npmatch ground-truth master (1,502-org benchmark)

Query-level ground-truth labels for the benchmark: one row per source organization. This is the archival label key — the training/validation target that the evaluation frame joins onto.

Usage

npmatch_groundtruth

Format

A data frame with 1,502 rows and 17 columns, including:

uei

Source (SAM) Unique Entity ID.

sample_type

"random" (1,000 representative) or "hard" (502 oversampled).

sam_name

Source organization legal name.

gt_is_match

TRUE if a correct BMF EIN exists.

gt_outcome_class

in_bmf_match / nonprofit_not_in_bmf / not_a_nonprofit / no_match_or_na / cant_determine.

gt_ein

The correct EIN (may be an inactive-BMF or externally-collected EIN).

gt_ein_in_bmf, gt_ein_bmf_active

Whether gt_ein is in the BMF, and active.

gt_method, gt_confidence, gt_notes

How the label was determined, confidence, and notes.

Source

See vignette("training-dataset", package = "npmatch").

See also

npmatch_eval