| Term | Meaning |
|---|
| AFD | Approximate functional dependency: a mapping with some exceptions. Reported with a violation count and the g1 error. |
| basis | measured identifies a count or direct comparison; estimated identifies a sample or model estimate; inferred identifies an interpretation such as a column role. |
| budget | A preset (quick, standard, or deep) for sampling and search limits. Exact column summaries use all rows under every preset. |
| category | A label such as uniqueness, dependencies, or gaps; the optional type field specifies a subtype. |
| conditional rate | Among rows satisfying a rule’s conditions, the proportion that also satisfy its outcome. |
| coverage | Status, sample size, and limits for each profiling pass. |
| detector | A component that performs a check. For example, dates produces gap and interval findings. |
| diagnostic partition | The fixed one-fifth row subset used to recount conditional rules. Discovery may already have used these rows. |
| effective cardinality | The number of equally frequent values with the same Shannon entropy as the observed distribution; calculated as exp(entropy). |
| evidence | A finding’s method, evaluated rows, validation, and metrics. A finding can include results from several methods. |
| finding | An entry in report.findings, containing a text summary, the columns involved, and supporting measurements. |
| form | A dependency’s representation, such as value_mapping, affine, or piecewise_linear. |
| g1 | Fraction of evaluated rows that would need to change for a functional dependency to hold exactly. |
| groups | Findings connected through shared columns. Check individual mappings before deciding that columns are redundant. |
| holdout | Rows excluded from fitting and used to evaluate a model. The evidence records their count. |
| interestingness | Score used to sort findings. A larger score means higher reporting priority, not a greater probability of correctness. |
| lift | A rule’s conditional rate divided by its outcome’s baseline rate on the same rows. |
| ξ (xi) | Chatterjee’s rank-based dependence coefficient. The profiler computes both directions because ξ(X,Y) can differ from ξ(Y,X). |
| scope | The rows used for a measurement, including input size, evaluated count, and missing-value handling. |
| statement | What a finding asserts, identified by its category, type, subject, and specification. Candidates with matching normalized fields are combined. |
| support | For a rule, joint_count counts rows satisfying both its conditions and its outcome. joint_support divides that count by eligible_rows. |
| UCC | Unique column combination: columns whose joint values uniquely identify the evaluated rows. |
| validation | The check used to assess a result, such as direct comparison, cross-validation, or a holdout. It also records whether the check helped select the result. |