Text with claim. [^10] [^11] [^12] More text continues here.

[^10]: 2018. [Chronic nicotinamide riboside supplementation...](). Published: 2018 | Updated: N/A

class PDFExtractor:
    def extract(self, pdf_path: str) -> Dict[str, Any]:
        """
        Extract all content from a PDF.

        Returns:
            Dict with:
            - metadata: PDF properties
            - pages: List[ExtractedPage] with text per page
            - full_text: Complete document text
            - structure: DocumentStructure (sections, TOC, refs location)
            - footnotes: Position-detected footnotes
            - references_text: Text from references section
            - tables: Extracted tables
            - is_image_based: Whether OCR would be needed
        """

def extract_latex_footnotes(self, text: str) -> List[Dict[str, Any]]:
    """
    Extract footnotes from LaTeX-generated PDFs where footnotes are
    embedded inline in the extracted text.

    LaTeX footnotes appear as: "1U.S. Food and Drug Administration..."
    where the number is immediately followed by the citation text.
    """
    # Pattern: number at start of line followed by capital letter
    # Examples:
    #   1U.S. Food and Drug Administration, "FDA Approves..."
    #   10Martens CR, et al., "Chronic nicotinamide..."

    match = re.match(r'^(\d{1,3})([A-Z][A-Za-z].*)', line)

def _clean_markdown_citations(self, text: str, citations: List[Dict]) -> str:
    """
    Clean up markdown text by:
    1. Converting inline citation refs (e.g., "text.10 11 12" -> "text. [^10] [^11] [^12] ")
    2. Removing footnote definition lines from body text
    """

    # Key challenges solved:
    # - Distinguish decimals ($1.5 billion) from citations (text.15)
    # - Handle citation clusters (10 11 12) vs single refs (15)
    # - Remove footnote definitions that appear inline in LaTeX PDFs
    # - Proper spacing around citations

def replace_end_of_sentence_refs(match):
    before = match.group(1) or ""  # Character before punct
    punct = match.group(2)  # . or , or ; or :
    numbers_str = match.group(3)
    after = match.string[match.end():match.end()+20]

    ref_nums = re.findall(r'\d+', numbers_str)

    # Check if this looks like a decimal number
    if before and before.isdigit() and punct == '.':
        # Check for monetary/unit context
        if re.match(r'\s*(billion|million|thousand|percent|%)', after, re.IGNORECASE):
            return match.group(0)  # It's a decimal with units

        # Single digit after period is likely decimal
        if len(ref_nums) == 1 and int(ref_nums[0]) < 10:
            return match.group(0)  # Keep as decimal

    # Convert to citation format with proper spacing
    valid_refs = [n for n in ref_nums if n in known_refs]
    if valid_refs:
        refs_formatted = ' '.join(f'[^{n}]' for n in valid_refs)
        return (before or '') + punct + ' ' + refs_formatted + ' '

class CitationCorrectorAgent:
    def process_pdf(self, pdf_path: str) -> Dict[str, Any]:
        """
        Process a PDF: scrape then correct citations.

        Steps:
        1. Run the scraper to extract citations
        2. Review each citation with Claude to fix parsing errors
        3. Identify missing citations from the document text
        4. Output corrected citations in our standard format
        """

# Citation format types
CitationFormat = Literal[
    "footnotes",      # Numbered footnotes at bottom of page
    "endnotes",       # Numbered endnotes at end of document
    "bibliography",   # Author-date style with bibliography
    "bracketed",      # [1], [2], [3] style inline
    "superscript",    # Superscript numbers
    "mixed",          # Multiple formats detected
    "unknown"
]

class ParsedCitation(TypedDict, total=False):
    """A citation extracted from a PDF and converted to our standard format."""
    original_ref: str           # Original reference marker
    original_text: str          # Full original citation text
    title: str                  # Parsed title
    author: Optional[str]
    date: Optional[str]         # YYYY-MM-DD format
    date_display: Optional[str] # Human-readable date
    url: Optional[str]
    our_format: str             # Citation in our [^n]: format
    confidence: float           # 0.0-1.0 parsing confidence

class ParsedPDFData(TypedDict, total=False):
    """Complete data from a parsed research PDF."""
    pdf_id: str
    full_markdown: str
    sections: List[ParsedPDFSection]
    citations: List[ParsedCitation]
    citations_markdown: str
    parsing_confidence: float

# Parse a single PDF
python cli/parse_research_pdf.py path/to/research.pdf

# Parse to specific output directory
python cli/parse_research_pdf.py report.pdf -o output/parsed/

# Parse multiple PDFs
python cli/parse_research_pdf.py data/research/*.pdf -o output/parsed/

# Skip table extraction (faster)
python cli/parse_research_pdf.py report.pdf --no-tables

# Scrape and correct citations
python cli/correct_pdf_citations.py path/to/research.pdf

# With custom output directory
python cli/correct_pdf_citations.py report.pdf -o output/corrected/

# In company data file (data/{Company}.json)
{
  "research_pdfs": [
    "data/research/McKinsey-Report.pdf",
    "data/research/Industry-Analysis.pdf"
  ]
}

# Parsed citations available for citation_enrichment agent

from src.scrapers.research_pdf import scrape_research_pdf

result = scrape_research_pdf("path/to/research.pdf")
if result["success"]:
    citations = result["data"]["citations"]
    markdown = result["data"]["full_markdown"]