How the methodologies work

Shared discipline across the suite

Veridi, Pragma, and Praxis each address a distinct question, but they share a common discipline:

• Explicit procedures. No reliance on instinct or pattern-recognition; every step is documented.
• Evidence ranked, not trusted. A four-tier source hierarchy + an Institutional Reliability Index applies across all three products.
• Gaming detection. Pragma’s 14 vectors (canonical count after the v1.2 audit), 6 native + 8 inherited Praxis vectors. Veridi shares the source-evaluation gaming surface.
• Calibrated confidence. Each product produces verbal bands and numeric ranges that map to documented evidence strength, not to a subjective sense of certainty.
• Outcome tracking. v1.3’s calibration feedback loop closes the gap between recommendation and realized outcome.

The detailed Veridi process is described below. Pragma’s process is on the for-policymakers pages; Praxis’s process is on the for-advocates pages.

How Veridi works

The short version

A claim comes in. Veridi classifies it by domain and complexity. It gathers evidence through structured searches. It evaluates each source and assigns it a quality tier. It checks for disinformation techniques. It applies decision trees to determine a verdict. It calibrates confidence against the strength of the evidence. It runs a quality-assurance checklist. It outputs an assessment that shows its reasoning at every step.

The process in detail

1. Claim classification

Every claim is categorized by domain - scientific, legal, medical, financial, electoral, historical, technological, or propaganda - and by complexity:

• Simple: Single verifiable fact, clear sources exist. Resolved directly.
• Moderate: Multiple sub-claims or requires context. Resolved with search, uncertainty flagged.
• Complex: Domain expertise needed, propaganda deconstruction required, or high stakes. Routed to a specialist framework.

2. Source hierarchy

Evidence is ranked on a four-tier system:

The confidence ceiling is structural: a claim supported only by Tier 3 sources cannot receive confidence above 65%, regardless of how many such sources agree. Multiple low-quality sources do not substitute for one high-quality source.

3. Gaming countermeasure scan

Before any verdict is issued, the methodology checks for twelve documented disinformation attack patterns. These range from confidence laundering (multiple outlets tracing back to a single unreliable source) to substrate self-reference (the assessor model evaluating a claim about itself, its developer, or the operator’s own system). Each pattern has defined detection procedures and specific consequences for the assessment.

At Standard tier and above, the assessment surfaces the top three claim-relevant vectors with an explicit assessment of each, rather than a simple detected/not-detected flag. Full and Forensic tiers display all twelve vectors with individual assessments.

Detailed gaming countermeasures

4. Verdict determination

Veridi uses nine verdict categories:

Verdicts are determined using decision trees designed to handle the most confusable distinctions - particularly Misleading vs. Lacks Context, and Mixed vs. Mostly False.

5. Confidence calibration

Confidence is presented as a verbal band — Near-Certain, High, Moderate, Low, or Speculative — rather than a raw integer percentage. Each band corresponds to a range, and the structural ceiling imposed by the source tier is shown alongside it as context (e.g., “High confidence · ceiling: Tier 2 sources only”). This makes it clear both how strong the evidence is and what structural factor limits the rating.

Confidence ratings reflect the strength of the evidence, not certainty about the verdict. A claim can receive a clear verdict (FALSE) with Moderate confidence if the evidence is strong but sourced only from Tier 2.

The methodology also applies field reliability coefficients - disclosed annotations based on published replication rates - to provide context for claims in fields with known reproducibility challenges. A psychology claim and a physics claim supported by similar evidence will receive the same confidence rating, but the psychology assessment will note the field’s lower replication rate.

6. Quality assurance

Before an assessment is published, a checklist is completed covering evidence completeness, source independence, gaming countermeasure application, confidence calibration, and output format compliance.

Confidence accuracy is tracked over time using a Brier score framework. Outcomes are defined as correspondence to external ground truth — election results, court rulings, scientific replications, and similar independently verifiable events — rather than verdict persistence (whether the system would produce the same answer again).

7. Post-generation validation

After the assessment is drafted but before it’s presented, a mandatory validation pass checks the output mechanically: confidence does not exceed the ceiling allowed by the source tier, all required fields are present for the verification tier, government and institutional sources have been checked against the Institutional Reliability Index, and the verdict is consistent with the decision tree logic. If the validator catches a violation, it corrects the assessment in-place and appends a transparent note explaining what was changed and why.

Verification depth

Four tiers of verification are available:

• Quick: Simple claims with clear answers. One authoritative source sufficient.
• Standard: Most claims. Multiple sources, structured search, gaming scan.
• Full: Complex or high-stakes claims. Specialist frameworks, comprehensive sourcing, full gaming countermeasure analysis.
• Forensic: Maximum rigor. Reserved for claims with significant public impact or legal implications. (This label is not intended to imply accreditation or licensing, which the system does not have in any jurisdiction.)

The methodology can auto-escalate: a claim submitted at Standard tier will be bumped to Full if gaming flags are detected during the initial scan.