How Kredulous works

Methodology

Kredulous is a nonpartisan, blind candidate-alignment tool. It reports how your responses aligned with sourced records. It does not endorse candidates, does not tell anyone how to vote, and never names an ideal candidate. This page documents the system; every race also publishes its own auto-generated methodology page with its exact numbers.

During a session, cards never show a candidate name, party, photo, slogan, campaign branding, recognizable quote, or telling biographical detail, and never say which candidate holds which position. The data sent to your browser during a session contains no candidate fields at all, and no source links: the link to a source is fetched only after you explicitly accept an identity warning, because original documents can reveal their owner. Identity appears only on your results page.

Sourcing

Every card links to a source you can open, with a stored excerpt, publisher, and retrieval date. Sources are ranked in six tiers from election authorities to reputable secondary coverage; campaign-owned material is always labeled as a candidate-stated position and never presented as fact. Where feasible each source carries an archived snapshot; failures are logged as waivers and retried.

Read the full source hierarchy

Neutral wording and review

Cards describe concrete policy and records, in parallel sentence structure for both candidates on an issue. A loaded-language lint blocks a list of slogans and approval words unless a reviewer logs an override, and an identity-leakage check scans every field a voter sees, in both languages, for names, party terms, and campaign voice. The leakage check can never be overridden. Balance is structural: equal differentiating item counts per candidate and comparable issue coverage. Where reviewed sources show no position from a candidate, the table says so instead of inventing symmetry.

The publish gate

A race cannot go live until it passes nine checks: equal item counts per candidate; comparable issue coverage; a valid, traceable source on every card; an archive snapshot or logged waiver on every source; the loaded-language lint passing or an override logged; zero identity leakage; no unresolved source conflicts; every score impact explainable in the results table; and a generatable methodology page. The gate runs at seed time, on every publish, and on demand from the command line.

Scoring

Agree adds 1 point toward the candidate who holds the position. Disagree subtracts 1 point from that candidate, which in a two-candidate race counts as an item favoring the other candidate. Skip and unanswered items count 0 and are excluded from percentages. Positions both candidates share are shown and counted in your answered total but marked non-differentiating; they cannot move your result toward either candidate. Optional issue weights, when a user sets them, rescale only that user's own view and are always shown next to the unweighted result. Results are reported as ratios and percentages of answered items with a per-issue breakdown, and every per-item impact is listed in the results table. Kredulous never produces a single opaque score, never crowns a winner, and never names an ideal candidate.

Community results

There is no tally anywhere before you complete your own session. Aggregates render only above a minimum of 25 completed sessions, count once per device, are never segmented into small cells, and are always labeled as self-selected users, not a poll.

Privacy

The core experience needs no account and collects no name, email, phone, address, or party preference. Response data is treated as sensitive: never sold, never used for ad targeting, never shared with campaigns or parties at any granularity. Address lookups for ballot matching are forwarded once to the Census Bureau geocoder and never stored. Usage is measured with privacy-friendly, cookieless analytics that count visits, pages, and referrers in aggregate. No cookies are set, no advertising or cross-site trackers are used, and no personal data is stored.

Corrections

Anyone can flag any card from their results page. A person reviews every report, and every resolution is published in the public corrections log and on the affected race's methodology page.

See the corrections log