matiasmolinas/dental-analysis/record-normalization

Record Normalization Skill (HISTORA integration layer)

Turn raw, fragmented dental and medical records into one structured record whose oral and systemic fields are co-present and time-aligned, so downstream relations are representable.

Normalization discipline

• Map every source field to the canonical schema keys used by

oral-systemic-analysis (demographics, shared_risk, medical_cv, periodontal). Do not invent keys.

• Preserve original units and add the canonical unit next to any converted value;

never silently rescale.

• Keep oral and systemic data co-present in the output (do not emit one domain

and drop the other) — cross-domain relations only surface when both are visible.

• Build a shared timeline: align periodontal events (probing, treatments,

radiographs) and medical events (labs, diagnoses, medication changes) by date so progression and temporal coincidence are readable.

Completeness discipline

• For each required-but-absent field, emit an explicit MISSING marker, not a

blank — a missing mediating datum (e.g. hs_crp) must be visible downstream.

• Never impute a patient-specific value. Missing data becomes a collection flag

handled by oral-systemic-analysis, never a guess.

• Distinguish "absent from source" from "explicitly negative/normal"; do not

collapse them.

Provenance discipline

• Tag each normalized field with its source record so traceability-audit can

verify downstream claims back to the input.

• Flag internal contradictions (e.g. conflicting smoking status across records)

rather than picking one silently.