Why We Stopped Pursuing Digital Voting
Electora spent a significant part of its early research effort on secure digital voting. We are no longer pursuing it. This is an explanation of why, written in more detail than a pivot usually gets, because the reasoning is more interesting than the decision.
What We Were Actually Working On
VoteKey was an architecture for end-to-end verifiable digital balloting: cryptographic ballot casting with individual verifiability, homomorphic tallying, zero-knowledge proofs separating voter identity from ballot content, and a tamper-evident audit record designed to support independent verification by observers without requiring them to trust the system operator.
None of that is novel in isolation. The cryptographic primitives are decades old and well studied. Scantegrity II ran in real municipal elections in Takoma Park in 2009 and 2011. Prêt à Voter, STAR-Vote, and ElectionGuard all explored adjacent territory. The academic foundations are solid. Our contribution was going to be in the integration and the operational envelope. That was the bet.
What We Found
We took the architecture to election officials. The response was not skepticism about the cryptography. In most cases the cryptography never came up. What came up instead:
The threat model does not close.End-to-end verifiability proves that a ballot was recorded and counted as cast. It does not prove that the ballot reflects the voter's intent, because it cannot see the endpoint. A compromised phone or laptop can display one choice and transmit another, and the verification receipt will faithfully confirm the transmitted choice. This is not an implementation gap. It is a property of the problem. Nobody has solved it, and the systems that claim to have solved it have generally solved a narrower problem than the one that matters.
Coercion resistance degrades outside a polling place. The secret ballot is enforced physically. A booth is a place where nobody can watch you. Remove the booth and you remove the enforcement mechanism, and you create conditions for vote buying and household coercion that the paper system was specifically designed to prevent. Cryptographic coercion resistance schemes exist. They require voters to execute deceptive protocols under adversarial observation, which is not a realistic expectation.
Dispute resolution has no digital equivalent. When a recount is contested, the resolution is physical: humans look at marked paper in a room with observers from both parties. Every stakeholder understands what is happening. A cryptographic proof is stronger evidence in a formal sense and weaker evidence in a practical one, because the number of people who can independently verify it is small and the number who will accept it on trust after a contested election is smaller.
Certification timelines exceed product timelines. VVSG 2.0 was adopted in 2021 and became effective for new certifications in late 2023. Certification through an accredited test lab is measured in years and costs are substantial. For a system architecture that does not fit cleanly into existing certification categories, the path is longer and less certain than for a conventional system.
And the demand was not there. This is the plainest finding and it deserves to be stated without hedging. Election officials did not want internet voting. Not at a better price, not with better cryptography, not as a pilot. The National Academies said in 2018 that internet voting should not be used, and the professional consensus has not moved. We were selling into a market that had already decided, on the merits, and had good reasons.
What We Concluded
A system can be technically sound and operationally unviable, and the second condition is dispositive. We could have kept building. Startups do this routinely, and there is a whole vocabulary for it: the market is early, the buyers are conservative, education is needed. Sometimes that vocabulary is accurate. Here it was not. The objections we encountered were correct. Election officials were not failing to understand our architecture. They understood the class of system it belonged to and had reached a defensible conclusion about it, and the endpoint compromise problem in particular has no answer we could have supplied with more engineering. Continuing would have meant persuading people out of a position that better evidence supported. That is not a business we want to be in.
What We Kept
Nearly all of the underlying work transferred, which is why this reads as a redirection rather than an abandonment. The cryptographic lifecycle research became our post-quantum governance work, and the ballot secrecy problem turned out to be the sharpest available illustration of harvest-now-decrypt-later. Ballot secrecy is permanent, which sets the required confidentiality lifetime to infinity, which makes the procurement gap impossible to argue away. That framing came directly out of the voting research.
The infrastructure monitoring and audit workflow design became Infraxion Guard, aimed at water, wastewater, and energy operators, where the buyers have appropriated budget, a regulatory driver, and close to no in-house security capability. The procurement analysis carried over intact. Election technology RFPs and water utility RFPs fail in the same way: cryptographic requirements copied from templates written years earlier, no crypto-agility language, no vendor disclosure obligations, and lock-in periods that exceed the useful life of the cryptography being specified. That is a solvable problem in both sectors.
On Election Security Specifically
We are not leaving the domain. We are leaving one approach within it. The work that helps election administration is unglamorous and mostly not a product: risk-limiting audit support, voter registration database integrity, phishing resistance for county staff, incident response planning for offices with no IT department, and better cryptographic language in procurement. That work has real demand and almost no venture interest, which is an uncomfortable pairing but not a reason to pretend otherwise.
If remote digital voting becomes viable, it will not be because the cryptography improved. The cryptography is already adequate. It will be because the endpoint problem got a real answer, because coercion resistance found a form that ordinary voters can actually execute, and because a dispute resolution mechanism emerged that a losing candidate's supporters will accept. Those are not engineering milestones. Until they arrive, restraint is the correct engineering position, and saying so publicly is more useful than another architecture diagram.