Checked results
Compare observers inside a declared task.
The tables preserve access, provenance, and result status. They do not collapse unlike tasks into one headline number.
| Order | Observer | Status | Access | Metric |
|---|---|---|---|---|
| Loading checked results… | ||||
Order is not automatically an official rank. Reference bounds receive no position. A submitted result receives a rank only within a track that fixes task version, access boundary, budget, controller, and loss.
Read two numbers
Prediction quality and action quality answer different questions.
Prediction metrics ask whether an observer ranks or estimates held-out cases accurately.
Action metrics ask what happens after the fixed controller spends its limited intervention or audit budget on that ranking.
A method can improve one and lose on the other. ObserverBench reports both.