Twelve realistic asks from the emission-eval suites — six ops, six forms — each answered by GPT-5.6 cold: the catalog's prompt-pack as system prompt, the ask below as
the only user message, captured by the harness and replayed here unedited. This is the
product claim in running form: models emit these vocabularies without fine-tuning or retries.
the ask — the only user message the model saw
Your monitoring shows: api.acme.dev uptime over the last 24h is 99.2% which is healthy; p95 latency is 340ms and has risen 45ms in the last hour, which is concerning; and there is a scheduled database failover tonight at 02:00 UTC that may cause a brief connection blip. Render a status view for the on-call engineer.
auri