How the published record is built — and what it will never do

Methodology

Models propose; independent reviewers publish. Every profile is pinned to an immutable evidence snapshot, every answer carries a receipt, and missing evidence is an answer — never a blank to be filled. This page is the contract the machinery holds itself to.

Browse the published recordMethodology JSONTrust model
Food security first

The first domain connects kitchens, food banks, delivery partners, agencies, co-ops, and public reviewers around transparent allocation decisions.

  • Current picture: needs, capacity, inventory, delivery windows
  • Evidence trail: source, freshness, confidence, audit record
  • Public view: aggregate dashboards, maps, exports, outcomes
  • Human review: objections, approvals, overrides, follow-through
Public-safeResidents see aggregate information, not private case details.
Human-governedRecommendations require accountable review before action.

Guarantees

These are the promises the machinery enforces — not editorial aspirations. Each one is checked in code and pinned into every receipt.

  • facts assertions and inferences are separate
  • missing evidence is not neutral
  • corrections supersede instead of erasing history
  • campaign submissions cannot replace the independent record
  • donor relationships do not alone establish capture

Claim kinds

Every published sentence decomposes into typed claims. Facts, assertions, and inferences never blend.

  • documented fact
  • politician assertion
  • campaign assertion
  • independent assessment
  • reconstructed position
  • forecast
  • value judgment
  • contested claim
  • unknown

The influence evidence ladder

A funding relationship alone never establishes influence. Each rung requires its own proof, and no source is made to climb higher than its record.

  1. documented relationship
  2. concentration or dependency
  3. access documented
  4. policy alignment documented
  5. evidence consistent with influence
  6. direct influence evidence
  7. official finding

Alignment arithmetic

Deterministic and reproducible: 100 * sum(importance * (1 - abs(userTarget - position) / 2)) / coveredImportance. Stated and observed positions are scored separately; unknown dimensions reduce coverage instead of becoming neutral agreement; dealbreakers are never averaged away. No language model controls any score or ranking.

Machine-readable version

This page renders methodology version civicsyn.political-intelligence.v1. The canonical JSON document is public: /api/public/political-intelligence/methodology.