Support for experiment, beta, and generally available features

GitLab sometimes releases features as experimental or beta, and users can opt in and test the new experience. Some reasons for these kinds of feature releases include:

  • Validating the edge-cases of scale, support, and maintenance burden of features in their current form for every designed use case.
  • Features not complete enough to be considered an MVC, but added to the codebase as part of the development process.

Please note that some features may not be aligned to these recommendations if they were developed before the recommendations were in place, or if a team determined an alternative implementation approach was needed.

All other features are considered to be generally available (GA).

Experiment

Experimental features:

  • Are not ready for production use.
  • Have no support available. Issues regarding such features should be opened in the GitLab issue tracker.
  • Might be unstable.
  • Could be removed at any time.
  • Might have a risk of data loss.
  • Might have no documentation, or information limited to just GitLab issues or a blog.
  • Might not have a finalized user experience, and might only be accessible through quick actions or API requests.

Beta

Beta features:

  • Might not be ready for production use.
  • Are supported on a commercially-reasonable effort basis, but with the expectation that issues require extra time and assistance from development to troubleshoot.
  • Might be unstable.
  • Have configuration and dependencies that are unlikely to change.
  • Have features and functions that are unlikely to change. However, breaking changes can occur outside of major releases or with less notice than for generally available features.
  • Have a low risk of data loss.
  • Have a user experience that is complete or near completion.

Generally available (GA)

Generally available features:

  • Are ready for production use at any scale.
  • Are fully supported and documented.
  • Have a complete user experience aligned with GitLab design standards.

All features are in production

All experimental, beta, and generally available features are available on GitLab.com, so they are all considered to be “in production”.

GitLab experiment and beta development guidelines

Teams should release features as GA from the start unless there are strong reasons to release them as experimental or beta versions first.

Product development teams should refrain from making changes that they reasonably believe could create significant risks or friction for GitLab users or the platform, such as:

  • Risking damage or exfiltration of existing production data accessed by our users.
  • Destabilizing other parts of the application.
  • Introducing friction into high Monthly Active User (MAU) areas.

Experiment features

In addition to the experiment details for users, experiments should:

  • Offer a way to opt in with minimal friction. For example, needing to flip a feature flag is too much friction, but a group or project-level setting in the UI is not.
  • Link out to the GitLab Testing Agreement in the opt-in.
  • Have documentation that reflects that the feature is subject to the GitLab Testing Agreement.
  • Have UI that reflects the experiment status.
  • Have a feedback issue to engage with internal and external users.
  • Not be announced in a release post.
  • Be promoted in the user interface through discovery moments, if needed.

All experimental features that meet the review criteria must initiate Production Readiness Review and complete the experiment section in the readiness template.

Beta features

In addition to the beta details for users, beta features should:

  • Not be required or necessary for most features.
  • Have documentation that reflects the beta status.
  • Have UI that reflects the beta status.
  • Be behind a feature flag that is on by default.
  • Be behind a toggle that is off by default.
  • Be announced in a release post that reflects the beta status, if desired.
  • Be promoted in the user interface through discovery moments, if needed.

All beta features that meet the review criteria must complete all sections up to and including the beta section in the readiness template by following the Production Readiness Review process.

GA features

Generally available features that meet the review criteria must complete the Production Readiness Review and complete all sections up to and including the GA section in the readiness template.

Provide earlier access

Our mission is “everyone can contribute”, and that is only possible if people outside the company can try a feature. We get higher quality (more diverse) feedback if people from different organizations try something, so give users the ability to opt in to experimental features when there is enough value.

Where possible, release an experimental feature externally instead of only testing internally or waiting for the feature to be in a beta state. We’ve learned that keeping features internal-only for extended periods of time slows us down unnecessarily.

Experimental features are only shown when people/organizations opt in to experiments, so we are allowed to make mistakes here and literally experiment.

Experiment and beta exit criteria

To ensure the phases before general availability are as short as possible each phase of experiment, beta, and limited availability should include exit criteria. This encourages rapid iteration and reduces cycle time.

GitLab Product Managers must take the following into account when deciding what exit criteria to apply to their experimental and beta features:

  • Time: Define an end date at which point the feature will be generally availabile.
    • Consider setting a time-bound target metric that will define readiness for exit into GA. For example, X number of customers retained MoM over 6 months after launch of experiment, X% growth of free and paid users in three months since launch beta, or similar.
    • Be mindful of balancing time to market, user experience, and richness of experience. Some beta programs have lasted one milestone while others have lasted a couple of years.
  • Feedback: Define the minimum number of customers that have been onboarded and interviewed.
    • Consider also setting a time bound when using user feedback as an exit criteria for leaving a phase. If a given time period elapses and we can not solicit feedback from enough users, it is better to ship what we have and iterate on it as a GA at that point rather than maintain a pre-GA state.
  • Limited Feature Completion: Determine if there is functionality that should be completed before moving to general availability.
    • Be wary of including “just one more” feature. Iteration is easier and more effective with more feedback from more users, so getting to general availability is preferred.
  • System Performance metrics: Determine the criteria that the platform has shown before being ready for general availability. Examples include response times and successfully handling a specific number of requests per second.
  • Success criteria: Not all features may reach GA. It is OK to pivot if early feedback indicates that a different direction would provide more value or a better user experience. If open questions must be answered to decide if the feature is worth putting in the product, list and answer those.

For the exit criteria of AI features, in addition to the above, see the UX maturity requirements.