Nimbus Features
Nimbus is a full-featured experimentation platform that provides configuration, analysis and client libraries for both experiments and rollouts.
Experimenter Console#
- Simple experiment configuration UI
- Pre-launch testing support
- Multifeature experiments
- Rollouts
- Simple targeting
- Advanced customizable targeting
- Remote settings integration
- Experiment Review / Approval Workflow
- Dashboards
- Monitoring and Analysis
Experiment Types#
| Branches | Control | Co-existence / Avoidance | Enrollment | Analysis | Good for? | Permanent Change | |
|---|---|---|---|---|---|---|---|
| Experiment | A/B - equal sizes. Size fixed at launch - cannot grow. | Y | One per feature at a time. Avoids other live experiments on a specific feature. Keep as small as possible to learn. | Fixed timeframe. 7 day typically. Analysis only AFTER enrollment ends | Automated for default metrics. Can select outcomes or create new custom | Stat sig differences from control + absolute counts. | No - all changes delivered through nimbus reverted. |
| Message experiment | Same AND each branch is given a messageID that the messaging system uses. | Same as experiment | Same as experiment AND message system throttles max # per time frame | Same as experiment | Same as experiment | Same as experiment | N Same as experiment |
| Multi-feature experiment | Same as experiment | Same as experiment | Avoids other live experiment with EITHER FEATURE - **restricted audience | Same as experiment | Same as experiment | Same as experiment | N Same as experiment |
| Holdback Experiment | Go right to end stage of rollout - 90% treatment | Y - 10% | Same as experiment AND no other feature can be tested in this feature for the duration. | Same as experiment | Same as experiment | Decision made to deploy quickly - but want data. A brief step before shipping as default in tree. | No - all changes delivered through nimbus reverted. Fast follow shipping in tree. |
Feature Delivery#
| Branches | Control | Co-existence / Avoidance | Enrollment | Analysis | Good for? | Permanent Change | |
|---|---|---|---|---|---|---|---|
| Rollout Today | Size increases 5%, 25%, 50%, 100% | N | Unrestricted co-existence Does not avoid other rollouts or experiments. | N/A | None - teams can make and monitor their own dashboards - but no “stat sig analysis” offered. | Risk mitigation for explosive failures (looking for crash, bugs, media issues) | Yes. Changes do not revert on targeted versions. Teams ship as default in tree in the release at 100%. |
| Branch Promotion | 1 - 100% | N | On an experiment with a winning branch - you can select “promote to rollout” which clones the ticket with the winning configuration filled in | N | N | Get winning branch to users quicker than trains | Same as Rollout Today |
| Do No Harm Monitoring Platform Experiment | Same as experiment | Same as experiment | Same as experiment. | No fixed timeframe means there is no automatic analysis trigger. | Not automated. Requires OpMon configured or team using their own custom dashboards. | Code changing/fixes landing during. Risk mitigation for quick major breaking stability / perf changes. Looking at adding search metrics as well. | N Same as experiment |
Multiple language Client integrations#
- Android (Kotlin)
- iOS (Swift)
- Firefox Desktop Frontend (JS)
- Firefox Platform (C++)
Client-side functionality#
- Consumer opt-out
- Integration with console preview testing
- View experiments (about:studies)
Current client integrations#
- Firefox Desktop
- Frontend via JS SDK
- Platform
- Windows installer
- Firefox Mobile via Nimbus-SDK Rust Component
- Fenix
- Firefox iOS
- Focus Andriod
- Focus iOS
Requesting feature support#
If you aren't sure we have what you need, pop into #ask-experimenter with your questions or file an issue