# What secondary NPI addresses add to an organization map

Build a set P from nonempty location_key values within each organization CSV. Build a set S by splitting secondary_practice_addresses on semicolons, trimming values and removing empty values. Count P, S, S minus P and P union S. The increase percentage is |S minus P| divided by |P| times 100. Reuse the retained address-key strings without applying a second normalization algorithm. Compare only within a legal-name profile.

These are address-key sets, not independently verified facilities. A secondary field can be incomplete or outdated. Zero additional keys means no additional values under this extraction and comparison rule; it does not prove a single-location business or complete coverage. String differences may survive normalization, and shared strings can conceal separate suites or departments. No key establishes current opening status or ownership.

Reproduce: python3 scripts/build_research_drafts.py organization-npi-secondary-addresses; then python3 scripts/build.py. Inputs and SHA256 hashes are in provenance.json.