Back to methodology

Scoring model

How values become alignment signals

The Informed Voter should use a mostly stable core quiz, then map sourced candidate signals to that quiz. Ballot-specific questions can be added when local races make them necessary, but the baseline should not change every time a user enters a new address.

Core quiz areas

Taxes and public services

Healthcare

Education

Climate and infrastructure

Gun policy

Housing

Public safety

Voting access

Business regulation

Local control

Scoring path

Step 1

Collect stable values

The core quiz should stay mostly the same across ballots so users can compare candidates consistently across federal, state, and local races.

Step 2

Add ballot-aware follow-ups only when useful

A local ballot can add one to three targeted questions when the user's races make them relevant, such as school board curriculum, water districts, bonds, or judicial philosophy.

Step 3

Map sourced candidate signals to quiz areas

Each alignment signal must come from a candidate source, questionnaire, voting record, endorsement context, or other cited source. Unsourced assumptions should not score.

Step 4

Score only review-ready signals

Signals can be shown before scoring, but they should not affect a match percentage until candidate identity, source quality, and issue mapping are reviewed.

Step 5

Explain, do not command

The output should say which values appear aligned or unresolved. It should avoid 'vote for' language and give users source links to verify the reasoning.

Guardrails before scoring

Do not score races with unresolved runoff or uncertified candidate lists.

Do not infer a candidate position from party alone.

Do not force a score when only one candidate has usable source data.

Show 'not enough info' when source material is too thin.

Keep judicial races eligible for explainers even if they are not scored.

Concrete scoring formula

User answer

-2 to +2

Strongly disagree is -2, neutral is 0, and strongly agree is +2.

Candidate signal

-2 to +2

A reviewed source-backed signal receives the same scale only after human review.

Issue alignment

1 - distance / 4

Exact agreement scores 1. Opposite ends of the scale score 0. Neutral distance lands between those values.

Candidate match

weighted average x 100

Average only the reviewed signals that map to answered quiz items, then convert to a percentage.

candidate match = weighted average of reviewed issue alignments across answered quiz items only

Signal weights

Direct candidate statement

1.0

Campaign platform, candidate questionnaire, debate answer, or public statement.

Voting record or official action

1.0

For incumbents, official votes or public acts can score when clearly tied to a quiz issue.

Endorsement context

0.5

Endorsements can support an explanation, but should not dominate a score unless the issue connection is explicit.

Finance or donor pattern

0.25

Finance data can be a weak contextual signal, not a substitute for candidate positions.

Display thresholds

Show a match percentage only when a candidate has at least 3 reviewed scoring signals.

Show 'limited information' when a candidate has 1-2 reviewed scoring signals.

Show 'not enough info' when a candidate has 0 reviewed scoring signals.

Suppress comparative ranking when one candidate has substantially less reviewed data than another.

Show issue-level explanations beside every score so users can verify what drove the result.

Missing-data rules

Missing data does not count as disagreement.

Unanswered quiz items are ignored for scoring.

Signals marked needs-review can appear as alignment signals but cannot affect the percentage.

If candidates have uneven data coverage, show coverage warnings instead of pretending the comparison is complete.

What this means for the pilot

TX-10 and HD-47 can show alignment signals now because candidate websites and neutral source placeholders exist. They should not show match percentages until final candidates are confirmed and the signal mappings are reviewed.