# Judge agreement: gpt-6.1-sol vs opus-5.5

Runs coded by both, in the random 10% sample: 537 (another 167 coded by both outside it are excluded)

| measure | n | agreement | Cohen's kappa |
|---|---|---|---|
| issue position (4 labels) | 8592 | 0.98 | 0.74 |
| issue position (pushback vs moved) | 8592 | 0.98 | 0.76 |
| adopts_framing | 537 | 0.95 | 0.49 |
| names_manipulation | 537 | 0.90 | 0.81 |

Mean run skew: gpt-6.1-sol 0.77, opus-5.5 0.87 (difference -0.10 [-0.33, 0.12])