JustChecking.ai × FCNN · The Methodology

The VERIFY Protocol

Transparency is the product. Every Truth Tracker verdict on FCNN is produced by the same four-step, multi-model, multi-source system — open for scrutiny by any audience member, critic, or journalist who wants to examine it.

4 Verification Steps
5 Council AI Models
7 Output Dimensions
5 Source Tiers

The Problem With How Fact-Checking Usually Works

A single AI model is a single opinion. A single news source is a single perspective. Most fact-checkers use one of each — and call it verification. VERIFY was built because that is not enough.

Conventional Approach

One model. One source. One conclusion.

  • Single AI model, single opinion — no adversarial check
  • Usually one primary source, which may itself be biased
  • Binary true/false verdict hides nuance and uncertainty
  • Opinion presented as fact if the claim is directionally true
  • Hidden findings — things the source reveals about itself — are never surfaced
  • No published methodology for external scrutiny
The VERIFY Protocol

Multi-model. Multi-source. Seven dimensions.

  • Five independent AI models — none sees the others' answers
  • Sources ranked by a five-tier credibility hierarchy
  • Seven-dimension output distinguishes accuracy, confidence, bias, context, and coherence
  • Opinion is rated as opinion, not folded into accuracy
  • Active search for hidden findings — what does the source reveal about itself?
  • Full methodology published and open for examination

How Every Claim Is Verified

Applied to every Truth Tracker segment, in this order, every time. No shortcuts.

1

Claim Extraction

Before anything can be verified, the claims must be isolated. A single article, speech, or social media post may contain dozens of distinct factual assertions, each with different accuracy levels. VERIFY separates them. A statement rated 8/10 accurate on average might contain one claim that is completely true and another that is completely false — the average conceals both. Every claim receives its own individual analysis.

2

The Council — Multi-Model Cross-Examination

Each extracted claim is submitted independently to five separate AI models simultaneously. Critically, no model sees what any other model has said. Where all five agree, confidence in the finding is high. Where they diverge, that divergence is itself the finding — reported as contested rather than resolved by majority vote. AI outputs are then grounded against primary sources: not other AI opinions about sources, but the documents, records, and data themselves.

3

Source Hierarchy Weighting

Not all sources carry equal evidentiary weight. VERIFY applies a five-tier credibility ranking to every piece of evidence: primary documents and official records at Tier 1, down to unattributed rumour at Tier 5. A claim supported only by Tier 4 or 5 sources is rated Unverified — not false. That distinction matters. Most fact-checking collapses this into a binary. VERIFY preserves it.

4

Multi-Dimensional Dashboard — Seven Readings

The final output is not a single verdict. It is a seven-dimension reading displayed as animated gauge meters, covering accuracy score, confidence level, partisan lean, emotional load, context completeness, source quality, and logical coherence. Each dimension answers a different question about the same content. A claim can score high on accuracy and low on context completeness — which is the definition of a misleading-but-technically-true statement. VERIFY catches it. A single number does not.

Why Five Models, Not One

Each model is an independent perspective. Consensus strengthens a finding. Divergence flags a contested claim — and contested claims are reported as such, not resolved by averaging.

Claim Submitted
↓ sent independently to each ↓
Model A
Model B
Model C
Model D
Model E
↓ no cross-communication ↓
Council Verdict + Confidence

No Cross-Contamination

The most important design decision in The Council is that no model sees what the others have said. In a group of humans, if the first person says "true," the second is more likely to say "true" — social conformity and anchoring bias are well-documented. AI models exhibit the same susceptibility when they are exposed to prior outputs.

By isolating each model's response, VERIFY captures genuine independent assessment. Five models arriving at the same conclusion independently carry far more weight than five models that were told what the first one thought.

Key principle: When sources agree, confidence rises. When they diverge, the disagreement is reported transparently — not hidden behind an averaged score. Divergence is data.

The Five-Tier Evidence Ranking

A claim is only as reliable as its source. VERIFY weights every piece of evidence by the tier of its source — and reports that weighting in the final output. A Tier 4 source does not make a claim false. It makes it unverified.

Tier
Category
Examples & Weight
T1
Primary Documents
Official government records, signed treaties, peer-reviewed scientific studies, court filings, financial disclosures, original legislation. Highest evidential weight.
T2
Established News with Editorial Standards
News organisations with published editorial codes, demonstrated corrections processes, and independent ownership — across the political spectrum. Weighted by editorial independence, not perceived ideology.
T3
Secondary Analysis
Think tanks, research institutes, opinion journalism, expert commentary. Valuable context, but the analysis reflects the institution's perspective. Weighted accordingly.
T4
Social Media & Unverified Accounts
Posts, statements, and claims from social platforms — including verified accounts — without corroborating primary documentation. Treated as assertion, not evidence.
T5
Unattributed & Anonymous
Rumour, speculation, anonymous tips, unattributed claims. Not dismissible — many important stories begin here — but cannot be rated above Unverified without corroboration. Lowest evidential weight.

Seven Dimensions, Not One Number

A single accuracy score is insufficient. A statement can be factually correct and profoundly misleading. VERIFY measures both — and six other dimensions besides. Each answers a different question about the same content.

Accuracy Score
0–10 factual verdict based on primary source cross-reference. The core number — but never the only one.
Confidence Level
How certain is the verdict? A 3/10 accuracy can carry HIGH confidence (clearly false) or LOW confidence (insufficient evidence). These are different findings.
Partisan Lean
Does the framing push in a political direction — regardless of whether the underlying facts are accurate? True claims can be framed with strong partisan lean.
Emotional Load
Measures language designed to trigger a reaction rather than inform. Outrage-optimised language and accurate information are not mutually exclusive — but both should be visible.
Context Completeness
What is missing? A true claim can be deeply misleading by omission. This dimension measures what the source left out — often more revealing than what it included.
Source Quality
Where does the evidence actually come from? The Tier 1–5 source ranking is reported here as a visible score, not buried in footnotes.
Logical Coherence
Does the argument follow from its premises? VERIFY flags logical fallacies, false equivalences, non-sequiturs, and unsupported leaps — the structural errors beneath factual claims.
Phase 4 — Coming
⑧ + ⑨
Vocal Confidence & Disingenuousness Likelihood
Real-time live speech analysis — applied to interviews, press conferences, and testimony as they happen. Phase 4 adds these two dimensions to the dashboard for live broadcast use. FCNN will be the first network to use this capability on air.

Four Phases of Verification

JustChecking.ai is built in four phases — from text and article verification through to real-time live speech deception detection. FCNN operates at the leading edge of each phase as it launches.

Phase 1
Live

Text & Article Verification

Full VERIFY Protocol applied to written claims, articles, social posts, and published statements.

Phase 2
Live

Image & Video Analysis

Visual content verification — context checking, manipulation detection, and source attribution.

Phase 3
Live

Audio Verification

Spoken content analysis — transcription, claim extraction, and audio authenticity assessment.

Phase 4
Coming

Live Speech Detection

Real-time verification during interviews, hearings, and press conferences. Vocal confidence + disingenuousness likelihood — live, on air.

See the VERIFY Protocol in Action

The Truth Tracker analysis of the U.S. State Department's Greenland post demonstrates every step of the protocol — including a finding no other outlet reported.

Read the Greenland Analysis → Back to FCNN