Skip to contents

Fill the resolved NTEE triple on entity

Source: R/md-ingest-ntee.R
md_ingest_ntee.Rd

Populates ntee_code, ntee_rule, ntee_agreement, ntee_n_distinct, ntee_source and nteev2 (§4) by combining the BMF snapshot with the canonical NTEE crosswalk.

Usage

md_ingest_ntee(
  version,
  crosswalk = NULL,
  con = NULL,
  crosswalk_version = "2026-08"
)

Arguments

version

The pinned bmf_unified version (for the active-vintage test).

crosswalk

Path to ntee_resolved_crosswalk.csv.

con

A connection from md_db().

crosswalk_version

Pinned ntee_crosswalk version, used when crosswalk is NULL.

Value

A table of rule counts, invisibly.

The resolution rule

  1. bmf_current - the organisation is in the current active BMF release and the BMF carries an NTEE code. What the IRS says now wins for an organisation that currently exists.

  2. modal - otherwise, the code observed in the most vintages. Robust to a single mis-keyed year, which a most-recent rule is not.

  3. most_recent - if there is no modal value.

  4. bmf_only - for EINs absent from the crosswalk entirely, the BMF code, whatever its vintage.

  5. NULL with ntee_rule = 'none' when no source has a code.

What the measurements say about this rule

Three findings shaped it, all against the pinned BMF and crosswalk.

The BMF and the crosswalk's ntee_current are the same value, not two opinions. Among active EINs where the crosswalk has a current code, it agrees with the BMF on 1,382,208 and differs on 608 - 99.96%. So there is no choice to make between "the BMF version" and "the crosswalked current version"; the real choice is between now and history.

The BMF's NTEE is missing far more often for active organisations than inactive ones - 29.4% against 3.6%. The fallback is therefore not an edge case: rule 1 leaves 581,028 active organisations to be resolved by rule 2. The inversion is not a defect; legacy rows inherit NTEE from NCCS series that assigned it, while the current IRS extract often does not.

For active organisations the BMF code and the modal code disagree 38.8% of the time (741,207 of 1,911,130), so rule 1 versus rule 2 is a consequential choice rather than a formality. For inactive organisations modal and most-recent differ only 4.1%, so rule 3 is nearly free.

nteev2 is recomputed, and how well depends on the rule

fiscal::get_nteev2() is applied to the chosen code rather than any stored rendering, because both the BMF and the crosswalk ship the specialty double-encoding defect (?md_ntee_canonical).

One honest limitation. Rule 1 recomputes from ntee_code_raw, which preserves the legacy five-character form and so gets the B8443 -> B43 rule right. Rules 2 and 3 only have the crosswalk's three-character code, where that information is already gone. ntee_source records which, so rows whose NTEEv2 rests on a truncated input are identifiable rather than merely suspected.