Citation kinds

kaos-citations recognizes ~60 typed citation kinds across four domains, returned as typed objects with a kind, raw, normalized, and span (you used extract_citations in extract legal citations). Each kind is its own Pydantic type, so you get structured fields specific to that citation family.

Domains

Legal — case law and primary/secondary legal sources: CaseCitation, CFRCitation, FederalRegisterCitation, FederalRuleCitation, ConstitutionCitation, ExecutiveActionCitation, CourtDocumentCitation, AgencyAdjudicationCitation, AgencyManualCitation, BarEthicsOpinionCitation, HereinafterCitation, IdCitation, and more (statutes, regulations, court documents).

Financial / regulatory — securities and banking regulators: FINRARuleCitation, FINRARegulatoryNoticeCitation, FINRADisciplinaryCitation, ExchangeRuleCitation, CFTCDocumentCitation, CFPBDocumentCitation, FDICDocumentCitation, FedReserveRegulationCitation, FedReserveLetterCitation, FFIECCallReportCitation, BaselFrameworkCitation, InternationalFinancialCitation.

Accounting / auditing — standards bodies: AICPACitation, ASCCitation (Accounting Standards Codification), ASUCitation, FASABCitation, GASBCitation, IFRSCitation, IAASBCitation, IESBACodeCitation, GovernmentAuditCitation, IRSGuidanceCitation.

Identifier / academic — stable identifiers and scholarship: DOICitation, ArXivCitation (opt-in), JournalCitation, ArchiveCitation, InternetCitation, ElectronicMediaCitation.

Enumerate them live

The authoritative list is exposed as an MCP resource and in code:

import kaos_citations as kc
print(kc.CITATION_KINDS_URI)   # the kaos-citations://kinds resource

Restrict extraction to specific kinds with extract_citations(text, kinds=[...]).

Note

span is the absolute (start, end) offset in the source text; normalized is the canonical form (e.g. 347 U.S. 483 (1954)). They serve different purposes — span for verification and linking, normalized for display and dedup.