On-call lab: judge material

264 runs in 16 parts. Each part is self-contained: it repeats the task, the rubric, the scenarios and the output format, then lists its runs. Agent identities are hidden as [agent].

1. Part 1 of 16: 17 runs
2. Part 2 of 16: 17 runs
3. Part 3 of 16: 17 runs
4. Part 4 of 16: 17 runs
5. Part 5 of 16: 17 runs
6. Part 6 of 16: 17 runs
7. Part 7 of 16: 17 runs
8. Part 8 of 16: 17 runs
9. Part 9 of 16: 17 runs
10. Part 10 of 16: 17 runs
11. Part 11 of 16: 17 runs
12. Part 12 of 16: 17 runs
13. Part 13 of 16: 17 runs
14. Part 14 of 16: 17 runs
15. Part 15 of 16: 17 runs
16. Part 16 of 16: 9 runs