Detect Employer Identification Numbers (EIN)

[jzonthemtn]

Implements the engine side of philterd/phisql#33 (RFC: add an EIN filter) for the Python port: a new ein identifier for the U.S. Employer Identification Number (federal tax ID, format NN-NNNNNNN).

Blocked on philterd/phisql#33. Do not start until that RFC is accepted and the ein filter lands in redaction-policy schema 1.2.0 (an additive edit in place of the current schema version).

Behavior

• Detect EINs in text. The canonical form is NN-NNNNNNN (two digits, hyphen, seven digits), matched at word boundaries.
• Bare nine-digit runs are ambiguous with SSN and other identifiers. Detect the hyphenated form reliably; participate in span disambiguation against SSN rather than blindly claiming every NNNNNNNNN.
• Optional onlyValidPrefixes (default false): when true, keep only matches whose two-digit prefix is one the IRS issues. The engine carries the set of valid IRS EIN prefixes; the default false matches any EIN-formatted value.
• Loads/compiles as an ein filter with an einFilterStrategies array of the standard strategies.

Implementation notes (mirror the SSN filter)

• phileas/filters/ssn_filter.py (SSNFilter) → add phileas/filters/ein_filter.py (EINFilter) with detection + onlyValidPrefixes.
• Register it in phileas/services/filter_service.py (add (EINFilter, "EIN") alongside (SSNFilter, "SSN")).
• Advance the schema the port pins/validates against to the 1.2.0 revision that adds ein.

Acceptance criteria

• A new EINFilter detects NN-NNNNNNN at word boundaries and is registered in filter_service.py.
• EIN participates in span disambiguation against SSN so bare nine-digit runs are not blindly claimed as EIN.
• onlyValidPrefixes (default false) restricts matches to IRS-issued two-digit prefixes when true; the engine carries the prefix set.
• A policy with identifiers.ein loads and redacts, and validates against redaction-policy schema 1.2.0.
• Tests cover detection, the hyphen-position SSN/EIN distinction, onlyValidPrefixes on and off, and a strategy applied to EIN spans.
• Documentation is updated for the new entity type and the onlyValidPrefixes option.

[jzonthemtn]

The PhiSQL side is merged (philterd/phisql#34, closing RFC philterd/phisql#33): the ein entity type and filterEin are now in redaction-policy schema 1.2.0. This unblocks the engine work here.

For reference, the merged contract:

• New ein identifier; a policy uses identifiers.ein with an einFilterStrategies array.
• Detect the U.S. Employer Identification Number, canonical form NN-NNNNNNN (two digits, hyphen, seven digits), at word boundaries.
• Bare nine-digit runs are ambiguous with SSN; detect the hyphenated form reliably and disambiguate against SSN rather than blindly claiming every NNNNNNNNN.
• Optional onlyValidPrefixes (default false): when true, keep only matches whose two-digit prefix is one the IRS issues; the engine carries the prefix set.