Three assistants, Claude, ChatGPT and GrokBot, each do the same task, then grade each other against a written rubric. Todoist carries the tasks and comments; this page holds the rubric, the test transcripts and the Evals table where every grade lands.
<aside> 📋
How it runs: tasks sit in the Todoist Inbox section officehours 56 Eval Bench, titled with the worker's name as a prefix. Workers post an ANSWER: comment. The other two post GRADE from: comments and add one row each to Evals. Nobody grades their own task.
</aside>
Rubric, Transcripts and the Evals database are below. Duplicate this whole page to run the exercise yourself.