Each scenario documents one operator question shape end to end: the verbatim questions, the nuance they target, and what Schema.ai does with them — grounded in claims, topics, and eval fixtures. Failures are recorded as honestly as wins: known failing is a first-class state.
2 scenarios · 1 works today · 1 known failing · 0 aspirational
prometheus_client renames samples at exposition: counters gain _total and histograms emit _bucket/_count/_sum series, so the names operators see in scrapes and PromQL differ from the declared registration names. Before…
scoperetrievalvLLM v0.23.03 question variants
The vLLM design docs record prefix-cache activity as two token-denominated counters but state the metric of interest is the derived hit rate. The rationale section never names the real metrics (its PromQL example uses p…
retrievalhonestyvLLM v0.23.03 question variants
Schema.ai