Explicit decisions, compared.

Compare JEV, OpenAI, and Microsoft-Decision-1 on the same support-ticket decisions: category, urgency, and human review.

Methodology: all evaluation calls have a 60-second timeout and no automatic retries. Request durations include provider and network time. JEV and Microsoft return typed probabilities; OpenAI generates an explanation. Agreement and three repeated trials measure consistency; accuracy requires labeled ground truth. Latency summaries include successful calls. Failed trials are excluded from agreement and stability. Providers may have different calibration.