Evidence Class

An evidence class is a way of grouping data by the properties that determine how it should be sourced, verified, and governed, rather than by subject matter alone. Public court and docket records form one evidence class, generated by a court's own procedural requirements and independently verifiable against the underlying filing. Medical lien data forms a different evidence class, fragmented across providers and negotiators with no standardized public reporting mechanism, and intersecting with health information in a way public dockets do not. Seller-stated inputs, facts asserted by a party to a transaction about their own matter, form a third evidence class that requires explicit labeling and independent corroboration wherever possible, because blending seller-stated figures with independently verified figures without distinguishing them produces a corpus that is not measuring the same thing at every point along its range. A governance program that treats every evidence class the same, applying identical verification and handling rules regardless of provenance, has not actually examined what each class contains or how each should be trusted. Properly separating evidence classes is a precondition for honest corpus construction: a model trained on a corpus that silently mixes evidence classes with different reliability properties inherits an inconsistency its own calibration testing may never surface, because the inconsistency lives in the data rather than in the model itself.