◆ METHODOLOGY

How we score credibility

VeriVote AI scores every tracked elected official on a 0–100 credibility index built from primary-source statements, contradictions, and responses to corrections. The formula is public. The inputs are audited.

Last revised · Apr 7, 2026v2.3
◆

TL;DR

  • →We score officials on a 0–100 scale from five weighted inputs; 35% is contradiction rate.
  • →Every published contradiction has two independent human reviewers and a 72-hour right-of-reply window.
  • →The score does not rate policy correctness. A consistent bad idea scores the same as a consistent good one.
  • →The formula is identical across countries, parties, and chambers. No adjustments. Ever.
◆ THE FORMULA

Five inputs. One composite.

Each input produces a 0–100 sub-score multiplied by its weight and summed. Sub-scores are normalized per country and tenure.

35%
Contradiction rate

Direct reversals and materially shifted positions per 100 indexed statements, normalized by tenure.

Example: Sen. A. Reyes: 12 contradictions / 342 statements = 3.5% rate.
20%
Verification depth

How many independent primary sources confirm the original statement. Floor, hearing, and signed document beat transcript and news recap.

Example: ≥ 3 primary sources = full weight. Single-source = 0.3×.
20%
Source diversity

Range of outlets and venues across the record. Officials who only speak to friendly press score lower on this input.

Example: 7 distinct outlets over 12 months = full weight.
15%
Response to corrections

Did the official acknowledge the contradiction on the record within 30 days? Silence is penalized; good-faith clarification is rewarded.

Example: Within 48h on-record = +1.5. Ignored for 30d = -0.9.
10%
Time horizon

Weighs consistency across years, not just weeks. Long-tenured officials with stable positions score higher here.

Example: Consistent on core issue 2019-2026 = full weight.
◆ WORKED EXAMPLE

A real number, line by line

Senator Elena Hart (illustrative) scored 72. Here is every component.

EH
Senator Elena Hart
US SENATE · CA · DEM · WORKED EXAMPLE
72
Composite
Contradiction rate· 35% weight
23.8 / 35.0
Verification depth· 20% weight
18.0 / 20.0
Source diversity· 20% weight
15.0 / 20.0
Response to corrections· 15% weight
9.0 / 15.0
Time horizon· 10% weight
6.2 / 10.0
Composite: 23.8 + 18.0 + 15.0 + 9.0 + 6.2 = 72.0 / 100
◆ LIMITS

What this number is not

Knowing what a metric is not is half of using it well.

  • Not policy correctness — The score does not evaluate whether a position is right or wrong, only whether it is consistent with the record.
  • Not an endorsement — A 94 does not mean we recommend this official. It means their statements are internally consistent and well-sourced.
  • Not a prediction — Past consistency is not a forecast of future behavior. We index history; we do not forecast votes.
  • Not party-adjusted — We do not adjust scores by party, country, or ideology. The same contradiction costs the same points regardless of who made it.
  • Not an opinion survey — No journalist, editor, or reader can move a score. Only primary-source evidence and the formula can.
  • Not a popularity metric — Audience size, media coverage, and public approval do not factor in. A backbencher and a president are scored identically.
◆ REVIEW PROCESS

Three stages. Zero shortcuts.

Every contradiction passes through AI detection, two human reviewers, and a 72-hour right-of-reply window before entering the public record.

1

AI detection

Three language models (current + two diverse architectures) flag candidate contradictions with a confidence score and cited clips.

Typical: 4–12 hr after publication · 94% recall
2

Human review

Two independent editors review every candidate. A contradiction only publishes if both agree — disagreement escalates to a third reviewer.

SLA: 24 hr · 100% of publishes
3

Right-of-reply

Before publication, the official and their office receive a 72-hour right-of-reply window. Any response is published verbatim alongside the contradiction.

72 hr window · responses shown inline
◆ OVERSIGHT

Advisory board

Six independent scholars and journalists review our methodology quarterly. None hold equity; all publish their reviews on the record.

JM
Dr. Julia Marchetti
Columbia Journalism School · Research
Former NYT standards editor. Led fact-check methodology rewrite of 2020.
HA
Prof. Hany Abdelrahman
LSE · Political communication
Published on contradiction detection in parliamentary speech across 12 democracies.
YK
Prof. Yael Kaufman
Hebrew University · Law & ethics
Specialist in right-of-reply doctrine and defamation in digital media.
RS
Ravi Subramanian
Stanford Internet Observatory
Builds adversarial benchmarks for political LLMs. Independent reviewer since v1.0.
MO
Maya Okafor
Knight Foundation · Trust in news
Oversaw the 2023 public methodology audit for two major fact-check organizations.
LK
Linnea Karlsson
Former Swedish ombudsman
Advises on editorial independence and partisan-bias firewalls for news platforms.
◆ FAQ

Ten questions we get most

Weekly cadence prevents score whipsaw from normal news cycles and gives officials a fair right-of-reply window. Daily updates rewarded volume over substance in our v1 trials.
← Back to the ranking