Navigation

Account
Loading account...
🧮 Formal Logic

Not "That Sounds Right." Proven, or Disproven.

Most reasoning tools return a confident-sounding paragraph. Formal Logic returns a verdict with a witness — a concrete model or counterexample. This is where the model stops guessing and the math answers.

"Plausible" Is Not "Correct"

Language models are pattern matchers. Ask whether a conclusion follows from your premises, or whether a set of rules can all hold at once, and you get a fluent answer that is sometimes wrong in ways that read exactly like being right.

Some questions don't want a probability. They want a proof.

Formal Logic routes those questions to a deterministic solver. Same input, same verdict, every time — with the model or counterexample that justifies it.

How It Actually Works

The upstream LLM translates your question into formal statements, then invokes a solver that returns a definite answer:

tools = [
    entails,             # Do the conclusions follow from the premises?
    is_satisfiable,      # Can these constraints all hold at once?
    check_consistency,   # Does this set of rules contradict itself?
    fol_prove,           # First-order / propositional proof
    ltl_check,           # Temporal properties over sequences
]

The result is a verdict — entailed / not entailed, satisfiable / unsatisfiable — accompanied by a witness: the assignment that makes it work, or the counterexample that breaks it. No "I think so."

When to Reach for This

  • A claim needs to be proven or disproven, not estimated
  • You want to check a set of requirements for contradictions
  • Correctness matters more than a plausible-sounding answer
  • You're verifying that a spec actually entails what you think it does

Use Cases

Requirement Consistency

Prove a set of rules, constraints, or ADRs doesn't contradict itself before you build on top of them.

Policy & Access Logic

Verify that authorization rules entail the access you intend — and provably forbid what they should.

Spec Verification

Confirm a design actually satisfies the properties it claims, with a counterexample when it doesn't.

Your AI Estimates. This One Proves.

$20/month for entailment, satisfiability, and temporal-logic checks that return a verdict — not a confidence score.