Verification Engines

QWED provides 8 specialized verification engines, each using deterministic methods first and LLM only as fallback.

Engine Overview

Engine	Technology	Key Features
Math	SymPy + Decimal	Calculus, Matrix ops, NPV/IRR, Statistics
Logic	Z3 Theorem Prover	ForAll/Exists quantifiers, BitVectors, Arrays
Code	Multi-Lang AST	Python, JavaScript, Java, Go security analysis
SQL	SQLGlot AST	Complexity limits, Cost estimation, Schema validation
Stats	Wasm/Docker Sandbox	Secure code execution with AST validation
Fact	TF-IDF + NLP	Semantic similarity, Entity matching, Citations
Image	Deterministic + VLM	Metadata extraction, Size verification, Multi-VLM
Reasoning	Multi-LLM + Cache	Chain-of-thought validation, Result caching

Deterministic-First Philosophy

All engines now follow a deterministic-first approach:

1. Try deterministic methods first (100% reproducible)
2. Fall back to LLM only when necessary
3. Discount LLM confidence when used

# Example: Fact verification is now deterministic!
result = client.verify_fact(
    claim="Paris is in France",
    context="Paris is the capital of France."
)
# Uses TF-IDF similarity + entity matching
# No LLM needed for most claims!

Engine Selection

QWED auto-detects the appropriate engine:

Content Pattern	Detected Engine
2+2=4, sqrt(16), derivative	Math
(AND ...), ForAll, Exists	Logic
SELECT, INSERT, DROP	SQL
```python, import, function	Code
Claims with context	Fact
Image bytes + claim	Image

Or specify explicitly:

result = client.verify(query, type="math")

Engine Documentation

Deterministic Engines

• Math Engine - Calculus, Matrix, Financial
• Logic Engine - Quantifiers, Theorem Proving
• Code Engine - Multi-language Security
• SQL Engine - Complexity Limits

Data Verification Engines

• Stats Engine - Wasm Sandbox Execution
• Fact Engine - TF-IDF + Citations
• Image Engine - Deterministic Verification

Orchestration Engines

• Reasoning Engine - Multi-LLM Validation
• Consensus Engine - Parallel Execution