Release changelog
3.0.0 — 2026-10-10
Data and compatibility
- Build a complete 2010–2024 tract history for the 50 states and DC, with
30 estimates and 30 accompanying 90% margins of error.
- Preserve each ACS year’s published tract identifiers and boundaries, including
Connecticut planning-region identifiers from 2022 onward. Remove geographic
crosswalks and boundary harmonization. Join on
census_tract_id and year;
the previous census_tract_vintage column is removed.
- Limit the current builder to census tracts; historical ZCTA assets remain
separate releases.
- Replace embedded field metadata and tabular-data-resource packaging with three
matching assets: Parquet data, a
.measures.yml catalog, and a .CPI.csv input.
Consumers using the previous dpkg interface should read Parquet directly.
- Use the 2024 R-CPI-U-RS input and express adjusted dollars in 2010 purchasing
power. Revised CPI inputs can change earlier adjusted values.
Corrections
- Correct proportion MOEs when the estimate equals one; use the ratio
approximation only when the subset formula’s radicand is negative.
- Interpret controlled-estimate MOE markers (
***** and -555555555) as zero
sampling error while retaining missing values for suppressed estimates.
- Correct the pre-1970 housing variable mapping at the 2015 table change and
describe poverty as a person-based measure.
- Include inflation-adjusted MOEs for income and home value. Apply estimate
rounding after calculations and retain MOE precision.
Build and documentation
- Drive definitions, availability, calculations, and source-variable mappings
from a validated shared YAML catalog. Reject missing final-year rules and
invalid geography settings.
- Read sequence-based Summary Files through 2020 and table-based files from
2021 onward, with persistent
stow caching and optional offline builds.
- Build one year at a time and write one Parquet row group per year. Protect
existing release assets and record coverage, row counts, columns, and sources.
- Declare build dependencies in
DESCRIPTION and record the R version,
platform, and build-package versions in each newly built release catalog.
- Add fixture-based regression checks for parsing, calculations, catalog
validation, and release writing.
- Expand fixtures for malformed Summary Files, historical lookup offsets and
sequence keys, missing calculation components, zero-count MOEs, CPI errors,
yearly tract unions, and reader caching.
- Prepare source inventories before annual calculations and pass them explicitly
to the release writer; metadata serialization no longer fetches lookup files.
- Generate Markdown catalog tables and expandable source/calculation details for
the release description, alongside the data assets, without committing
generated documentation to the repository.
- Add an MIT license for original code and documentation, source-data reuse
guidance, a suggested citation, and this changelog.
Historical releases
These entries summarize changes verified in the repository’s tagged history.
Tag names retain their historical spelling.
- Add a separate ZCTA-level dataset.
- Add 2023 ACS data and update the build and metadata workflow.
- Add 2022 ACS data and update CPI inputs and release metadata.
- Change
fraction_snap to use B19058, measuring households receiving public
assistance income or food stamps/SNAP, and update its definition accordingly.
- Add housing-unit measures. The same commit is also tagged
0.2.0.
- Add 2020 ACS data and introduce the historical
census_tract_vintage field.