Methodology

Akashic separates reported facts, derived measures, model outputs, and editorial interpretation. The methodology contract is to preserve that distinction from source ingestion through the public page.

Election results

County, state, and national presidential results are assembled from the source family appropriate to each era. Certified vote totals are not modified to improve a narrative. Candidate and party labels may be normalized for comparison while literal source detail is retained where available.

The complete public presidential record claim is 1892–2024. Earlier returns are partial.

Derived election measures

Margin, vote share, swing, trend, rank, and similarity are derived from reported or reconciled inputs.

A two-party margin is:

(democratic_votes / two_party_votes) - (republican_votes / two_party_votes)

Equivalent implementations may compute (democratic_votes - republican_votes) / two_party_votes. The stored core election margin uses a -1 to 1 proportion; the interface usually displays percentage points.

Sub-county geography

Cities, districts, Census places, and custom Places may require precinct-to-block or block-group processing. The pipeline distinguishes:

• native precinct observations;
• exact aggregation of complete source units;
• reconciled disaggregation from a larger certified unit;
• estimated cycles where local detail is incomplete;
• missing cycles that remain missing.

Every estimate must reconcile to its authoritative parent total and retain a method flag. Reconciliation makes totals consistent; it does not recreate unknowable historical local variation.

Boundary changes

Historical results can be attached to modern identifiers through a documented crosswalk. This supports longitudinal comparison on a consistent modern geography, but it can differ from the exact historical boundary footprint.

Connecticut planning regions, renamed counties, changing districts, and user-drawn Places require explicit treatment rather than silent identifier substitution.

Demographics

Demographic measures use the stated Census or ACS table, vintage, universe, and transformation. Percentages with different universes are not directly interchangeable. The data dictionary identifies common denominators such as total population, population age 5+, adults age 25+, households, and citizen voting-age population.

Religion and sparse geography

Religion counts are based on their source model and may reflect congregation location. Service-shed context can help explain sparse-county undercount, but must be labeled as regional context rather than a direct county measurement.

Similarity

Political Twins and related similarity tools compare standardized voting trajectories over a defined set of elections. Similarity is descriptive. It does not establish that two places have the same demographics, institutions, causes, or future behavior.

Forecasts

Forecasts are probabilistic model outputs, not certified results. A public forecast record begins with editions actually published and locked prospectively. Retrospective backtests are labeled research. Model inputs, uncertainty, calibration, and scoring belong to the forecast methodology for the relevant edition.

Editorial interpretation

Editorial text follows the same evidence and tonal standards for all parties and places. It may synthesize documented patterns but must not invent causal certainty, demographic motives, or unsupported local history.

Provenance model

A complete provenance path includes:

1. upstream source and license;
2. source vintage or certification date;
3. raw geographic unit and identifier;
4. transformations and crosswalks;
5. validation or reconciliation result;
6. generated artifact and Akashic release;
7. public field or page using the value.

The existing detailed methodology page remains available at /methodology/ during migration. This hub will absorb its deep sections into topic-specific pages without losing detail.