Package index • npmatch

Run projects & stage drivers

Scaffold a run directory, address files inside it, drive each stage, and report on progress. See the Workflow article for the layout these produce.

np_project_init(): Create a run project directory
np_project_use(): Point the session at an existing project
np_project_root(): Locate the active npmatch project
np_project_path(): Build a path inside the active project
np_project_status(): Report the status of a run project
np_stage_schema(): The shared outcome schema
np_stage1_run(): Run stage 1 - the probabilistic cascade
np_stage2_run(): Run stage 2 - LLM adjudication of the MAYBE queue
np_stage3_run(): Run stage 3 - LLM research of the NO cases
np_final_run(): Roll the stages up into the final deliverables
np_gate_rules(): Entity-gate rules
np_entity_gate(): Classify source records before research
np_merge_decisions(): Merge adjudication decisions onto a review frame

Match pipeline

Tiers 3–6: pre-prep the reference corpus, generate candidates by blocking, compare and score them, apply the veto rules, select and tier. np_cascade() orchestrates all of it, and takes a matchdb reference directly.

np_cascade(): Match with a progressive blocking cascade
np_match(): Run the full linkage pipeline
np_block(): Generate candidate pairs by blocking
np_block_union(): Union candidate-pair sets from several blocking passes
np_ref_index(): Precompute the reference-side token index for blocking
np_compare(): Generate and compare candidate pairs
np_score(): Score candidate pairs
np_veto(): Apply veto rules to candidate pairs
np_select(): Select best matches per query record
np_tier(): Sort selected matches into YES / MAYBE / NO tiers
np_route(): Route a tiered result to accept / review / unmatched

Entity families

Roll a query up to the family its candidates belong to — a chapter of a federation, a hospital system, a school and its PTO — and tell a contest between strangers from one within a family. Both take cluster membership from matchdb (cluster_member joined to cluster).

np_cluster_rollup(): Match one level up: roll candidates up to their family
np_pair_links(): Are a query's top two candidates one family?

Batched matching

Run the stage-1 cascade over a large source in evenly sized chunks.

np_batch(): Split a data frame into batches
np_run_batches(): Run stage-1 matching in batches against a shared reference

Setup & standardization

Tier 2: map any dataset onto the canonical fields, from which every later tier works. The standardization itself — np_normalize(), np_name_core(), the name, token and address statistics, and the signature that pins them — lives in npnormalize and is re-exported here unchanged. np_map_sam() / np_map_bmf() are the maps for the SAM-to-BMF deployment; write your own for another one.

np_config(): Configuration for a linkage run
np_fields: npmatch canonical fields
np_query(): Construct a query frame
np_reference(): Construct a reference frame
reexports np_normalize np_normalize_signature np_name_core np_name_placeholder np_name_rules np_stopwords np_name_freq np_token_idf np_addr_freq: Objects exported from other packages
np_map_bmf() np_map_sam(): Built-in schema maps
np_flag_active(): Flag which reference EINs are active
np_geo_profiles(): Geographic profiles
np_detect_geo(): Detect the geographic profile of a query frame

Rules, distinctiveness & training

Do-not-match rules, token/name statistics, and model benchmarking.

np_default_rules() np_rule_legal_form(): Default veto rule set
np_rule(): Build a veto rule
np_default_passes(): Default progressive blocking passes
np_default_weights(): Default per-profile field weights
np_default_geo_weights(): Default location-granularity credits
np_train(): Train a supervised match combiner
np_benchmark(): Benchmark scoring methods by cross-validation
np_auc(): Area under the ROC curve (Mann-Whitney form)
np_evaluate(): Evaluate scored pairs against known labels
np_tune_thresholds(): Tune YES / MAYBE thresholds from labelled data

Review & labeling hand-off

Tier 6: package what the algorithm could not settle. Review frames, LLM prompts, and calibration/labeling tools.

np_label_frame(): Build a labelling frame for training and evaluation
np_calibration_frame(): Build a labelling frame for threshold / weight calibration
np_label(): Attach labels to candidate pairs
np_preview_pair(): Preview the field-by-field comparison for one candidate pair
np_veto_audit(): Extract vetoed pairs for auditing and training
np_bmf_review_fields(): Default BMF context fields for the review queue
np_sam_review_fields(): SAM context fields for the review queue
np_as_prompt(): Render a review case as an LLM adjudication prompt
np_as_prompts(): Render every review case as a prompt
np_review_report(): Render a review frame as a static HTML report

Data management, fetching & reporting

Resolve data assets, fetch/prepare source files, run incremental diffs, and report on a run. These build a reference from raw NCCS BMF and SAM files; matchdb is the shorter path to a reference, and supplies tier 1 (alias enumeration) as well.

np_data_root() np_data_path() np_data_init(): npmatch data asset directories
np_manifest() np_manifest_add(): Record and read the data-asset manifest
np_source_urls(): Known npmatch data sources
np_fetch(): Download a data file
np_fetch_bmf(): Fetch the reference BMF (pinned archive)
np_fetch_sam(): Fetch a SAM extract (pinned archive or a URL you supply)
np_read_sam(): Read a raw SAM public extract
np_sam_layout(): SAM public extract column layout
np_flag_nonprofits(): Filter a SAM extract to nonprofits
np_prepare_sam(): Download-and-prepare a SAM extract for matching
np_diff_unmatched(): Retain only source records not already matched (incremental matching)
np_run_report(): Summarise a cascade run as a report

Datasets

The hand-labeled 1,502-org training/evaluation benchmark.

npmatch_eval: npmatch evaluation frame (1,502-org benchmark)
npmatch_groundtruth: npmatch ground-truth master (1,502-org benchmark)

Geography helpers

np_clean_geo(): Clean and standardize city and state fields
np_city_aliases(): Common city aliases
np_augment_geo(): Augment records with ZIP-level geography
np_ruca_urban(): Collapse a RUCA code to an urban / rural flag
np_load_hud(): Load a HUD USPS ZIP crosswalk file
np_load_ruca(): Load a USDA ERS ZIP-code RUCA file
np_load_zipcode_db(): Build a ZIP crosswalk from the zipcodeR package