channel (including non-standard channels like funnelcake).
4. Expose debugging view for previewing the set of recipes being sent to a specific browser/location configuration.

Measuring success

How do we know that we've won? What do we want to be measuring?

Bugs

• Firefox client-side: https://bugzilla.mozilla.org/show_bug.cgi?id=1233804

https://bugzilla.mozilla.org/show_bug.cgi?id=1234393

Future phases

Possible future improvements may include:

• A debugging view for previewing a specific recipe's functionality.
• Automatically modify sample rates to reach a certain minimum amount of users over time. For example, "Reach 500 users per day, spread out over the entire day".
• Chain recipes with auto-disable rules, such that when one is disabled the next one is enabled.