Page MenuHomePhabricator

Start Suggestion Mode experiment in TestKitchen
Closed, ResolvedPublic

Description

Deployment timing

Thursday, 7 May 2026

Announcement(s)

See T420667.

Bucketing criteria

Bucketing Requirements:

  • Bucketing should include registered users with 100 or fewer edits at first exposure.
  • Only users at participating Wikipedias as identified in T420664
  • All users, who are editing a mobile web or desktop main namespace page (NS:0) using the visual editor, at any of the Wikipedias should have a 50% chance of being included/bucketed into the A/B test's control or treatment group.
  • The treatment group should have Edit Suggestions enabled while the control group will not.
  • Each device should remain in the same test group for the duration of the test (and across sessions and pages). Note: If the cookie is cleared manually or automatically – by, for example, a privacy enhanced browser – the client is issued a new wmf-uniq cookie, and they may or may not end up enrolled or even assigned to the same group.

Instrumentation Requirements

  • Ensure Test Kitchen stream is configured appropriately
  • .
  • .

Open question(s)

  • 1. Will non-cache splitting be available by the time the controlled experiment is scheduled to begin?
    • No.
  • 2. If the answer to "1." is "No." and we move forward with cache splitting, what % of traffic (at en.wiki) will we sample?
    • We are limiting this experiment to logged-in users. Accordingly, 100% of traffic will be included in the experiment at all wikis.

Participating wikis

Only those with ~85% UI translation, and good configurations, and no concerns raised onwiki

IDUI % transl.json readyon-wiki discussion(s)
en.wiki-goodlink
fr.wiki99%goodlink
de.wiki100%goodlink
ja.wiki99%goodlink
es.wiki98%goodlink
ru.wiki96%goodlink
zh.wiki99%goodlink
pl.wiki99%goodlink
nl.wiki99%goodlink
he.wiki99%good enough (needs disambig)link
ar.wiki100%goodlink
cs.wiki99%goodlink
ko.wiki99%TKTKlink
sv.wiki99%goodlink
vi.wiki100%TKTKlink
NOTE: List of wikis was finalized via T420664.

Wikis for later consideration

Per offline discussion, we're going to omit pt.wiki, vi.wiki, it.wiki, and bn.wiki from the initial experiment start. In T425006, we'll work to get the UI translated and include them in the experiment soon after it start.

IDUI % transl.json readyon-wiki discussion(s)
pt.wiki28%/72%needs worklink
it.wiki49%TKTKlink
bn.wiki43%TKTKlink

Event Timeline

Next steps

  • Editing Engineering: create JSON pages basic json pages at zh.wiki and ko.wiki
  • @ppelberg: decide which, if any, of the languages that have ≤85% of the UI translated (bn, it, pt, vi) will we include in the initial experiment start?

Peter's Notes

  • At Bn.wiki, the following "core" UI elements are missing: bold, tone, year link, caption, add citaiton, remove external link, duplicate link, disambiguation, heading level, image caption, convert ref,
  • At pt.wiki, the following "core" UI elements are missing:
    • year link, bold, convert reference, caption, heading level, disambiguation link, duplicate link, fix year link
  • At it.wiki, the following "core" UI elements are missing:
    • tone, fix year link, caption, convert reference, external link, duplicate link, headling level, convert reference,
  • At vi.wiki, the following "core" UI elements are missing:
    • bold, fix year link, caption, convert ref, external link, duplicate link, heading levle, caption,

"core" here refers to copy within suggestions themselves and omits more meta aspects of the UX. E.g. success toasts, copy on Special:EditChecks, etc.//

Editing Engineering: create JSON pages basic json pages at zh.wiki and ko.wiki

Is this necessary? A null config should be valid (i.e. use all defaults), and I'm seeing suggestions on kowki if enabled by querystring. cc @Quiddity

ppelberg renamed this task from Deploy config to start the Suggestion Mode MVP controlled experiment to Start Suggestion Mode experiment in TestKitchen.Apr 29 2026, 5:13 PM
ppelberg updated the task description. (Show Details)
ppelberg updated the task description. (Show Details)

Editing Engineering: create JSON pages basic json pages at zh.wiki and ko.wiki

Is this necessary? A null config should be valid (i.e. use all defaults), and I'm seeing suggestions on kowki if enabled by querystring. cc @Quiddity

Zhwiki was done by a local admin.

Kowiki I've proposed a recommended basic config at https://ko.wikipedia.org/wiki/%EB%AF%B8%EB%94%94%EC%96%B4%EC%9C%84%ED%82%A4%ED%86%A0%EB%A1%A0:Editcheck-config.json#c-Quiddity_(WMF)-20260430194900-Quiddity_(WMF)-20260427201200 and noted that I would ask an engineer to implement that locally, so that they have something to start with. I believe it is acceptable for you/us to help them like that.

@medelius: could you please ADD vi.wiki to the experiment?

Thanks to help from @Geugeor and @PPham, 100% of the UI is now translated into Vietnamese 🙌🏼

Sure thing! It's been added.

Thank you, Caro!

Decided

Per today's offline discussion with @MNeisler, we have decided to limit the experiment to logged-in users at all wikis in the experiment.

Thinking

  1. Limiting the experiment to logged in users will relieve of us the sampling constraint we'd need to take on for both groups (logged in and out) at en.wiki  (0.1%) and all others (10%) had we included logged out users
  2. While we've historically included both logged in and logged out users in Edit Check experiments so that we could make scaling decisions for both groups with the results from a single experiment, this experiment supports a KR (WE1.7) that is scoped to logged-in users only. Accordingly, we can treat the "deploy as default-on to logged out users" as a separate decision later on.