Smart search for tagging: prioritize relevant users in @ mentions

[SylvainChevalier]

Closes #4280

Summary

• Add optional post_id param to /users/ search endpoint to prioritize relevant users (commenters, authors, project members) in @ mention suggestions
• Thread postId through frontend mentions context to API call
• Add backend tests for relevance-based user search

Test plan

• Run python -m pytest tests/unit/test_users/test_search.py -v
• Test @ mentions in comment editor on a post with known commenters
• Verify that commenters appear first when searching by username
• Verify existing @ mention behavior still works without post context

Generated with Claude Code

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Post copyable unit tests in a comment
• Commit unit tests in branch claude/issue-4280-smart-search-for-tagging

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[SylvainChevalier]

@claude keep the requirement of 3 characters before suggestions.

[claude]

Claude finished @SylvainChevalier's task in 54s —— View job

---

Verifying 3-character minimum for @ mention suggestions

• Gather context: review PR changes related to character threshold
• Identify where the threshold is set
• Verify the 3-character minimum is preserved

Result

No changes needed — the 3-character minimum requirement is already preserved in this PR at both layers:

1. Frontend (front_end/src/components/markdown_editor/plugins/mentions/utils.ts:33):

   if (query.length < 3) {
     return clientSearch(query, fallbackMentions);
   }

   This is unchanged from the original code. When fewer than 3 characters are typed, only local fallback results are shown (no API call).

2. Backend (users/serializers.py:206):

   search = serializers.CharField(required=False, min_length=3)

   The API endpoint enforces min_length=3 on the search parameter, so even if the frontend were to send a shorter query, the backend would reject it.

The smart search changes (passing postId, relevance sorting) only affect the API call that fires after 3+ characters are typed. No modifications required.

--- |

[github-actions]

🚀 Preview Environment

Your preview environment is ready!

Resource	Details
🌐 Preview URL	https://metaculus-pr-4327-claude-issue-4280-smart-search-preview.mtcl.cc
📦 Docker Image	ghcr.io/metaculus/metaculus:claude-issue-4280-smart-search-for-tagging-2b72ee7
🗄️ PostgreSQL	NeonDB branch preview/pr-4327-claude-issue-4280-smart-search
⚡ Redis	Fly Redis mtc-redis-pr-4327-claude-issue-4280-smart-search

Details

• Commit: 6a331fbefd1fb904cc97fd1a01c507984282d6ed
• Branch: claude/issue-4280-smart-search-for-tagging
• Fly App: metaculus-pr-4327-claude-issue-4280-smart-search

---

ℹ️ Preview Environment Info

Isolation:

• PostgreSQL and Redis are fully isolated from production
• Each PR gets its own database branch and Redis instance
• Changes pushed to this PR will trigger a new deployment

Limitations:

• Background workers and cron jobs are not deployed in preview environments
• If you need to test background jobs, use Heroku staging environments

Cleanup:

• This preview will be automatically destroyed when the PR is closed

[SylvainChevalier]

@claude what does the @ member tag do?

[claude]

Claude finished @SylvainChevalier's task in 35s —— View job

---

What does the @ member tag do?

• Gather context on the @ mention feature
• Provide explanation

The @ mention feature is a user tagging system inside the comment editor on Metaculus. When you're writing a comment and type @ followed by a username, it triggers a typeahead/autocomplete dropdown that lets you tag other users.

Here's how it works step-by-step:

1. You type @ in the comment editor — this activates the mentions plugin (BeautifulMentionsPlugin from lexical-beautiful-mentions).

2. With fewer than 3 characters typed, only local/fallback results are shown (no API call). These are pre-loaded defaults like recent commenters on the current post.

3. Once you type 3+ characters, the frontend fires an API call to GET /users/?search=<query> to search for matching usernames.

4. The dropdown shows matching users — you can select one to insert an @username mention into your comment.

5. Group mentions are also available: @moderators, @admins, @members, and (for curators/admins) @predictors.

What this PR adds — the "smart search" enhancement:

Previously, the search just matched usernames alphabetically. This PR adds a post_id parameter so the backend can prioritize relevant users for the post you're commenting on. The priority order is:

Priority	User type
1st	Users who have commented on this post
2nd	The post author and coauthors
3rd	Users with permissions on the post's project
4th	All other matching users

This is implemented in users/services/common.py:24-117 via Django ORM annotations (is_commenter, is_author, has_permission) that drive the sort order. The postId is threaded from the CommentFeed component → context provider → mentions plugin → API call.

---

View job |

[SylvainChevalier]

@claude I know that, I'm asking who receives notifications when @ members is used.

[claude]

Claude finished @SylvainChevalier's task in 1m 36s —— View job

---

Who receives notifications for @members?

• Gather context: find group mention and notification code
• Determine who receives notifications for @members
• Provide answer

Result

@members is shown in the frontend dropdown but is NOT actually implemented as a group mention in the backend. It will not notify anyone as a group.

Here's what happens when you type @members in a comment:

1. The frontend allows selecting @members from the mention dropdown (front_end/src/components/markdown_editor/plugins/mentions/utils.ts:13-17)
2. When the comment is saved, the backend processes mentions in comments/utils.py:35-64
3. The backend only recognizes these group mentions:
   • @admins → notifies all users with ADMIN permission on the post's project
   • @moderators (or @curators) → notifies all users with CURATOR permission on the post's project
   • @predictors → notifies all users who have forecasted on the question (only if the comment author is a curator or admin)
4. @members is not handled — it falls through to the username lookup fallback (comments/utils.py:67), where it tries to find a user literally named "members". If no such user exists, nobody is notified.

In short: @members currently notifies nobody (unless there's a user with the username "members"). It's a frontend option without a backend implementation.

The relevant code paths:

• Group mention extraction: comments/utils.py:15-69
• Notification dispatch: comments/services/notifications.py:17-28
• Email sending: notifications/services.py:720-752

---

View job

[SylvainChevalier]

@claude ok thanks. Two more refinements then:

1. Remove the @ members tag from the front end.
2. When making the list of all other users, only keep users who're active, and have a non-deleted comment in the last year.

[claude]

Claude finished @SylvainChevalier's task in 3m 51s —— View job

---

Implementing refinements to smart search

• Gather context: read relevant files
• Remove @members tag from frontend
• Filter "all other users" to active users with a non-deleted comment in the last year
• Update tests for new filtering behavior
• Commit and push changes

Summary

Both refinements have been implemented in commit 7848daf:

1. Removed @members from frontend (utils.ts:13-16)

• Removed { value: "members" } from the usersGroupMentions array — it had no backend implementation and notified nobody.

2. Filtered non-priority users to recently active only (users/services/common.py:43-53)

• Added a recently_active_user_ids subquery that finds users with at least one non-deleted comment (is_soft_deleted=False) created in the last 365 days.
• When post_id is provided with search: non-priority users (those who aren't commenters, authors, or project members) must be in the recently-active set to appear.
• When searching without post_id: all results are filtered to recently-active users only.
• Priority users (commenters on the post, authors/coauthors, project permission holders) still always appear regardless of recent activity.

3. Updated tests — added 3 new test cases:

• test_inactive_users_excluded_from_search — users without recent comments don't appear in non-priority results
• test_old_comments_dont_count_as_recent — comments older than a year don't qualify
• test_deleted_comments_dont_count_as_recent — soft-deleted comments don't qualify

Existing tests updated to give "other" users a recent comment via _make_recently_active() helper so they pass the new filter.
·

[SylvainChevalier]

@claude functionality looks good, now fix the issues from the unit tests.

[claude]

Claude finished @SylvainChevalier's task in 22m 52s —— View job

---

Fixing unit test issues

• Read CI logs and identify failures
• Fix Prettier formatting in utils.ts
• Fix Flake8 linting errors in test_search.py (unused imports, unused variable)
• Fix Black formatting in test_search.py and users/services/common.py
• Commit and push fixes
• Verify CI passes

Result

All CI checks are now passing. Fixed in 2 commits:

5308720 — Prettier, Flake8, and initial Black fixes:

• utils.ts: Collapsed 2-element array to single line (Prettier)
• test_search.py: Removed unused imports (pytest, User), fixed unused variable inactive_user, added trailing commas on multi-line factory_project() calls
• common.py: Collapsed author_ids.extend() to single line

2b72ee7 — Black formatting cleanup:

• Collapsed function signatures that fit within 88-char limit to single lines
• Collapsed anon_client.get() and assert expressions to single lines where they fit
• Converted long docstring to multi-line format
  ·