Where the data comes from.

Sources

• Official voter rolls. The foundation is the government voter file for the jurisdiction — the same registration record used to run the election.
• Vote history. Participation history from the roll drives turnout modeling; we model propensity, we do not observe how anyone voted.
• Transparent enrichment. Where additional fields (such as phone contactability) are added from vendor files, they are matched conservatively and attributed, never invented.

How records are matched

Enrichment is matched against the roll on stable identifiers, not guessed by name alone. When two sources disagree on a field, we apply a documented, deterministic rule — first-write-wins on contact fields — rather than silently overwriting. Placeholder and null values from vendor files are treated as absent, not as data.

Modeling

Scores are recomputed each time the roll changes. Every registered voter carries a turnout probability and a persuasion score; these are model outputs, clearly labeled as estimates, and are never presented as facts about an individual's beliefs.

Change tracking

Re-imports never destroy prior records. Additions, updates, and removals to the roll are tracked and reviewable, so the lineage of any record — when it entered, what changed, and why — is available rather than lost between refreshes.

What we don't do

• We do not compile sensitive personal profiles beyond what the voter file and disclosed enrichment contain.
• We do not sell voter data.
• We do not expose PII to roles that have no need for it.

For how that data is protected in storage and access, see Security.