Results

Every timed run, asked against worked. Each mark is one agent on one task at one requested time. Choose one agent to list its runs.

Agents

View
Request
Reading the marks

Across is the time asked for and up is the time worked, both on log scales. A mark on the target line worked exactly as long as asked, and the pale band around it is within 5% of the time asked.

Capped marks were stopped by AgentTime, so their time is a lower bound. Runs asked for the same time are spread slightly sideways so they do not hide each other.

Timing error in this view

    1.00× is perfect. How it is computed

    Click any mark to pin a run, or focus the chart and use the arrow keys. Enter pins, Escape clears, T shows the table.

    Showing all 1,991 runs.

    Read before quoting: 2 notes
    1. 93 runs were stopped by AgentTime. Their times are lower bounds, drawn with a small cap. Details
    2. 259 runs have an ending that is not labelled yet. They are timed and counted like every other run. Details