The Trust Model
QWED is built on a fundamental insight:LLMs are probabilistic translators, not reasoning engines.
How Translation Works
This is the key to understanding QWED. Letβs walk through a real example:Example: Verifying a Mathematical Claim
User asks: βIs it true that the sum of interior angles in a triangle equals 180 degrees?βTranslation Examples
| Natural Language | QWED-Logic DSL | Engine |
|---|---|---|
| βx is greater than 5β | (GT x 5) | Z3 |
| βIs 2+2 equal to 5?β | (EQ (PLUS 2 2) 5) | SymPy |
| βx squared plus y squared = 25β | (EQ (PLUS (POW x 2) (POW y 2)) 25) | Z3 |
| βIf raining, take umbrellaβ | (IMPLIES raining umbrella) | Z3 |
Key Principles
1. Determinism over Probability
QWED uses symbolic engines (Z3, SymPy) that provide mathematical guarantees:| Engine | Technology | Guarantee |
|---|---|---|
| Math | SymPy | Algebraic correctness |
| Logic | Z3 | SAT/UNSAT proof |
| Code | AST | Pattern matching |
| SQL | Parser | Syntax validity |
2. Verification, Not Generation
QWED doesnβt generate answersβit verifies them:3. Transparent Proofs
Every verification produces a verifiable proof:Verification Statuses
| Status | Meaning |
|---|---|
VERIFIED | Claim is correct, proof generated |
FAILED | Claim is incorrect |
CORRECTED | Claim was wrong, correction provided |
BLOCKED | Security violation detected |
ERROR | Verification engine failed |
Attestations
QWED can produce cryptographic attestations (signed JWTs) that prove a verification occurred:- Embedded in documents
- Stored on blockchain
- Verified by third parties