catchmentACS

• cacs_set_cache(scope = "global") no longer writes to the R startup file. It sets the option for the session and shows the line to add to the startup file. confirm is kept so that code written for earlier versions still runs, but it is not used.
• cacs_cache_dir() no longer creates the cache folder unless create = TRUE.
• as_tibble() on a cacs_run() result keeps the rows in their order by default. rate_first = TRUE puts the rate rows first within each site and drive time, and options(catchmentACS.rate_first_default = TRUE) does the same for conversions called from your own code, while conversions made inside other packages keep the order. As a result, dplyr::semi_join(), dplyr::anti_join(), and summaries after dplyr::rowwise() return the right rows.
• The map functions (cacs_plot_site_isochrone(), cacs_plot_site_intersection(), cacs_plot_site_weighted(), cacs_plot_site_rates(), and cacs_plot_site_pipeline()) now draw the standard OpenStreetMap map by default (tiles = "OpenStreetMap"). The CARTO tiles used before need an API key, which these functions do not send, and they showed a notice asking for one. Any other provider in leaflet::providers can still be chosen with tiles.

Bug fixes

• cacs_intersect_weight() and cacs_run(acs = ) now treat the Census Bureau's annotation codes (-222222222, -333333333, -555555555, -666666666, -888888888, and -999999999) in the estimates and margins of error of ACS data given to them as missing, with a warning that names the codes found; they used to enter the sums. A code in place of a margin of error therefore makes that margin of error NA, and a rate that uses the variable NA as well. Data from cacs_acs_prefetch() and the bundled data give the same results as before.
• cacs_acs_validate(), cacs_intersect_weight(), cacs_run(acs = ), and cacs_acs_prefetch() now stop with an error when a tract has two rows for the same variable; such a row used to be added twice.
• cacs_intersect_weight(), and cacs_run() through it, now recognize saved results by the drive-time areas themselves, so a subset or an edited copy of a cacs_isochrone() result no longer returns rows saved for another subset. The key also no longer depends on how the row names of the areas are stored. Results saved by earlier versions are computed again once.
• cacs_isochrone() now stops at the first site that gets HTTP status 429, 401, or 403 and sends no request for the remaining sites; before, it tried every site first.
• cacs_isochrone() no longer reads a port number in a connection error, such as "port 443", as an HTTP status. A connection failure or a timeout is therefore tried again, up to three times, as the help page describes; before, the site was not tried again.
• cacs_acs_prefetch() likewise no longer reads a port number or a waiting time in a connection error as an HTTP status. A connection failure is tried up to three times and then stops with an error of class catchmentACS_error_network; before, it stopped at once with an error saying that the Census Bureau had refused the request.
• cacs_isochrone() no longer saves a result in which routing failed for a site in a way that may be temporary (no answer, a timeout, or an HTTP 5xx status), and it does not use such a result saved by an earlier version, so a temporary failure is not returned again from the cache. A site that the routing service rejects with another HTTP 4xx status is saved with the result, and a warning says so when the saved result is used.
• A server set with options(osrm.server = ) now stays in use over repeated cacs_isochrone() calls; before, loading the osrm package during the first call replaced it with the public demo server for later calls, and the record kept with the result named that server.
• cacs_acs_prefetch(), and so cacs_run(), no longer stops with "package lwgeom required" when spherical geometry is turned off with sf::sf_use_s2(FALSE): the check for tracts without area now measures the tracts in EPSG:5070 in that case, so the lwgeom package is not needed.
• cacs_intersect_weight(), and so cacs_run(), now leaves the option sf_use_s2 as it was. It turns spherical geometry off while it runs, as before, but it used to set the option to TRUE afterwards when the option had not been set.
• cacs_run(output = "list_column") no longer gives the warnings Unknown or uninitialised column when sites is an sf object built on a tibble without lon and lat columns. The two columns of the result are NA, as before, and are read only from columns with exactly these names.
• cacs_acs_prefetch() now reads back a result it saved together with its message about removed water tracts also when the package is installed with its sources kept (for example with R_KEEP_PKG_SOURCE=yes). The saved message kept a reference to the package source that changed when the file was read, so the saved copy never matched its checksum file and each call downloaded the data again.
• The map functions now find the bundled cacs_alabama_sites table when they are called as catchmentACS:: without library(catchmentACS), and an object of the same name in the global environment no longer takes its place.
• cacs_derive_rates() now gives an error for an argument passed through ..., such as level = 0.95, which it used to ignore; the confidence level is set in cacs_propagate_moe().
• cacs_intersect_weight(), and so cacs_run(), now stops with an error of class catchmentACS_error_geometry when the geometry of no row of the drive-time areas or of the tracts can be repaired; before, an error in writing the message took its place.
• cacs_summary_as_markdown() now names the confidence level of the result in the caption of rates_per_site_moe, which always said 90%; summary() records the level on that table.

Other changes

• cacs_intersect_weight() has a new argument cache_dir, as cacs_isochrone() and cacs_acs_prefetch() have, and cacs_run(cache_dir = ) now applies to all three steps.
• cacs_clear_cache() and cacs_cache_status() now handle only the files named the way the package names its saved files, and they include the GeoPackage files written by cacs_acs_prefetch(write_gpkg = TRUE). cacs_clear_cache(confirm = TRUE) in a non-interactive session now says that it deleted nothing because it could not ask for confirmation.
• The phase values that cacs_capture_conditions() returns for conditions from cacs_run(), cacs_derive_rates(), and cacs_propagate_moe() are now "run", "rates", and "moe"; before, they were longer labels that began with a section number of an internal design document. Several warnings and messages that had no phase now have the name of their step, such as "isochrone" or "intersect".
• Error, warning, and progress messages were rewritten to say what happened and what to do. They no longer refer to development versions, to sections of an internal design document, or to functions and arguments that do not exist (such as cacs_derive_rates_custom() and crs_area). The messages about the OSRM res value used when res is not given say which value was used and how to choose another. The condition classes are unchanged.
• cacs_validate_iso() gives plainer fix_hint and expected text for ring topology, provider values, and missing columns. The check values and the columns of the table are unchanged.
• The suggested packages mapboxapi, r5r, httptest2, covr, lintr, pkgdown, and roxygen2 were dropped; none of them was used by the package. Building the vignettes needs knitr 1.35 or later.
• The warning of cacs_derive_rates() about rates that are NA or use a replacement formula no longer names an internal function as its call. It names the call of cacs_derive_rates(), or no call when cacs_run() calls it.
• A cacs_run(output = "list_column") result no longer has the attribute iso_was_filled, which was used only inside the package.

Documentation

• The vignettes are R Markdown documents built with knitr instead of Quarto documents, so building the package no longer needs the Quarto command-line tool. Their formulas are written as MathML: in the vignettes opened with vignette(), longer formulas were shown as TeX code before and are now displayed as formulas, also without an internet connection.
• Every exported topic has an example that runs; ?cacs_isochrone and ?cacs_acs_prefetch show results of those functions stored in the package.
• The help pages describe the changes above and, where they apply, the limitations below.
• The new file COPYRIGHTS names the bundled data files that hold OpenStreetMap or Census Bureau data and gives the notices for those data.

Known limitations

• cacs_isochrone() and cacs_run() do not build drive-time areas with provider = "mapbox" or provider = "r5r", which are not implemented and give an error, and cacs_derive_rates() accepts only the five built-in rates.
• cacs_isochrone(provider = "ors") has been checked only with simulated responses from openrouteservice.
• cacs_isochrone() sets no time limit on its requests. A routing server that accepts the connection but does not answer makes the call wait, and a server that cannot be reached is tried three times for each site, so a call with many sites can take minutes before it returns a result in which every site failed. cacs_validate_osrm_endpoint() checks an OSRM server with a time limit.
• cacs_acs_prefetch() downloads American Community Survey (ACS) 5-year estimates for the census tracts of one state per call.
• cacs_intersect_weight() and cacs_run() decide how to combine a variable from its ACS code alone. Codes in tables B19013 and B25077 are treated as medians, codes in table B19301 as per-person values, and every other code made of "B", five digits, an underscore, and three digits as a count, which is added up over the tracts. A median or per-person value with a code of that form from any other table therefore comes out as a sum of tract values, with weight_basis = "coverage" and no warning.
• A rate from cacs_derive_rates() or cacs_run() divides the total of its numerator by the total of its denominator, even when the two totals come from different tracts. This happens when the ACS data have a tract's row for one of the two codes but not for the other, for example after rows with a missing estimate were removed. The rate can then be far from the share for the area, even above 1, without a warning; n_tracts_num and n_tracts_den then differ, unless the two codes lack the same number of tracts.
• When the ACS data have no row for a tract and a variable, cacs_intersect_weight() and cacs_run() leave that tract out of the variable's total or average without a warning, so a total comes out too small, and cacs_acs_validate() does not report the missing row. n_tracts on that variable's row is then smaller than on the rows of the variables that have the tract.
• The part of a drive-time area that has no tracts in the ACS data, such as the part across a state line, adds nothing: counts come out too small, and rates and medians come from the remaining tracts. The only warning compares the bounding box of all the drive-time areas with the bounding box of all the tracts, so it is not given when the uncovered part lies inside that box or reaches less than 0.05 degrees (about 5 km) beyond it.
• The drive-time areas in the example file legacy_2025_isochrones.rds are circles with a radius of 1 km per minute of driving, not areas computed by a routing service, yet their routing columns hold fixed values such as provider = "osrm", routing_engine_version = "OSRM 5.27.1", and osm_snapshot_status = "explicit". Results computed from them therefore name OSRM as the provider, in their columns and in the output of print() and cacs_describe().