How the Congressional Capture Portal computes and assigns risk tiers to elected officials.
The Capture Risk Score is a composite index (0–100) estimating how susceptible an elected official may be to regulatory capture — the phenomenon where public officials prioritise narrow private interests over their constituents. The score is not an accusation; it aggregates publicly documented signals into a single comparable metric.
Scores are relative to the rest of Congress. Each component is the member's standing among every scored member: 80 means the member's count is at or above 80% of their colleagues'. A zero count always scores 0. The weighted sum of the five standings is the overall score.
| Component | Weight | What It Measures | The count that is ranked |
|---|---|---|---|
| Contradiction Risk | 30% | Contradictions between stated commitments and the record | (count × 15) + (high-severity × 20), plus +3 per finding and +8 per high-severity finding from the last 180 days |
| Donor Influence | 20% | Distinct documented donors: PACs from FEC Schedule A filings and other primary or secondary interests. Inferred donors do not count. | documented donor count |
| Connection Density | 20% | Mapped relationships to lobbyists, contractors and organised interests. Donor links are left out, since donors are their own component. | non-donor connection count |
| Silence Risk | 15% | Topics where the member's donors have documented interests and the member has made no statement or floor action | active silence flags backed by a documented donor interest; inferred flags do not count |
| Intelligence Volume | 15% | Documented facts from primary and secondary sources | fact count |
| Constituency Deviation | 0% | Shown, not weighted: the district priorities were a model's estimate, not a record | avg_deviation + (high_deviation_count × 10), capped at 100 |
| Voting Misalignment | 0% | Shown, not weighted: needs platform commitments, which few officials have on file. See the roll-call record on /votes instead. | (contradicted / total_votes) × 200, capped at 100 |
Tiers are ranks, not fixed score thresholds. Until 2026-09-28 they were fixed thresholds, and once every member's FEC money was on file 507 of 542 officials crossed the top one, which told a reader nothing.
| Tier | Share of scored members | Interpretation |
|---|---|---|
| CRITICAL | top 5% | The strongest combined documented signals in Congress |
| HIGH | next 10% | A documented pattern well above most colleagues |
| ELEVATED | next 20% | Above the typical member in one or more components |
| MODERATE | next 30% | Within the typical range for an active legislator |
| LOW | the rest | Fewer documented signals than most colleagues |
| INSUFFICIENT | — | Not enough substantive evidence (facts, contradictions, votes, or baselines) to produce a meaningful score |
Contradiction findings from the last 180 days receive a recency boost (+3 per recent finding, +8 per recent high-severity finding). This ensures that freshly documented reversals carry more weight than historical ones, while not discarding older evidence entirely.
High-severity contradictions (semantic analysis score ≥ 70/100) receive an additional per-hit bonus of 20 points (vs. 15 for standard contradictions), reflecting that verified direct conflicts between stated positions and votes are the strongest capture signal.
Officials must have at least one of: ≥ 2 documented facts, ≥ 1 contradiction analysis, ≥ 1 voting record, or ≥ 1 constituency baseline. Without substantive evidence, the system assigns "Insufficient Evidence" rather than a misleading numeric score. Edge-only signals (silences, donors, connections) do not satisfy the gate.
This score measures documented signals, not proven corruption. A high score means the public record contains patterns consistent with capture — it does not prove intent. Officials with extensive public records may score higher simply because more data is available. The score is recalculated on every request; it changes as new intelligence is documented. Components with sparse data across the population (donor interests, constituency baselines, voting misalignment) carry reduced weight until more data is ingested.
