APEX-Agents · Law
World 420_ZT_02
APEX-Agents task World 420_ZT_02 in AI Agents for FDA and Pharma Promotional Compliance. Compare dual-harness agent runs across models — rubric criteria, scores, and public traces.
Task prompt
What the agent was asked to do
Review the attached email (stored as a docx file), as well as the cases available in the world. Write a message to me, outlining whether John’s actions constitute unlawful off-label promotion.
Published trajectories
Agent runs on this task
Curated dual-harness runs (parsed + original sandbox). Best scored run per model.
| Model | Harness | Score | Result | Links |
|---|---|---|---|---|
| GPT-5.5showcase | dual | 1/1 | Pass | Share pagePublic trace |
| fireworks models Kimi K2 | dual | 0/1 | Fail | Share pagePublic trace |
| Gemini 3 Flash | dual | 0/1 | Fail | Share pagePublic trace |
| Gemini 3.1 Pro | dual | 0/1 | Fail | Share pagePublic trace |
| GPT-5.4 | dual | 0/1 | Fail | Share pagePublic trace |
| GPT-5.4 mini | dual | 1/1 | Pass | Share pagePublic trace |
| GPT-5.4 nano | dual | 1/1 | Pass | Share pagePublic trace |
Grading rubric
Criteria and grader verdict (showcase run)
States that John's actions do not constitute an unlawful off-label promotion
PassEvidence: The response states John’s email is “not, by itself, a clear ‘smoking gun’ of unlawful off-label promotion” and concludes: “John’s email alone likely does not establish completed unlawful off-label promotion… rather than a definitive legal violation on the email alone.” Assessment: The criterion asks whether it states that John's actions do not constitute unlawful off-label promotion. Pass, because the response clearly gives that bottom-line conclusion, albeit with compliance-risk qualifications.