Skip to content

Run Management

Jump to bottom
mcmahj45 edited this page Jul 24, 2026 · 5 revisions

Run Management

A competition run is created from a competition template. Once created, it can be started and managed from the Evaluation Runs view in the DRES frontend.

Starting a Run

From the Evaluation Runs list, click the play button next to a run to open the admin view. From here the operator can:

  • Start and end the overall run
  • Select which task to run next
  • Start and end individual tasks
  • Increase or decrease the time remaining on the current task
  • Download the current scores as CSV, or the full run as JSON, via buttons in the admin toolbar

Runs created from an ASYNCHRONOUS template have slightly different admin-view gating around starting/ending tasks than SYNCHRONOUS runs, since teams may be on different tasks at the same time.

The Run Viewer

The run viewer is the main display shown to participants and observers during a competition. It shows:

  • The current task's query hints (image, video clip, or text) as they are revealed over time
  • The scoreboard, updated live as submissions arrive
  • Recent submissions for the current task

The viewer updates without a manual page refresh: it currently polls the server every second for state changes. (A WebSocket-based push update is in development on the pr/reintroduce-websockets-2 branch and will replace polling once merged.)

Scoreboard

The scoreboard displays team scores and updates live during a task. There are two views:

  • Task group view — shows scores for the current task group only
  • Competition overview — shows a stacked breakdown of scores across all task groups

Score History

The score history page (/evaluation/scores/{runId}) is intended to show a time-series graph of how each team's score developed over the course of the competition, with a dropdown to switch between different score groupings. Note: on the current branch, the frontend code that populates this graph's data series is disabled, so the page does not currently render score history — this is a known gap, not a configuration issue.

Judgement

For tasks with a JUDGEMENT target type, submissions are not automatically evaluated — they are queued for a human judge to review. See the Judgement page for details.

Audience Voting

For tasks with a VOTE target type, non-judge users can cast votes on submissions via the audience voting viewer, rather than a single judge deciding. See Judgement for details.

Clone this wiki locally