How Litmus works
The chain, end to end
Every KPI you submit goes through a fixed chain. Each stage produces a concrete object that the next stage consumes. If a stage fails, the next stages know why.
Interpret
Your sentence (‘Reduce churn below 5%’) becomes a derived requirement: the metric, the underlying concepts, the formula, the target, and Litmus’s confidence that it understood you.
Match
Each concept (‘customers_lost_in_month’) is matched to a real column in your schema. A deterministic pass handles the obvious ones; an LLM handles the leftovers.
Profile
Mapped columns get profiled: completeness, validity, uniqueness, freshness. Each rolls into a single quality score per field.
Evaluate
A KPI is possible only if every concept is mapped AND every mapped field is good enough. Otherwise: degraded (data isn’t trustworthy), blocked (a concept is missing), or not measurable (no source has this data at all).
Report
Two headline numbers, schema coverage and quality-adjusted, plus a blocker list ordered by how many KPIs each blocker is killing.