Lines, 2026 edition -- Call the Map (callthemap.com/lines)

Eleven alternative congressional maps for all 50 states, 435 districts on
the 2020 apportionment, each drawn by a computer toward one goal:
Republican and Democratic gerrymanders in two versions (durable: as many
districts as possible that lean the party's way by at least 12 points in an
even national year, so each holds through a 12-point national loss, within
1.3 times a typical map's borders and county splits; aggressive: as many
districts as possible that hold unless the party loses the national vote by
more than 6 points, with no regard for the courts beyond the
majority-minority floors below, within 2 times a typical map's borders and
county splits); a proportional map
(within 1.2 times a typical map's borders and county splits) and a
competitive one (the most districts within a point of even, within 1.5
times a typical map's borders and county splits); and five
maps that use no voting data: straight lines (Warren Smith's shortest
splitline: each state cut again and again by the shortest straight line
that divides its people in the right ratio, a line whose sides would not
hold together by land passed over for the next shortest, then only the
fewest precinct moves to join stray pieces and balance population), town and country (each district
as alike in density as possible, within 1.5 times a typical map's borders),
slices (each district the same mix of city and country, within 2 times) and
county lines (the fewest county splits, then the shortest borders; in
AZ, CA, FL, IL, MD, MI, NY, PA and WA it was searched county by county, so its lines follow
county lines) and centered (called circles until 2026-10-05, and still
"circles" in file names and in atlas.json; every district's people as close
as possible to its population centre: a population-balanced power diagram on
the precincts' population points, then moves that shorten the people-weighted
mean distance to the centre, then a border polish that keeps it within 1%;
built from whole precincts under the contiguity rules, so its districts are
tight but not round; counties and towns are not its goal; atlas.json
"meanDist" gives that distance in km for every map). In MA straight lines is Smith's procedure adjusted for the pinch rule below (86 precincts moved). Still the previous edition's maps, not yet replaced by this one: town and country in MT. The seven geography maps that an earlier search had
grown from a vote-drawn map (CO splitline, CT slices, IA towncountry, NV splitline, WV county, WV slices, WV splitline) were redrawn from starts that use
no votes.

A typical map's borders are the middle count of precinct borders cut among
the state's maps drawn with no goal, and its county splits the middle count
of splits, each ranked on its own; every state has an even number of these
maps, and the limits use the higher of the two middle counts (atlas.json
"neutral" p50). Every district's land is joined by land borders, with
crossings of water only to islands, on every map this edition redrew;
no other map joins a district across open water
(atlas.json "water"). A pinch rule is being added: no
district may hang together by a single land border shorter than 1 km with
1% or more of its people beyond it. The last searches (rounds 6 and 7)
searched the state maps with necks again under it, straight lines aside, and
drew the centered map under it; 217 of the published state maps are
their results. AL's proportional map is still round 5's: its search under the rule left it far less compact. WA's proportional map is still round 5's: its search under the rule left it far less compact. GA's town and country map is still round 5's: its search under the rule left it with more necks.
98 such necks remain in 48 state maps: 61 in
their results, where they could not remove them, 33 in straight
lines, a fixed procedure that is not searched, and 4 in maps still
to be searched under the rule; three are 50 m or narrower with 5% or
more beyond (atlas.json "pinch": district, metres, % of its people beyond;
"pinchSearched").

Contiguity (round 8, 2026-10-04) comes before every map's goal. A district is
contiguous when all of its land is connected over land: its precincts touch
along a shared land border of positive length, a precinct drawn as several
polygons is counted polygon by polygon, and open water joins nothing. The only
crossings of water are to land that no land border joins to the rest of its
state (an island, a detached shore, a peninsula reached only by a bridge); a
land mass with fewer people than a district goes to one district, across water
its precincts and that district's are neighbours over in the precinct data.
Every crossing is listed, 3,201 across 242 state maps, in two classes
(atlas.json "landMasses": each state's land beyond its mainland as county,
km2, people, class; a map's "crossings": district, land mass, km2, people,
kind): 1,749 are to the 159 land masses (class 0: 1 km2 or more, or an island
with water all around it), and 1,452 attach the 132 precinct fragments (class 1:
under 1 km2, holding 67,074 people in all), pieces of precincts that lie against
their neighbours but are drawn apart from them in the source shapes; each is
kept with a district beside it. 233 state maps were repaired to meet this
("onePieceMoved": the precincts a repair moved); a repair that cost a map
seats, compactness or a neck was kept. Every district of all 484 state maps is one piece of land.
They are hypothetical maps, not proposals and not a forecast.

Round 5's search kept each map's goal and county splits
and then kept towns whole: the fewest Census places of 5,000 or more people
split between districts, then the fewest pieces they are cut into, then, in
CT, MA, ME, NH, RI, VT, NY, NJ, PA, MI, WI, MN and OH, where townships and
towns are local governments, the fewest of those split, and only then the
shortest borders. The gerrymanders and the proportional and competitive maps
also had to keep at least today's number of districts where Black, Hispanic
or all minority voting-age people together are a majority (round 4's number
where that was already lower); every one holds it, and six stay below today's count where round 4's map already was (AZ gop-durable, LA fair, LA gop-durable, MI fair, OH competitive, OH dem-aggressive). The proportional maps of AZ, NY, OH and WI are still round 4's, not held to the floors: round 5's were far less compact and split more counties and towns. The counts use a 50% voting-age
threshold and are not a legal analysis. atlas.json gives each state map's
"towns" (places split, place pieces, subdivisions split, from the redraw's
Census 2020 tables), "vra" (Black-, Hispanic- and coalition-majority
districts) and, for the maps held to it, "vraFloor".

Every district is built from whole 2020 Census voting districts (VTDs; block
groups in CA, HI, OR and WV; county subdivisions in parts of Maine), is
contiguous as defined above, and is within 0.5% of its state's ideal population (DEV_PCT
gives the exact deviation). Enacted maps are balanced to the person by
splitting census blocks, which a whole-precinct map cannot do: to adopt one
of these maps, rebalance it at block level first.

FILES
  atlas-2026-<map>.zip           one map's district polygons as a shapefile
                                 (.shp/.shx/.dbf/.prj/.cpg), NAD83
                                 longitude/latitude, simplified to about 250 m
                                 and clipped to the shoreline (Census 2025
                                 cartographic boundaries)
  atlas-2026-vtd-assignments.zip every map, exactly: one row per 2020 VTD
                                 (GEOID20, STATE) and one column per map giving
                                 its district -- the file redistricting tools
                                 import (Dave's Redistricting, Districtr and
                                 others accept VTD assignment files)
  atlas-2026-districts.csv       every drawn district's numbers, all maps
A single state's map as GeoJSON is one click on its close-up at
callthemap.com/lines; `ogr2ogr -f GeoJSON out.geojson <map>.shp` converts a
whole map.

FIELDS
  STATE, STATEFP, MAP
  DISTRICT     the number of the enacted 2026 district it shares the most
               2020 population with, conflicts settled to keep the most
               people in all under their own number; a district sharing
               nobody with the number left to it takes one in reading order
  POP20        2020 Census population
  DEV_PCT      deviation from the state's ideal district population, %
  VAP20        2020 voting-age population
  WHITE_PCT, HISP_PCT, BLACK_PCT, ASIAN_PCT, OTHER_PCT
               share of voting-age population: non-Hispanic white alone,
               Hispanic, non-Hispanic Black alone, non-Hispanic Asian alone,
               everyone else (non-Hispanic)
  MARGIN24     2024 presidential margin, Democrat minus Republican, points
  DWIN_PCT     chance a Democrat wins in a year when the national popular
               vote is even (the Atlas's default), from the 2024 margin with
               the curve fitted to Call the Map's House forecast (atlas.json
               "calibration"); an average relationship, not a race forecast.
               The page counts a district as competitive when each party has
               at least a 1-in-6 chance (DWIN_PCT from 16.7 to 83.3), as
               FiveThirtyEight's 2018 Atlas did, and as likely Democratic or
               likely Republican past that

SOURCES
  Dave's Redistricting, VTD data (github.com/dra2020/vtd_data): 2020 Census
  shapes, population and voting-age population; precinct election returns
  from the Voting and Election Science Team (CC BY 4.0) and the
  Redistricting Data Hub. Where 2024 precinct returns are not in that
  package (AR, CT, ID, ME, MI, NJ, OK, OR, PA and the at-large AK, ND, SD),
  2020 precinct returns are carried forward by each county's actual
  2020-to-2024 swing.
  Method and code: scripts/atlas/ in the Call the Map repository.
