Experimentation Framework Roadmap
以下を希望される場合:
- コードの貢献を提出する
- バグの報告または修正
- 機能や改善の提案
- ドキュメントへの貢献
これらのページの英語版のガイドラインに従ってください。
このページの英語版にアクセスしてください。
We’re actively investing in improving the experimentation platform based on feedback from teams running experiments at scale. The full roadmap is tracked in Improve Growth experiment process and velocity.
Planned improvements
Sticky candidate assignments across exclusion boundaries
Currently, control assignments are cached and remain sticky, but candidate assignments can be overridden by exclusion rules in subsequent experiment blocks. We’re working on allowing cached candidate assignments to take precedence over exclusion logic, simplifying multi-step experiment flows - especially in registration and onboarding scenarios where context is built up progressively.
Related: gitlab-experiment#91
Forced variant assignment for testing and validation
Engineers need a way to force themselves into specific experiment variants during UAT and staging validation - including for anonymous entry points and backend-only experiments where there’s no clear UI parameter to pass. We’re exploring approaches through segmentation rules and operational tooling.
Related: gitlab#579133
Improved event validation and observability
Verifying that experiment tracking events are structured correctly and arriving in analytics pipelines is currently a late-stage, manual process. We’re working on shifting event validation left in the development cycle and providing near-realtime observability for staging and production environments.
Related: gitlab#579150, gitlab#579137
Graceful experiment transitions
When experiments conclude and are either promoted or reverted, users can experience a jarring shift in their experience. We’re improving the cleanup process to account for these user experience transitions and provide guidance on handling them smoothly.
Related: gitlab#579148
Goals
- Reduce time from experiment design to implementation
- Fewer tracking-related issues during rollout
- Improved developer experience across the experiment lifecycle