Running an Origin Trial

For the full context on origin trials, please see the explainer. This is the feature author guide for Blink contributors.

Here, we describe what is involved in running an origin trials experiment for a new browser feature. Most importantly, origin trials are integrated into the launch process for new web platform features. You should be following that overall process (maybe you ended up here from that page).

Should you run an origin trial?

Origin trials are intended to be used to ensure we design the best possible features by getting feedback from developers before the standard is finalized. They may also be used to prove developer interest in a feature proposal that is otherwise undesired due to an expected lack of interest. Typically, you should have a specific hypothesis that you wish to test by collecting data.

If you answer “yes” to any of the following questions, you should consider running an origin trial (see caveat in [1]).

• Is there disagreement about how well this API satisfies its intended use case?
• Are you unsure about what API shape will be the most ergonomic in real world scenarios?
• Is it hard to quantify performance gains without testing on real world sites?
• Is there a reason that this API needs to be deployed to real users, rather than behind a flag, for data to be meaningful?

[1] Origin trials should be run for a specific reason. These questions are guidance to identifying that reason. However, there is still debate about the right reasons, so the guidance may change. You can join the conversation in this doc.

If you're planning to run an origin trial please contact the origin trials team (OT team) to quickly talk over your feature and the reason for running the trial. You can email origin-trials-support@google.com.

How do origin trials work in Chrome?

The framework will enable features at runtime, on a per-execution-context basis (practically, this will be per-document or per-worker). Features are disabled by default, and only be enabled if a properly signed token, scoped to the origin that it is being presented on, and scoped to the specific feature name, is present in either:

• an HTTP Origin-Trial header in the server response,
• an HTML <META> tag in the document's head, or
• (for Dedicated Workers only) the HTTP response or document head of the parent document.

The logic for enabling includes a check of your runtime feature flag (even if the origin trials framework isn't being used). This means you can easily test your feature locally, even without any trial tokens. Origin trials are being enabled in documents (for both inline and external scripts), and in service, shared, and dedicated workers. (Note that for service workers and shared workers, HTTP headers are the only way to enable trials. Dedicated workers will also inherit any trials enabled by their parent document). If an experiment gets out of hand (way too popular to be an experiment anymore, for instance), we’ll be able to turn it off remotely, for all origins. Similarly, if there turns out to be major problems with the implementation of the framework itself, we’ll be able to turn it off completely, and disable all trials. (Hopefully we don’t have to do that, but we're still in the early stages of origin trials, and we’re being careful.)

Is your feature ready to be an origin trial?

Before running an origin trial experiment, your feature needs to be ready for both web developers and users. Your feature must satisfy the following:

• Have an explainer for your feature
  • There needs to be a description of the problems and use cases the feature is intended to address.
• Have a draft spec for your feature
  • Ideally this would be in the form of an actual spec -- or PR against an existing spec -- in the format expected by the eventual standards body (although the draft might be hosted elsewhere while discussions are ongoing), but a detailed explainer laying out all the details may suffice. The level of detail needs to be enough so that (a) developers participating in the origin trial know how it works, and (b) spec editors for the relevant eventual specifications can see exactly how it would affect that spec when added.
• Be approved by the internal Chrome launch review process
  • Users may be exposed to your feature without opting in, so the appropriate measures must be taken for privacy, security, etc.
• Have a way to remotely disable the feature
  • Origin trials provides infrastructure to disable a feature (or a specific origin), but this only applies to the exposure as an origin trial. That means, any interface(s) controlled by the trial will be disabled, but it will still be possible to enable the feature via its runtime flag. As well, all of the token validation/revocation happens in the renderer.
  • If the previous point is not sufficient for disabling the feature, you should implement a kill switch that allows your feature to be disabled remotely via Finch.
    • This can use the existing functionality in PermissionContextBase or base::FeatureList, or be a feature-specific implementation.
  • Consult the Chrome feature launch process for guidance for a feature flag (Googlers only). If you would launch your feature by default with a flag, then you should implement one for the origin trial.
• Have UMA metrics to track feature usage
  • You should record usage with UseCounter, as that can be automatically monitored by the origin trials infrastructure.
  • The feature must have a corresponding entry in the enum WebFeature.
  • For any JavaScript-exposed API, usage can be recorded easily via one of the [Measure] or [MeasureAs] IDL attributes.
  • If not exposed via IDL, the appropriate UseCounter::count*() method can be used directly from your feature implementation.
  • If it's not feasible to integrate with UseCounter (e.g. usage is best tracked outside a renderer, .etc), please contact the OT team.
• Have an established community for discussion of the feature
  • At a minimum, this should be a WICG group, Github repo, etc. Anywhere developers can find discussion or log issues about your feature.
  • The origin trials system will facilitate collecting feedback from web developers. However, the goal is to have web developers participate in the existing community around the feature.
• Prepared a blog post/article/landing page introducing the feature
  • There needs to be single link that will provide details about the feature.
  • Should make it clear how developers provide feedback/log issues for your feature.
  • This could be the README.md in your Github repo, or any other page of your choice.
  • Should include details about availability via origin trials.

What is the actual process to run an origin trial?

First review the Blink launch process. Please contact origin-trials-support@google.com with any questions about these steps. If you don't have access to any of the links below, the mailing list can help find someone to guide you. Running an origin trial requires the following:

• Make sure your feature is ready to run an origin trial experiment (see above).
• Review go/ChromeLaunchProcess and determine what launch approvals you require.
• Integrate with the origin trials framework (see below).
• Send an Intent to Experiment, via the ChromeStatus entry for your feature.
• After the intent is approved, request a trial for your feature at ChromeStatus by clicking the "Request Trial Creation" button in the Origin Trial section.
• Land the feature in Chrome prior to beta.
• Engage with external partners or large-scale developers for early testing.
  • Some issues may only found during large-scale use, so should be tested even before beta (if possible).
  • If feasible, this could include doing your own testing within the developer's environment (any of test/staging/production).
  • For an example, see crbug.com/709211.
• Publish a blog post on developers.google.com/web/updates about the feature when it lands beta.
  • The OT team will add your feature to the public developer console.
  • See below for more details.
• Update the feature's entry on chromestatus.com to set the status to "Origin trial".
• You can review the developer registrations for your feature (and renewals) by following the instructions in the feature author guide.
  • In particular, be aware of any registrations that expect >10,000 page views per day to be using the feature

Note that these steps are not meant to be sequential. For example, you can certainly start integrating your feature with origin trials prior to getting various launch approvals.

Adding your feature to the public developer console

The configuration of a trial in the developer console requires basic information, including some public landing page/documentation link (see above). In some cases, this may seem like a chicken-and-egg problem. For example, you may not want to publish a blog post until the feature is ready for developers. If the blog post has detailed information on joining the origin trial, it doesn't make sense to publish and have web developers unable to register. Fortunately, a trial can be configured in the developer console in advance, separately from making it available to the public.

Recommended process:

• Request setup of the trial with an interim documentation link (e.g. the README from the GitHub repo).
  • The OT team can configure the trial in the developer console (without making in public), and provide a permanent link to the trial detail page.
• Publish the blog post for the feature when the beta release is available.
  • Include instructions about the origin trial, and a link to either (1) the list of available trials or (2) the detail page for your trial.
• Notify the OT team with the final link to the blog post.
  • The link will be updated in the developer console and recorded in go/origin-trials-feature-pipeline.
• Notify the OT team when the trial is ready for registration.
  • OT team will activate the trial to make it available to the public.

How long do Origin Trials typically last?

Origin trials can run for up to 6 milestones (~24 weeks), and can be extended under certain conditions. See here for more details.

What is the process to extend an origin trial?

Origin trials are not intended as a mechanism for shipping a feature early without following the full launch process. This is one of the reasons that each trial has a predefined end date (rather than running indefinitely until the feature ships). That said, there are situations where it is beneficial to allow experimentation to continue beyond the planned end date of the trial.

There are two general cases where experimentation may continue beyond the planned end date:

1. Unexpected delays in releasing
   • With the 4 week Chrome release cycle, code may not land in the intended release, meaning it is not shipped until the subsequent release. Alternatively, the code may land in the release, but the Chrome stable rollout is delayed, meaning it not available/installed until much later than expected. When a trial typically runs for 12-16 weeks, such delays can significantly impact the availability of the experimental feature and the ability to collect sufficient data.
2. Feature changes, new areas experimentation, or reviving dormant features
   • As the origin trial progress, you may determine that a feature is not ready to be shipped, but do want to continue experimenting.
     • Examples include:
       • Feedback indicates that changes are needed to the feature (especially in API surface), but the changes would benefit from further feedback.
       • You uncover new hypotheses for experimentation, after proving or disproving the previous ones.
       • Returning to a dormant feature launch to start experimenting again after the previous trial has long ended.

Consult with the OT team to figure out if you're in a situation where it makes sense to continue experimenting. For unexpected delays (1), this generally means requesting an extension to the trial end date. For feature changes and such (2), this generally means starting a new origin trial, to follow the previous trial.

In order to be eligible for an extension, you must demonstrate substantial progress towards meeting the bar for shipping the feature. See here for more details.

How to setup an extension or continued experiment?

The process depends on the scenario for your feature, as described in the previous section. If desired, email origin-trials-support@google.com to consult on the appropriate approach.

• Scenario 1: Extending the end date of the existing trial.
  • Send an Intent to Extend Origin Trial, via the ChromeStatus entry for your feature.
  • Wait for 1 LGTM from at least one API owner (similar to the original Intent to Experiment)
  • Select the "Finalize Trial Extension" button that is now available in the Origin Trial section.
• Scenario 2: Continuing to experiment via a new trial.
  • This will officially be a new and separate trial, meaning a separate entry in the list of trials, etc.
  • Update the integration with the framework - the code must use a different trial name in Chromium (see integration below).
  • Add a new Origin Trials stage in the ChromeStatus feature entry and edit the fields as appropriate.
  • Request a new trial with the appropriate naming to distinguish the old and new trial at ChromeStatus by clicking the "Request Trial Creation" button in the Origin Trial section.
  • Click on the API Owners chip and then on Draft Intent to Experiment Email.
  • After the intent is approved, request a trial for your feature at ChromeStatus by clicking the "Request Trial Creation" button in the Origin Trial section.
  • Upon approval, the OT team will setup the new trial.

How to integrate your feature with the framework?

The integration instructions now live in the Chromium source repo: https://chromium.googlesource.com/chromium/src/+/HEAD/docs/origin_trials_integration.md

Roadmap

All of this may change, as we respond to your feedback about the framework itself. Please let us know how it works, and what's missing! To follow the most up-to-date progress and plans, filter in crbug.com for “component:Internals>OriginTrials”.