Labelled account numbers

ACCOUNT_NUMBER detects 6–17 contiguous ASCII digits immediately after account number, account no., acct number or acct no. (ASCII case insensitive). The value span excludes the label. Supported spacing, delimiters, boundaries and exclusions are defined in the rule contract. This is lexical coverage, not bank validation or universal account detection.

from datafog_core import scan_and_transform
result = scan_and_transform("bank account number: 00123456", {
    "scan": {"entities": ["ACCOUNT_NUMBER"]},
    "transform": {"default": {"strategy": "redact"}}
})
assert result.text == "bank account number: [ACCOUNT_NUMBER]"

const result = scanAndTransform("bank account number: 00123456", {
  scan: {entities: ["ACCOUNT_NUMBER"]},
  transform: {default: {strategy: "redact"}}
});
console.assert(result.text === "bank account number: [ACCOUNT_NUMBER]");

Rules run by default across locales. Bare numbers, generic account: labels, intervening prose, split numeric values and alphanumeric accounts are excluded. The masked-account contract also accepts account-labelled four-digit endings and bounded X/asterisk masks. For account ending in 1234, only 1234 is the value. For account number: XXXX1234, the whole XXXX1234 token is the value. A visible suffix is not a full account identity; workspace matching and recall use the existing normalized literal value and cannot recover hidden digits. Structured scanning uses the same text rule within each decoded string; field keys do not supply the label. Rust, Python, Node and browser WASM share Core semantics. Findings have no confidence score and use datafog-core/account-number provenance, original byte and codepoint ranges, and JavaScript UTF-16 ranges. Normal entity selection, allowlists and transformation strategies apply. Other detectors may overlap; select ACCOUNT_NUMBER when a transformation should use only that type. This partial rule preserves a configured model’s broader ACCOUNT_NUMBER coverage. A model session selected for this type still runs its model, combines rule/model findings and retains evidence from both for equal-span duplicates. Models can therefore still find unlabelled values that this rule cannot.

Get started

Core concepts

Guides

Reference

Project

Labelled account numbers