Location geocoding workflow

1. Pull and clean

library(locatr)

cleaned <- records %>%
  clean_addresses(
    id = `Location ID`, address = Address,
    city = City, zip = Zip, name = `Location Name`
  ) %>%
  flag_bad_addresses()

clean_addresses() adds *_clean columns and a full_address_clean string; flag_bad_addresses() sends PO boxes and placeholders straight to review. Only address and city are required. If your file has no ID, locatr generates row-number IDs; if it has no ZIP, zip_clean stays NA and the single-line address omits the ZIP instead of ending in NA:

cleaned_minimal <- records %>%
  clean_addresses(address = Address, city = City, state = "NJ") %>%
  flag_bad_addresses()

Missing ZIP is recorded as bad_address_flag == "missing_zip" for audit, but it does not block geocoding when address + city + state are present.

For one-off review, use geocode_address() to see ranked ArcGIS candidates for a single address:

if (interactive()) {
  geocode_address("1600 Pennsylvania Ave NW", city = "Washington", state = "DC")
}

2. Geocode with a guarded cascade

geocoded <- geocode_records(cleaned)   # Census -> ArcGIS -> name lookup

geocode_records() runs each tier in turn and validates against the configured region after each pass, so a later, fuzzier service only retries what is still unplaced. The geocode_pass column records which tier placed each row.

3. Geography join

with_geography <- add_county_muni(geocoded, state = "NJ")

The geography step is independent of the input address columns once coordinates exist. add_county_muni() builds Census TIGER/Line geography and attaches county/locality fields. Pass an sf boundary layer to adapt the geography join, e.g. add_muni_from_shapes(geocoded, muni_shapes = my_local_shapes), or use add_muni_from_key() when your records and geography share a code column.

4. Review, override, export

write_geocode_review(with_geography, "manual_review.csv")

final <- with_geography %>%
  apply_manual_overrides("manual_review_completed.csv") %>%
  export_location_crosswalk("location_crosswalk.csv")

final is ready for Tableau, GIS joins, or a reusable reference table, and every row carries the audit columns that explain how its coordinate was produced.

5. Reproducible runs and provenance

Because the cascade calls external services, reuse a locatr_cache() to make a run reproducible and cheap to repeat:

cache <- locatr_cache("geocode_cache.rds")  # omit the path for a memory-only cache

geocoded <- geocode_records(cleaned, cache = cache)

# a repeat run replays cached coordinates instead of re-querying
geocoded_again <- geocode_records(cleaned, cache = cache)

cache_info(cache)

The cache is keyed by the exact query and request parameters and stores one row per candidate result (with a no-match sentinel so misses replay too). Pass refresh = TRUE to re-query and overwrite. Nothing is written to disk unless you give locatr_cache() a path.

Every geocode_records() result also carries a run manifest and two per-row provenance columns:

geocode_provenance(geocoded)

geocoded[, c("record_id", "placed_at", "cache_status")]

cache_status is fresh, cached, reference, manual, or unplaced, and placed_at is when the coordinate actually entered the output (the cached timestamp for cached rows, not the current run). Both columns are carried into export_location_crosswalk(). The manifest is attached as an attribute, so read it with geocode_provenance() right after the run, before any later data-frame operation that might drop attributes.