Skip to main content

Catalog Classify

Category: Security
Node id: estate-classify
Maturity: beta
Risk tier: C

PROPOSES the sensitivity (PHI, PII, special category, credentials) and clinical meaning of each column from its name, its table's name and its profile. Proposals only — never confirms, and absence of a finding is not evidence of absence.

What it does​

  • Spanish and Catalan clinical vocabulary (NHC, CIP, APELLIDO1, FEC_NAC, DIAGNOSTICO)
  • Table context promotes ambiguous column names
  • Shape-based detection for columns whose names say nothing
  • Every proposal carries confidence, origin and evidence

Use cases​

  • Find which columns of a hospital database hold personal or clinical data
  • Propose semantic meaning to seed a FHIR mapping
  • Produce a sensitivity inventory for a DPO

Runtime​

  • Performance tier: fast

Input ports​

NameTypeRequiredDescription
assetsjsonNoAssets with fields, as emitted by introspection
profilesjsonNoColumn profiles, which sharpen the rules and enable shape-based detection

Output ports​

NameTypeRequiredDescription
proposalsjsonNoPer column: sensitivity, semantic meaning, confidence, origin and the evidence that fired
unresolvedjsonNoColumns no rule matched. Reported explicitly rather than defaulted to harmless
proposal_countnumberNoColumns with at least one proposal

Settings​

Basic​

SettingNameTypeRequiredDefaultAllowed valuesDescription
Table NametablestringNoTable the columns belong to. Supplies the context that promotes ambiguous names: NOTAS on a warehouse table is nothing, NOTAS on PACIENTE is clinical free text.
Minimum Confidencemin_confidencenumberNo0Drop proposals below this confidence. Raising it trades recall for reviewer time; for PHI, prefer more review over fewer findings.

Errors​

CodeDescription
no_columnsNeither assets nor profiles carried any column to classify

Runtime notes​

  • PROPOSALS ONLY. Nothing is confirmed; promotion to confirmed is a separate permissioned action, never this node.
  • Absence of a sensitivity finding is NOT evidence of absence of sensitivity — the rules are conservative and under-report by design.
  • Set 'table' whenever you can: it is what lets an ambiguous column name be judged in context, and it is the highest-value rule family in a hospital schema.
  • Passing profiles enables shape rules, which is how a column called DATO3 holding 8 digits and a letter gets proposed as a national identifier.
  • Columns no rule matched are reported as unresolved rather than as harmless: 'no rule matched' and 'checked and found safe' are different claims.
  • Deterministic and offline — no model is called, so the same input always yields the same proposals.

See Build any workflow for how to put nodes together and Ports, types and data flow for how port types connect.

Something unclear? Send documentation feedback by email.