Skip to content

Restore full server-side filtering on the Runs list page #1138

Closed

Description

Original author: Cedric Vidal (@cedricvidal)

Summary

Filtering on the Runs list page is currently only partially evaluated server-side. Several filter dimensions are applied client-side over the current paginated page only, so under server-side pagination and grouping the results are incomplete — a filter can show few or zero matches even when matching runs exist on other pages.

Full server-side filtering for the Runs list existed before the UX v2 refactor. This issue tracks restoring it.

Background (prior art)

  • Move runs grouping server side #568 — Move runs grouping server side (original issue)
  • feat: move runs grouping and filtering server-side #570 — feat: move runs grouping and filtering server-side (merged 2026-04-12, closed Move runs grouping server side #568). Moved grouping, filtering, and aggregation into the API: GET /api/v1/requests gained status, outcome, taskPromptId (and related) query params backed by a MongoDB $match, added migration 009 indexes for filtered/grouped queries, and removed the client-side .filter() for status/outcome/task entirely.
  • UX v2 #945 — UX v2 (merged 2026-06-01). Reworked the Runs list into the new master/detail layout. In the new apps/portal/src/pages/RunsList.tsx, filtering is now performed client-side via Array.filter() over the fetched page for most dimensions.

This is a factual record of where the behavior changed, not an attribution of fault.

Current behavior — apps/portal/src/pages/RunsList.tsx

Sent server-side, but only when exactly one value is selected and it isn't the (Unknown) sentinel (singleServerValue()):

  • worker, status, outcome, profile, taskPromptId, submissionId, criteria, turns, maxIterations

Evaluated client-side over the current page only (after fetch, in the filteredRuns memo):

  • model, OS/platform, priority, agent version, date range, free-text search
  • any multi-value selection of worker / status / outcome / profile
  • any (Unknown) sentinel selection

Because the client only holds one page (limit 10–100), these filters operate on an incomplete dataset. Client-side sorting has the same single-page limitation (tracked separately in #739).

Desired behavior

  • All Runs list filters evaluated server-side so they compose correctly with server-side pagination and grouping.
  • Multi-value selections supported server-side (e.g. status in [...]).
  • Server-side support for the dimensions that currently lack it: model, OS/platform, priority, agent version, and date range (date/time range is also tracked by It should be possible to search for runs by date and time range #907).
  • Free-text search behavior made consistent (server-side, or clearly scoped).
  • Maintain CLI ↔ Portal parity for any new filter params.

Pointers

  • API handler: apps/api/src/routes/requests/index.ts (filter construction ~L1050) and ListRequestsQuerySchema in packages/shared/src/schemas/request.ts. The API already accepts several of these params; the gaps are mainly (a) multi-value support, (b) the missing dimensions (model / OS / priority / version), and (c) the portal wiring them to the API instead of filtering locally.
  • Add indexes for any new filterable fields (see migration 009 / docs/architecture/database.md).

Related

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    area: portalScope web portal, pages, and user-facing components.

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions