Search

When a proof-checking kernel completes the checking of a formal proof, that kernel asserts that a specific formula follows from a collection of lemmas within a given logic. We describe a framework in which such an assertion can be made globally so that any other proof assistant willing to trust that kernel can use that assertion without rechecking (or even understanding) the formal proof associated with that assertion. In this framework, we propose to move beyond autarkic proof checkers-i.e., self-sufficient provers that trust proofs only when they are checked by their kernel-to an explicitly non-autarkic setting. This framework must, of course, explicitly track which agents (proof checkers and their operators) are being trusted when a trusting proof checker makes its assertions. We describe how we have integrated this framework into a particular theorem prover while making minor changes to how the prover inputs and outputs text files. This framework has been implemented using off-the-shelf web-based technologies, such as JSON, IPFS, IPLD, and public key cryptography.