Data & methodology
How Politeion works, in the open — the two halves of the site are measured differently, so they're documented separately.
Analyzer methodology
How the article Analyzer scores stance and what it flags. The exact scoring prompt is proprietary; this is a summary of the method.
The Daily Brief
The Politeion Daily Brief is drafted by synthesizing wire-service coverage across multiple outlets (a story is only used when ≥2 outlets carry it), then machine-certified: every item is scored by this same Analyzer and may publish only if it registers no detectable stance and no flags. An item that fails is rewritten once with its failing framing fed back, then held for human editing. Each published item links a neutrality receipt. The source list is public config (data/brief_sources.json).
Test methodology
The self-test places you on 22 independent axes. Each item contributes to one or more of those categories; the per-item weights are proprietary, so what's published here is the shape of the bank — how many items cover each axis and how evenly they're keyed.
Per-axis coverage
Candidates methodology
Candidate profiles apply the same 22-axis Analyzer to what candidates publish in their own voice — measuring rhetoric, not records. Each state registry is compiled and sourced independently; the registries currently loaded are listed below with their validation status.
- First-person sources only. A source qualifies only if it is the candidate's own words — an authored op-ed or essay, a floor/committee statement, a candidate questionnaire, or a verbatim interview transcript. Third-person campaign pages ("Smith's platform…") do not qualify, however long.
- Verbatim voice, never quote density. A news feature that merely quotes a candidate is not coverage of their voice; quote-dominated interview features are excluded in favor of transcripts and authored text. (In this pilot two such interview items were removed from one candidate's sources under this rule.)
- The ≥3 threshold. A per-axis score renders only once ≥2 analyzed sources support it, and the full profile renders only at ≥3 qualifying sources. Below that, the candidate shows a thin-corpus notice rather than a shaky profile.
- Rhetoric, not votes. Profiles measure what candidates say in their own published words — not their votes, conduct, or record. Every axis cites the verbatim evidence behind it.
- Isolated from the Analyzer's public data. Candidate analyses never enter the outlet/author
leaderboard or aggregates — a floor speech fetched from a
.govdomain never mints an outlet profile.
Refresh calendar. Pending slots
re-verify once their primary is held. The 2028 tracker is reviewed monthly. Full source-by-source
provenance is in docs/CANDIDATES_PROVENANCE*.md — one report per registry.