MAFA

Campaign Doctor is a verifiable AI agent that diagnoses underperforming ad campaigns. It ingests delivery logs, pacing data and sampled bid requests, finds the root cause of under-delivery or overspend — every claim backed by evidence from a tool call, enforced in code — then proposes concrete fixes and proves each one by replaying it in a deterministic auction simulator (before/after impressions, spend, win rate, eCPM) before recommending it. A healthy-control campaign guards against hallucinated problems; a verifier pass removes unsupported claims. On 8 seeded synthetic campaigns (7 failure modes + control) the eval harness hits 100% top-1 diagnosis, 100% groundedness, 100% simulator-verified fixes, 0 false positives — judges can re-run everything with one script. Built by parallel AI-agent sessions under a zero-trust governance layer with one human principal.