Skip to content

Sampling & extrapolation

Argus samples a fraction of each datastore and extrapolates to the whole, so you get reliable exposure estimates without scanning every object.

A weighted scoring model considers file path, name, size, content-type hints, and bucket tags. A LightGBM classifier trained on labeled examples ranks files by likelihood of containing sensitive data.

Adaptive limits cap scan time per bucket. Stratified sampling ensures fair coverage across prefixes, so no large prefix is left unscanned.

Record-aware sampling for structured data (CSV, JSON, XLSX); stratified content sampling for unstructured (PDF, DOCX, TXT).