RESEARCH SNAPSHOT · 14 OCTOBER 2026

What the map measures

The detailed map contains 124,362 Census block groups across 19 states, plus 3,514 historical archive polygons. All 50 states appear in the overview. Data for the other 31 states are not yet included. The national benchmark is available separately; it does not imply nationwide neighborhood coverage.

Households reporting internet access

The source is the American Community Survey, 2020–2024 five-year estimates, table B28002. The percentage is 100 × (total households − households with no internet access) / total households. This includes reported access with or without a paid subscription. It is a household measure, not the share of land with a signal, not a speed test and not a count of people.

A five-year estimate combines observations over the period. It is not a September 2026 measurement. Reporting internet does not establish usable speed, reliability, affordability or a suitable device. No causal conclusion about redlining is made.

Which area does each number describe?

Each local number applies to its entire Census block group, identified by its 12-digit GEOID. Some block groups cross historical boundaries. The map does not assign a block group's percentage specifically to the redlined portion, a street, a building, or a neighborhood. The selected Census area is outlined in yellow; historical Grade D boundaries remain red.

The verified Pleasant Hill example uses the original 1937 historical D13 polygon (archive area 4063). It is a historical neighborhood reference, not a current statistical unit. Its six linked block groups retain their full boundaries and whole-geography estimates. Most historical city views show archive extents, not municipal boundaries. The Pleasant Hill path uses the Macon-Bibb city and Bibb County boundaries and separate Census records: 1600000US1349008 and 0500000US13021. Their published values match in this release but are never added together. State and US benchmarks use their own published household totals, not an average of block-group percentages.

Pleasant Hill protocol v1.0

The path is historical D13 → Macon-Bibb city → Bibb County → Georgia → included South project states. City, county and state summaries use their own ACS B28002 household counts. The region is explicitly a custom subset of 13 project states, not the full Census South. Its rate is 100 × sum of published state internet-household counts / sum of published state household totals. No mean of block-group or neighborhood percentages is used. Aggregate regional uncertainty has not been calculated.

The pilot links require overlap greater than 1 m² in the retained source geometry. Below-60% and below-80% filters use unrounded point estimates; optional CV screening excludes values above 30% or unassessable values. The map, table and query CSV use the same selected GEOIDs. Excluded records remain unfilled and have a separate export with reasons. This is point-estimate screening, not a significance test.

Pilot block-group, city, county and Georgia panels display 90% margins using the published-MOE subset approximation, with an addition fallback for a negative radicand. CV uses the rate standard error (MOE / 1.645) divided by the rate. Intervals are capped to 0–100 for display. Block group 130210101003 remains 132 of 232 households (56.9%), with a 90% margin of ±28.7 percentage points. These calculations do not account for historical georeferencing error. Frozen pilot counts, geography and provenance.

Uncertainty

Outside the labeled pilot panels, the combined site preserves each state package’s estimates and approximate 95% intervals. Original state CSV downloads retain those 95% intervals; pilot query CSVs instead contain explicitly named 90% fields. Block-group uncertainty uses an 80-replicate first-order ratio approximation where available; invariant or unusable replicates use a published-margin-of-error subset approximation, with a conservative fallback where required. Intervals wider than 20 percentage points are flagged. Each selected record identifies its method. No statistical significance test or matching-based causal analysis is performed here.

These intervals address sampling uncertainty, not all measurement error or historical boundary uncertainty. A zero-household Census area has no rate and is gray. In the state and regional views, gray states are not yet included; zero-household block groups are separately identified in the legend.

Historical boundaries and streets

Historical polygons come from the Mapping Inequality archive service recorded in each state's source manifest. Standard residential grades require an archive grade of A–D, city_survey = 1 and residential = 1. Other archive categories remain available as a separate purple outline layer. Grade D was the historical “Hazardous” category; it is not a current assessment of the area.

Background states use 2024 Census cartographic state boundaries. Detailed Census geography and streets use TIGER/Line sources in the state packages. Street context is limited to the prepared historical study areas, generally within 4 km; it is not complete statewide street coverage. Jefferson County, Kentucky uses the previously documented 2023 road-source fallback.

For web display, most block-group geometry is simplified by 10 metres, historical geometry by 2 metres and streets by 3 metres. The six pilot block groups and D13 use the retained unsimplified input boundaries. Overview state geometry is simplified by 500 metres. Original statistical values are retained. Display boundaries are approximate and should not be used to resolve individual property membership. Alaska and Hawaii use inset positions and are not shown at a common scale with the contiguous states.

Connected historical-area views

The United States, South and West views retain access shading and historical areas. Red dots identify Grade D locations at broad scales; they are not polygon boundaries. Detailed state geometry and local streets load when a state is selected. The broad-scale display uses an additional topology-preserving simplification of 0.003 degrees for Census polygons and 0.0003 degrees for historical polygons. All rates are unchanged.

Each historical area has a table of Census block groups with positive-area overlap in the existing map snapshot; boundary-only contact is excluded. The table presents each full block group's estimate and uncertainty. It does not apportion households or calculate a rate for the redlined portion. Small overlaps and historical boundary uncertainty remain possible. Areas without overlap are explicitly marked, including five Georgia archive areas. Use the Census IDs to inspect the full statistical geography.

Use the historical-area selector or click an outline, then use the up button to move through city, state and region; the verified Pleasant Hill path also includes the county level. The 19 individual state entries below the map open the same connected data. Connected-view data review · Navigation checks.

Coverage and source records

These are the 19 included states, not all states geographically belonging to the South or West. Each link provides original source URLs, retrieval dates and hashes from its state package.

Review and delivery

The combined data review records source file hashes, counts and checks. All 124,362 GEOIDs, household identities, rates and zero-household handling were checked during integration. Earlier interactive selection was tested with a simulated DOM, native Canvas rendering, and representative live Chrome desktop checks. Those prior browser checks do not certify the new protocol controls; see the protocol integration checks and limits. See the browser review for the checked views and limits. Browser CSV file-save completion and mobile/touch interaction remain unverified. The standalone state GIS projects remain the source packages. This website does not require an ArcGIS trial or a live ArcGIS account.

The current phase includes reported internet access and historical boundaries. The full block-group atlas covers internet access. The separate Macon street map includes additional indicators and local facilities from the supplied release. Education, crime and flood analyses are outside the scope of this release. A missing historical archive area does not establish that redlining did not occur.

Return to the connected map →