feat: unify rewrite domain metadata into a single source

[memadi-nv]

Summary

Refactors rewrite-pipeline domain metadata into a single source of truth and updates the domain taxonomy.

Refactor

• Collapse three parallel structures (_DOMAIN_LIST, DOMAIN_SUPPLEMENT_MAP, DOMAIN_SUPPLEMENT_PRIVACY_MAP) into one DomainMetadata dataclass + one DOMAIN_METADATA tuple keyed by Domain.
• New _build_domain_index raises RuntimeError at import time on duplicate or missing entries — drift fails fast instead of surfacing later as a mid-pipeline KeyError.
• Rename rewrite_supplement → quality_supplement (the field has always been quality guidance, never rewrite-specific).
• privacy_supplement is str | None and defaults to quality_supplement via __post_init__, so domains without dedicated privacy guidance can omit it. Today only LEGAL has a distinct privacy supplement.

Taxonomy

24 prior Domain values → 21 new ones:

The Domain enum's serialized values have changed. Pipelines or downstream
consumers holding rows with prior values will fail validation. Migration:

Old value	New value
SECURITY_INFOSEC	SECURITY_INFOSEC
FINANCIAL	FINANCIAL
LEGAL	LEGAL
META_TEXT	META_TEXT
ENTERTAINMENT_MEDIA	ENTERTAINMENT_MEDIA
OTHER	OTHER
BIOGRAPHY	BIOGRAPHY_PROFILE
CLINICAL_EHR_MEDICAL	MEDICAL_CLINICAL
HR_PEOPLE_OPS	HR_EMPLOYMENT
MANAGEMENT_OPERATIONS	BUSINESS_OPERATIONS
NEWS_JOURNALISM	NEWS_PUBLIC_AFFAIRS
SCIENTIFIC_ACADEMIC	RESEARCH_SCIENTIFIC
TECHNICAL_ENGINEERING_SOFTWARE	TECHNICAL_SOFTWARE_ENGINEERING
EDUCATIONAL_PEDAGOGICAL	EDUCATION
FICTION_CREATIVE	CREATIVE_FICTION
ECONOMIC	ECONOMIC_ANALYSIS
POLICY_REGULATORY_COMPLIANCE	POLICY_REGULATORY
MARKETING_ADVERTISING, PRODUCT_REVIEW	MARKETING_COMMERCIAL
SOCIAL_CULTURAL_OPED, SOCIAL_MEDIA	SOCIAL_COMMENTARY
CHAT_EMAIL_CSAT	dropped (re-classify)
PROCEDURAL_INSTRUCTIONAL	dropped (re-classify)
TRANSCRIPTS_INTERVIEWS	dropped (re-classify)
None	INSURANCE
None	GOVERNMENT_PUBLIC_RECORDS
CHAT_EMAIL_CSAT	dropped
PROCEDURAL_INSTRUCTIONAL	dropped
TRANSCRIPTS_INTERVIEWS	dropped

Tests

• Refreshed existing tests for the new field name and renamed identifiers.
• Added two error-path tests for _build_domain_index covering duplicate entries and missing enum coverage.
• make test — 647/647 pass; no new typecheck diagnostics in edited files.

Related Issues

Closes #55

[greptile-apps]

Greptile Summary

This PR consolidates three parallel domain-metadata structures (_DOMAIN_LIST, DOMAIN_SUPPLEMENT_MAP, DOMAIN_SUPPLEMENT_PRIVACY_MAP) into a single DomainMetadata dataclass and DOMAIN_METADATA tuple, with import-time validation via _build_domain_index that catches drift between the enum and the metadata table immediately. The Domain enum is simultaneously retaxonomized from 24 to 21 members (renames, merges, two additions, three removals), making this a breaking change for any persisted domain-column values.

• Structural consolidation: DomainMetadata(domain, classification_description, quality_supplement, privacy_supplement) replaces three separate dicts; _build_domain_index raises RuntimeError at import time on duplicates or missing coverage, turning a formerly silent runtime KeyError into a loud startup failure.
• Enum rename/merge/drop: 14 members renamed, 2 merged into existing values, 3 removed, 2 added (INSURANCE, GOVERNMENT_PUBLIC_RECORDS); any pipeline row carrying an old string value (e.g. "BIOGRAPHY", "CHAT_EMAIL_CSAT") will fail Pydantic validation after deployment.
• Test coverage: Error-path tests for _build_domain_index are present but use a manual try/except/else pattern rather than pytest.raises.

Confidence Score: 4/5

Safe to merge for greenfield data; any existing pipeline data carrying old domain strings will break at validation time and needs a migration or compatibility layer before deploying.

The structural refactor and import-time validation are sound. The Domain enum rename is intentional and well-documented, but it is a hard break for persisted rows — there is no alias or migration layer, so deploying against existing data will produce Pydantic ValidationError failures in _enrich_domain and _enrich_domain_privacy for every row carrying an old string value.

src/anonymizer/engine/schemas/rewrite.py — verify that all upstream datasets have been migrated or that this is intentionally a clean-slate deployment before merging.

Important Files Changed

Filename	Overview
src/anonymizer/engine/schemas/rewrite.py	Domain enum renaming: 24 → 21 members, removing 3 (CHAT_EMAIL_CSAT, PROCEDURAL_INSTRUCTIONAL, TRANSCRIPTS_INTERVIEWS), adding INSURANCE and GOVERNMENT_PUBLIC_RECORDS. Breaking serialization change for any persisted rows (previously flagged in thread).
src/anonymizer/engine/rewrite/domain_classification.py	Three parallel data structures collapsed into DomainMetadata dataclass + DOMAIN_METADATA tuple, with import-time validation via _build_domain_index. Minor ambiguity in FINANCIAL classification description overlapping with ECONOMIC_ANALYSIS.
tests/engine/test_domain_classification.py	Tests updated to new field names; error-path tests for _build_domain_index added using try/except/else rather than the pytest.raises idiom.
tests/engine/test_rewrite_workflow.py	Mechanical rename of "BIOGRAPHY" → "BIOGRAPHY_PROFILE" throughout test fixtures; no logic changes.

_{Reviews (3): Last reviewed commit: "address feedback" | Re-trigger Greptile}

[asteier2026]

Looks good to me.

[lipikaramaswamy]

Looks good from an engineering perspective. I reviewed the refactor/taxonomy changes and ran the local test suite.

I didn't however run a live model-backed end-to-end rewrite pipeline locally, so my approval assumes you validated a full pipeline run with the new taxonomy.

[memadi-nv]

Looks good from an engineering perspective. I reviewed the refactor/taxonomy changes and ran the local test suite.

I didn't however run a live model-backed end-to-end rewrite pipeline locally, so my approval assumes you validated a full pipeline run with the new taxonomy.

Tested with a single record of the following datasets => returned DOMAIN in rewrite:

• NVIDIA_synthetic_biographies => BIOGRAPHY_PROFILE (Renamed domain from BIOGRAPHY to BIOGRAPHY_PROFILE in this branch)
• TAB_legal_sample25 => LEGAL (only domain were quality_supplement and privacy_supplement are different)
• Created a dataset with insurance context => INSURANCE (New DOMAIN)

All returned domains are as expected.