npi 0.3.0 (2026-07-01)

MINOR IMPROVEMENTS

npi_search() now normalizes address_purpose values to API constants and accepts case-insensitive input.
npi_is_valid() now supports vector inputs and returns logical vectors.
npi_search() returns a typed empty npi_results object when no records are found.

npi_summarize() now preserves input rows when NPI records have empty or missing nested address or taxonomy data.
Updated the package CITATION file to use bibentry() instead of the deprecated citEntry() format.

Clarified npi_summarize() taxonomy fallback behavior when no primary taxonomy is marked.

npi_flatten() is now covered by unit tests (#15 and #73).
npi_search() provides an informative error message when internet is off or the endpoint is unreachable (#52).
npi_search() now checks for legal use of special characters prior to submitting an API request (#53).
npi_search() now checks for legal use of wildcard characters prior to submitting an API request (#54).
User agent default now includes the installed package version (e.g., “npi/0.2.0”) (#55).
checkmate package is now in Imports rather than Suggests so tests are run on it (#57).
npi_search() now has test coverage for the case where empty results are returned (#59).
npi_search() now provides a message indicating how many records are being requested (#72).
npi_search() now handles epoch field conversion to date format, fixing a breaking API change. Thanks to @trang-n for reporting the problem (#74) and @parmsam for submitting a PR that fixed it (#75).

npi_summarize() now handles NPI records in which all provider taxonomy records have been flagged as non-primary (#51).
User functions are now backwards-compatible with tidyr <= 0.8.99 (#56).

Grouped the reference index to highlight the most important functions (#45).
Added API URL to DESCRIPTION file (#47).
Clarified description of the npis dataset (#48).
Added Advanced Use vignette (#49).
Clarified the API’s rate limitation (#50).
Removed unused internal functions delay_by() and remove_null() (#58).
Acknowledged rOpenSci reviewers, Emily Zabore (#63) and Mattias Grenié, in DESCRIPTION. Thanks! (#60)
README now clarifies that enumeration_type is returned, not provider_type (#64).
README warns that code results in README may not be reproducible since they reflect dynamic, stateful data (#65).
Vignettes organize search arguments by name instead of description (#66).
npi vignette details what npi_summarize() (#67) and npi_flatten() (#68) do.
REAMDE and npi vignette provide use cases for npi_is_valid() (#69).
npi_search() documentation now links to the NPPES data dictionary to help users understand the API results (#70).