Methodology
Evidence-First Approach
Wize Money operates on an evidence-first principle: no article is published unless its core claims are supported by multiple independent sources with a minimum natural language inference confidence score. We never start from a narrative and then find supporting evidence — we start from verified claims and synthesize them into structured reporting.
Claims classified as YMYL (Your Money or Your Life) — including price predictions, investment outcomes, and regulatory enforcement actions — require a minimum 0.85 NLI confidence score from at least two independent source classes before publication.
How the Pipeline Works
Stage 1: Source Ingestion
Claims are ingested from regulatory feeds (ESMA, SEC, BaFin, CONSOB), wire services (Reuters, Bloomberg), crypto-native media (CoinDesk, The Block), and on-chain data providers (Glassnode, DeFiLlama). Each claim is time-stamped, attributed, and classified by source reliability tier.
Stage 2: NLI Reconciliation
Every claim is cross-referenced against corroborating and contradicting sources using natural language inference. The system identifies entailment, contradiction, and neutral relationships between claims. Contradicting high-confidence claims trigger a manual review flag.
Stage 3: Article Synthesis
Verified claims are assembled into structured articles using one of four templates (Brief, Deep Background, Comparison, Live Timeline). The synthesis agent follows strict editorial rules: no investment advice, no speculation presented as fact, no unverified casualty numbers, no named-individual legal predictions.
Stage 4: Agent-12 Compliance Gate
Before publication, every article passes through Agent-12 — an independent compliance checker that verifies editorial red lines, source attribution completeness, financial disclaimer presence, and AI disclosure metadata. Articles that fail this gate are not published.
Three-Block Article Structure
Every Wize Money article is built around three clearly delineated blocks. This structure ensures readers can always distinguish between established context, verified facts, and editorial analysis:
Establishes context from previously verified information. Helps readers understand the background and significance of the current story.
Contains only NLI-verified claims with full source attribution. This is the factual core of the article — what we know to be true based on available evidence.
AI-generated editorial analysis and forward-looking commentary. Clearly separated from facts and marked with data-nosnippet to prevent search engines from presenting analysis as established fact.
Source Verification Process
Sources are classified into four tiers:
- Tier 1:Regulatory bodies, central banks, government agencies
- Tier 2:Major wire services, established financial media
- Tier 3:Crypto-native media, on-chain data providers, project documentation
- Tier 4:Social media, anonymous sources, unverified community reports (used only for context, never as primary evidence)
Hard Red Lines
The following editorial constraints are enforced by the pipeline and cannot be overridden:
- • No investment advice, buy/sell/hold recommendations, or price targets
- • No unverified casualty or victim numbers
- • No named-individual legal predictions or guilt presumptions
- • No content enabling financial harm, fraud, or market manipulation
- • No presentation of AI analysis as established fact
- • Financial disclaimer required on every article