Skip to contents

Construct a QuoteProvenance from a single Anthropic citation

Source: R/quote_provenance.R
make_quote_from_citation.Rd

Bridges one Anthropic citation object (the shape produced by R/02_ai_providers.R::.anthropic_extract_citations) to a QuoteProvenance object with citation_source = "anthropic_citations_api". The constructed quote is in the "unverified" state; chain through verify_quote to run the verification ladder.

Usage

make_quote_from_citation(
  citation,
  documents,
  attributed_theme_id = NA_character_,
  attributed_code_id = NA_character_,
  ai_model = NA_character_,
  ai_call_id = NA_character_,
  ai_paraphrase = NA_character_,
  source_doc_type_default = "data_entry"
)

Arguments

citation

A single citation object from ai_complete()$citations. Must have type, document_index, and the type-specific span fields populated.

documents

The same documents list passed to ai_complete(). Each element must have $id and $text; an optional $type field overrides the default source_doc_type.

attributed_theme_id, attributed_code_id

Optional attribution metadata. The bridge cannot infer these from the citation alone – the caller pairs each citation with the code/theme it supports.

ai_model, ai_call_id

The model and request_id from the ai_complete() response that produced this citation. Stored on the QuoteProvenance for audit log linkage.

ai_paraphrase

Optional paraphrase, if the AI rephrased rather than directly quoted. Defaults to NA_character_ since Citations API returns verbatim slices.

source_doc_type_default

Default source_doc_type when the document doesn't specify $type. Defaults to "data_entry" matching pakhom's coding pipeline convention.

Value

A QuoteProvenance object (unverified state) with citation_source = "anthropic_citations_api".

Details

Supported citation types:

  • char_location (plain text source) – maps directly to start_char/end_char. The cited_text is stored as exact_text; the verification ladder confirms it matches source[start_char:end_char] byte-for-byte.

  • page_location (PDF source) – not yet supported. PDF inputs aren't part of pakhom's current data model. Errors with a clear message rather than silently producing a malformed quote.

  • content_block_location (custom_content source) – not yet supported for the same reason. The Anthropic path uses plain_text source exclusively; if a future caller switches to custom_content, the bridge needs a block-index-to-char-offset mapping (caller would supply per-document block boundaries).

Document lookup: citation$document_index is 0-indexed into the documents list passed to ai_complete(). The bridge converts to 1-indexed for R.