Compare two unit lines as adjudication evidence

Compare two unit lines as adjudication evidence

Usage

pf_unit_evidence(a, b, street_a = NULL, street_b = NULL)

Arguments

a, b

Raw address line 2 values, the anchor's and the candidate's.

street_a, street_b

Optional street lines, passed through to pf_street_unit() so a box number is not read as a suite.

Value

A character vector, one per pair:

conflict

Both parse to a unit and they disagree. Two different doors at one street, so the two renderings are very likely different organizations and a triangulated label assigning them one EIN is probably wrong. This is the case worth a reviewer's attention.

agree

Both parse and match. Do not read this as support for one organization. Measured on the triangulation run it is close to worthless: SUITE 100, SUITE 300 and SUITE 400 are shared by unrelated tenants (MIDDLE EAST CHILDRENS ALLIANCE and IRISH REPERTORY THEATER both at Suite 100; GLOBAL FUND FOR WOMEN and THAKKER CHARITABLE TRUST both at Suite 400), a fiscal sponsor's suite is shared by everything it sponsors, and some values are not units at all — 15219 is a Pittsburgh ZIP code sitting on line 2. It is recorded for completeness, not as evidence.

one_sided

One parses, the other does not. Carries nothing — filers omit the line constantly, so silence is not disagreement.

absent

Neither parses.

Two units agree when their identifiers match and their designators fall in the same class: the room-level designators are interchangeable (.PF_UNIT_ROOM), everything else is not, so UNIT 1200 is STE 1200 but FL 2 is not STE 2. This is where pfmatch departs from npmatch, which compares street_unit_id alone and so reads a floor as a suite. Upstream the cost is a missing 0.97 rung in a max() ladder; here it would be a false agree, which no reviewer ever looks at.

Against npmatch's own pair annotations: conflict and agree are geo_unit == 0 and geo_unit == 1 under the stricter designator rule above, and one_sided and absent are the two halves of geo_unit == NA that geo_unit cannot tell apart. The crowding half of the annotation — addr_shared, addr_units, addr_unit_shared, addr_note — has no counterpart here and should be taken from np_compare() rather than reinvented; see dev/PLACES-SYNC.md §4.5.

conflict fires on 0.47% of candidate pairs in the triangulation run — 69 of 14,790, of which 19 sit on a MAYBE candidate and so reach the adjudication queue. It is a rare, sharp signal, not a routine one, and it concentrates: one address (a fellowship program writing a sequential counter into line 2, one per individual grantee) produced 31 of the 69. A conflict is a prompt for a reviewer, never an automatic verdict. See the "Name and address variants" vignette for the full rates.

See also

pf_street_unit()