The problemWhy certificates break trust
The solutionYour keys, checked by many
Who runs itNo single party in charge
Questions & answersShort answers, no jargon
Everything this site says in plain language rests on work you'll be able to check: a formal paper, machine-verified proofs, a reference implementation, and models that regenerate every number.
The full design is written up as an academic paper — “ConsensusPKI: Removing Unilateral CA Authority at Post-Quantum Signature Sizes” — currently being finalised for submission to a peer-reviewed security conference.
Following standard academic practice, the paper will be published on this page once it has been submitted. The design it describes supersedes a 2019 master's thesis by the same author; the paper documents, candidly, what that earlier design got wrong and why the new one is different.
Domain-authorised key sets (prevention, not detection) · unpredictable multi-vantage validation · witness-cosigned transparency with majority quorums · offline, private verification · and post-quantum operation with no per-connection penalty.
The two load-bearing security properties are machine-verified with the Tamarin prover — the standard tool for symbolic protocol verification. All seven lemmas across two protocol models are discharged automatically, with no manual proof steps.
For an established domain that has disabled remote recovery, no coalition of validators — even all of them — can cause a browser to accept a key the owner never signed. Machine-checked, including the subtle bootstrap conditions that make the claim honest.
Two different versions of the public record for the same hour cannot both be accepted unless a majority-breaking number of witnesses is compromised — the quorum-intersection property, machine-checked.
All four roles — owner tooling, validator, log, witness — implemented with a test suite of 110+ cases, every one of the design's "must reject" conditions included. The repository will be opened alongside the paper.
Every byte of every message is specified, and public test vectors lock the encoding: any change to the format breaks the build loudly instead of drifting silently.
Every size and cost figure — including the post-quantum ones — is generated by a published model from the specified field widths. Nothing on this site is a hand-typed estimate.
Now — reference implementation complete; formal models verified; measurement campaign (network-level experiments) in progress. Next — conference submission, after which the full paper, the repository and the artifact bundle appear on this page. Then — pilots with interested organisations.