Labelled street addresses

STREET_ADDRESS detects the bounded number-first English street-line format immediately after street address:, shipping address: or billing address:. Labels and values stay on one line. The value includes a supported direction and single trailing unit, and excludes city, region, postcode and country. Exact suffixes, bounds and exclusions are in the approved contract. This is lexical coverage, not universal postal-address validation.

from datafog_core import scan_and_transform
result = scan_and_transform("Street address: 12 Willow Road, Unit 4B, Anytown", {
    "scan": {"entities": ["STREET_ADDRESS"]},
    "transform": {"default": {"strategy": "redact"}}
})
assert result.text == "Street address: [STREET_ADDRESS], Anytown"

const result = scanAndTransform("Street address: 12 Willow Road, Anytown", {
  scan: {entities: ["STREET_ADDRESS"]},
  transform: {default: {strategy: "redact"}}
});
console.assert(result.text === "Street address: [STREET_ADDRESS], Anytown");

The rule runs by default across locales. Unlabelled addresses, bare address: fields, number-last formats, named premises, PO boxes, intersections and leading units are outside this baseline. Recognized malformed or additional unit continuations are rejected rather than silently reducing the finding to the street prefix. The contract lists the recognized markers; other comma components terminate the street span. Models and user review remain necessary for unknown unit vocabulary and broader coverage. Structured scanning applies the same rule within each decoded string; field keys do not supply context. Rust, Python, Node and browser WASM share the Core rule. Findings expose original byte/codepoint ranges, JavaScript UTF-16 ranges, no confidence score and datafog-core/street-address provenance. Normal selection, allowlists and transformation strategies apply; other detectors may overlap. This partial rule retains configured native/custom model STREET_ADDRESS findings. Small’s full ADDRESS stays a distinct entity with its existing model coverage. Equal-span, equal-type duplicates retain both sources of evidence; existing Core transformation selection handles different spans. No model full address is renamed or truncated to a street line.

Get started

Core concepts

Guides

Reference

Project

Labelled street addresses