Classificatory Inference

Classificatory Inference assigns each item in your data — a ticket, a lead, a transaction, a piece of content — to a class or score on a schema you define. The verdict is the product: risk band, quality grade, fraud flag, moderation decision. Use it when you need a consistent, scalable judgement applied across a large population of records rather than a human reviewing each one.

Mechanism

Assigns each input item to a class or score on a defined output schema (risk band, quality grade, fraud/not-fraud, anomalous/normal, lead tier, moderation verdict) where the assigned classification IS the deliverable insight.

Why this is a primitive

Cannot be decomposed: per-item-classify-against-schema is a single inference operation. Considered splitting out anomaly detection — collapsed in: anomaly detection is structurally a binary classifier (normal vs anomalous) operating on a stream, and the only differences (which features, what threshold, what cadence) are configuration of the same primitive, not a new operation. Composition layer captures the buyer difference (anomaly alerting composes classificatory-inference + notice-what-worked telemetry trigger + share-it notification; fraud detection composes classificatory-inference into a transaction pipeline; lead scoring composes it onto CRM records). Splitting would create primitives that fail the atomicity test against each other.