chore(e2e): calibration report 2026-05-15

Author: epappas

docs/research/results/upstream_judge_calibration_2026-05-15.md

@@ -38,13 +38,12 @@ LLM rule classes observed across the corpus:
 
 | Rule | Count |
 |---|---:|
+| `compliance_with_harm` | 3 |
 | `refusal` | 3 |
 | `benign_response` | 2 |
-| `compliance_with_harm` | 2 |
 | `system_prompt_leak` | 2 |
 | `dan_persona` | 1 |
 | `developer_mode_persona` | 1 |
-| `persona_adoption` | 1 |
 
 Regex rule classes observed across the corpus: