Where proof is the difference

Testing samples behaviour; verification proves it. Where a wrong answer is irreversible — money moved, inventory lost, a payment captured twice — the difference between "it worked in the tests" and "it holds for every valid state" is the difference between a bug report and a liability.

Money movement

Transfers that conserve funds, forbid overdrafts, and keep every balance non-negative — stated as predicates over the whole state space, not a table of examples.

  • Conservation — total balance is unchanged by every transfer.
  • No overdraft — every balance stays non-negative, before and after.
  • Frame discipline — the contract declares exactly what it reads and writes.

Exactly-once fulfilment

Orders that reserve inventory, capture payment, and emit events atomically — and idempotently, so a retried worker cannot double-charge.

  • Exactly-once — at most one worker observes success for a given order.
  • No lost inventory — a failed reservation consumes nothing.
  • Composable proof — the top-level guarantee is proved from sub-contracts.

Inventory & reservations

Reservation logic where failure leaves the system exactly as it was — safe retries by construction, not by convention.

  • Rollback on failure — no partial state on a failed reserve.
  • Safe retry — a failed attempt is provably invisible to the caller.

Regulatory compliance

Rules your system must never violate, stated as invariants the prover checks against every execution path — not as a checklist you hope the code obeys.

  • Provable invariants — every state value stays in its allowed set.
  • Exception paths — the error cases are specified too, with their triggers.
  • Auditable — the contract is the document you show the auditor.

Vericoding Services

Prove what the program does. Then make it faster.

Models now write more code than you can review line by line. You specify what the program must do. We prove the implementation meets that specification, and where latency or compute cost is the problem, we replace it with a faster program that still does.

See the services
Code checked by a proof, then measured for speed

Latest from the blog

View all posts →
AI-Guided Certified Program Refinement

AI-Guided Certified Program Refinement

Software's core problem is knowing what it should do. AI-guided certified program refinement — vericoding — makes specification the artefact we tinker with, and lets AI guess program transformations that a theorem prover certifies.

From Vibecoding to Vericoding: A Gradient, Not a Jump

From Vibecoding to Vericoding: A Gradient, Not a Jump

You do not have to go from zero to verified in one step. You can start with Gherkin scenarios, graduate to executable contracts with specsaver, and then bring in a theorem prover when you are ready. Contracts are the bridge.