Diagnostic prototype
What this tool can — and cannot — tell you
Honeycombing compares enacted district maps against a neutral hexagonal reference grid (Uber's H3) to make spatial patterns visible. The grid predates any election and was drawn for logistics, not politics — that neutrality is the point. But visibility is not proof. Everything here is descriptive; before citing any number, read the six limits below.
- 1
Not legal evidence
Nothing on this site demonstrates illegal intent or legal injury. Every readout is descriptive and exploratory. Partisan-gerrymandering claims are not heard in federal court at all (Rucho v. Common Cause, 2019), and the state-court and reform paths that remain have evidentiary standards this prototype does not meet.
Treat the maps as a way to see spatial pattern and ask better questions — not as a finding about any map's lawfulness.
- 2
Not a seat forecast
H3 hexagons are (roughly) equal area, not equal population, and they are not districts. Counting red versus blue hexes systematically understates dense urban populations, so raw hex tallies are never a measure of seats a party should win.
Where the site reports seat counts, they come from district-level aggregation of a documented plan or ensemble — never from hex counts.
- 3
Point-assignment shortcut
Precinct returns and census blocks are assigned to hexagons by a single interior point, not by splitting geometry across cell boundaries. Our own audits (Alamance and Mecklenburg counties, NC) rate this shortcut disqualifying for cell-level evidence: single-cell vote-share errors reach 9–16 percentage points.
The same audits show the error largely washes out at district-level aggregation (at most 0.12 percentage points in the ALARM calibration check), which is why district summaries are reported and single-cell values are orientation only.
- 4
District heat is county-derived
The district fill layer aggregates county-level 2020 presidential returns into districts, not precinct-to-district totals. It exists for coarse orientation; precise district vote shares require precinct-level aggregation that is not built yet.
- 5
Ensemble percentiles are not intent
A plan sitting at an extreme percentile of a comparison ensemble — for example, North Carolina's 2023 enacted plan (used in the 2024 election) at the 2.8th percentile of Democratic seats in the ALARM ensemble — is atypical relative to that documented constraint set. Nothing more.
Percentiles do not establish motive or unlawfulness, and they cannot show whether deviations were legally required (for example, Voting Rights Act compliance). The ensemble's constraints are always displayed alongside its numbers.
- 6
Partisan signal, not racial — and the law is moving
This tool measures a partisan seat outlier; it says nothing about race. The live challenge to North Carolina's 2025 map is a racial vote-dilution case, and the Supreme Court sharply narrowed the Voting Rights Act in Louisiana v. Callais (April 2026). Nothing here bears on those racial claims or on what the law now requires.
The comparison ensemble holds majority-minority districts near a 2020-cycle baseline — a simulation constraint, not a VRA analysis, and one that predates both the redraw and Callais.
Two further caveats apply everywhere: presidential returns are a partisan-lean proxy, not congressional performance; and no citizen-voting-age-population or racially-polarized-voting analysis is included, so no Voting Rights Act conclusion can be drawn from anything on this site.
Full method, data provenance, and the expert review packet live in the white paper.