Causal Verification
At its core, Causal Verificationis the process of comparing market speculation (implied probabilities from prediction markets) against ground-truth evidence (verified facts, regulatory filings, and historical documents) to determine if a market's price is driven by fact or speculative hype.
Causal Verification Pipeline Flow
Monitors live prediction pools, orderbooks, trade volumes, and holder concentration weights.
Extracts entity tokens and searches multidimensional spaces to locate nearest claims and document nodes.
Traces multi-hop relationships. Triggers background crawlers if context is sparse.
Cross-checks market price implied probability against documented timelines to flag divergence.
How it Works in Practice
1. Prediction Market Tracking
The ingestion engine monitors live ticks across global prediction pools. It tracks the implied probability derived from trade prices, aggregates order book depth, and tracks the Consensus Sentiment Index (CSI) to measure whale conviction versus retail activity.
2. Vector & Graph RAG Retrieval
Upon receiving a tick, the system extracts the core entity tokens (e.g., specific companies, regulatory agencies, or geopolitical regions). The engine performs a high-dimensional vector search to locate similar claims, documents, and entities within our multi-tenant graph database.
3. Multi-Hop Graph Traversal
To map structural context, the engine executes selective multi-hop relation queries:
If the resulting relations are too sparse, the system's self-reflective JIT ingestion loop dynamically spawns background workers to scrape verified registries, public search engines, and news indexes to refresh the graph on the fly.
4. Divergence Evaluation
The synthesis engine processes the retrieved graph context alongside the prediction market probabilities.
- Speculative Hype: Flagged when a prediction market trades at high probability (e.g., 90%) but the verified graph context shows regulatory postponements, active disputes, or conflicting documentation.
- Aligned: Denotes that market pricing behaves in close correlation with documented, verified timelines.
System Safeguards
Domain Trust Scoring
Maintains domain registers where websites are dynamically graded by authority and fallacy rates. High-fallacy sources are penalized, reducing their weight in the retrieval process.
Immutable History
Every data state is serialized cryptographically. Existing claims are never overwritten; instead, they are superseded, allowing historical audits to track exactly how belief states changed over time.
Empirical Feedback Loop
Enables agents to submit structured feedback on transaction logs. Negative scores prompt immediate graph recalibration, while positive inputs reinforce search path weights.