One design goal here was to make “knowledge work” verifiable in the same way code is. The rubric/verify loop was our attempt to give agents a signal beyond “sounds good,” especially for research or strategy tasks where correctness isn’t binary. Curious how others here handle verification for non-code agent workflows.
One design goal here was to make “knowledge work” verifiable in the same way code is. The rubric/verify loop was our attempt to give agents a signal beyond “sounds good,” especially for research or strategy tasks where correctness isn’t binary. Curious how others here handle verification for non-code agent workflows.