doczyai-pipelines/fieldExtraction/constants/regex_patterns.py

# Delimiter Patterns
PIPE_PATTERN = r"\|([^|]+)\|"
BACKTICK_PATTERN = r"`([^`]+)`"
TRIPLE_BACKTICK_PATTERN = r"```([^`]+)```"

# TIN and NPI Regex Patterns
# TIN_PATTERN = (
#     r"\b\d{9}\b|\b\d{2}[-.]\d{7}\b|\b\d{3}[-.]\d{2}[-.]\d{4}\b|\b\d{2}-\d{4}-\d{3}\b"
# )
# NPI_PATTERN = r"\b\d{10}\b"
TIN_PATTERN = r"\b(?:\d[\s\-.]*){9}\b"
FILE_NAME_TIN_PATTERN = r"(?<!\d)(?:\d{2}[-.]\d{7}|\d{9})(?!\d)"
NPI_PATTERN = r"\b(?:\d\s*){10}\b"


# OCR-Correctable Patterns
# Handles common OCR misinterpretations where letters are mistaken for digits.
# Common OCR substitutions: O/o→0, l/I→1, S/s→5, G/g→6, B/b→8

# OCR-Correctable TIN Pattern
# Strategy: Match 9-character sequences OR formatted sequences (with hyphens/dots)
# containing digits and OCR-prone letters, then apply corrections post-match.
#
# Examples that WILL be captured and corrected:
# - "12345678O" → "123456780" (unformatted, O at end)
# - "O23456789" → "023456789" (unformatted, O at start)
# - "1234O6789" → "123406789" (unformatted, O in middle)
# - "12-345678O" → "12-3456780" (formatted XX-XXXXXXX, O at end)
# - "O2-3456789" → "02-3456789" (formatted XX-XXXXXXX, O at start)
# - "12O-456789" → "120-456789" (formatted XX-XXXXXXX, O in first segment)
# - "123-4S-6789" → "123-45-6789" (formatted XXX-XX-XXXX, S in middle segment)
#
# Examples that will NOT match:
# - "12345678" (only 8 characters)
# - "1234567890" (10 characters)
# - "123ABC789" (non-OCR letters like A, C)
#
# Note: This pattern is intentionally permissive. Validation happens post-match
# by applying OCR_SUBSTITUTIONS and checking that result has exactly 9 digits.
TIN_OCR_PATTERN = r"\b[O0lI1S5G6B8gbo\d]{9}\b|\b[O0lI1S5G6B8gbo\d]{2}[-.][O0lI1S5G6B8gbo\d]{7}\b|\b[O0lI1S5G6B8gbo\d]{3}[-.][O0lI1S5G6B8gbo\d]{2}[-.][O0lI1S5G6B8gbo\d]{4}\b"

# OCR-Correctable NPI Pattern
# Strategy: Match 10-character sequences containing digits and OCR-prone letters,
# then apply corrections post-match to validate it becomes a proper 10-digit NPI.
#
# Examples that WILL be captured and corrected:
# - "123456789O" → "1234567890" (O at end)
# - "l234567890" → "1234567890" (l at start)
# - "12345O7890" → "1234507890" (O in middle)
# - "123456789l" → "1234567891" (l at end)
#
# Examples that will NOT match:
# - "123456789" (only 9 characters)
# - "12345678901" (11 characters)
# - "123ABC7890" (non-OCR letters like A, C)
#
# Note: NPIs are always unformatted 10-digit numbers, so no hyphen/dot variations.
# Validation happens post-match by applying OCR_SUBSTITUTIONS and checking
# that result is exactly 10 digits.
NPI_OCR_PATTERN = r"\b[O0lI1S5G6B8gbo\d]{10}\b"

# OCR Substitution Mapping
OCR_SUBSTITUTIONS = {
    "O": "0",  # Letter O to zero
    "o": "0",
    "l": "1",  # Letter l/I to one
    "I": "1",
    "S": "5",  # Letter S to five
    "s": "5",
    "G": "6",  # Letter G to six
    "g": "6",
    "B": "8",  # Letter B to eight
    "b": "8",
}