Security and verifiability
Where your vote is encrypted, what the operator can and cannot read, and what can actually be proven when someone disputes the result.
What this system guarantees, and what it asks you to trust
Two different things hold this system up, and they are worth telling apart. One is arithmetic: your ballot is encrypted in your own browser, which makes it unreadable in transit and unreadable to anyone who copies the database without the election key. The other is access control: that key sits on the server, and so does the record of which token was issued to which voter. Somebody who reaches the database directly could therefore read the votes and attach names to them. We would rather write that down than let you assume otherwise.
Everything below is written so that a technical advisor can verify it independently, and so that a non-technical reader can see which claims rest on mathematics and which rest on conduct.
Your vote is sealed before it leaves your device
The ballot is encrypted in your own browser. What travels to the server is already unreadable, and it stays that way in the ballot box.
How the browser does it
Your browser derives a shared secret with the election key (ECDH on curve P-256), stretches it with HKDF, and encrypts the ballot with AES-256-GCM. All of it comes from WebCrypto, the cryptography built into the browser itself.
How the server does it
On the server the same primitives come from libsodium, which ships inside PHP: X25519 key exchange with XSalsa20-Poly1305 sealing, SHA-256 for every published fingerprint, and the operating system CSPRNG for randomness.
What is deliberately absent
No cryptography library is downloaded from a CDN. No external service is called. No algorithm was designed by us.
The two implementations are checked against each other
An automated test feeds the same input to the browser cryptography and the server cryptography and compares the output byte for byte. If the two ever disagreed, the build would fail rather than silently produce ballots nobody can open.
What this costs you
Voting requires JavaScript. There is no way to encrypt something on your device without running code on your device, and we would rather state this limit than send your choice to a server in the clear and call it secure.
What the operator can reach, and what nobody else can
An earlier version of this system kept the voter list and the ballot box structurally apart, so that not even the operator could join them. That is no longer the case, and pretending otherwise would be the worst thing this page could do. Administrators need to see who has not voted yet and to replace a token somebody has lost, and both of those require knowing whose token is whose. So the record is kept for the whole election, and the operator can join the two.
| Who is eligible to vote | Stored, and published as a frozen list with a fingerprint so nobody can add names later. |
| Which ballot you cast | Recorded. Your ballot carries a credential reference, and the system keeps which credential was issued to you, so the two can be joined by anyone with direct database access. |
| Your IP address | Kept only as an HMAC fingerprint with an election-specific secret, never in the clear. |
| The contents of your ballot | Encrypted on your device. Readable to the operator, who holds the decryption key; unreadable to anyone else who obtains the database. |
The key sits on the server, and that is a deliberate trade
An earlier design split the key between three people so that no interim result could exist. That guarantee was real, and so was its cost: a lost share could void an election, and every count needed a ceremony with three participants.
This version keeps the key with the election record. Ballots are still encrypted on your device and still stored sealed, so anyone who steals the database without the key reads nothing.
What changed is who can open the box. The operator can, at any time, including while voting is open. No screen shows an interim tally and the software will not produce one, but that is a decision about conduct, not a fact about mathematics.
Key splitting is not available in this version at all. For a board vote among people who already trust their own secretariat, the simpler arrangement is usually the right one. If your election needs a guarantee that no interim result can exist even in principle, tell us before it is created and we will say honestly whether this is the right system for it.
What can actually be proven, and by whom
This is the part that matters in a real disagreement. Each row is a claim someone might make, the evidence that answers it, and the person who can check that evidence.
| The claim | The evidence that answers it | Who can check it |
|---|---|---|
| My vote was not counted | Look your own tracking code up on the election's lookup page. If it is there, your ballot is in the box. No sign-in, no request to anybody, and you can repeat it whenever you like. | The voter, alone |
| The result was altered after closing | The ballot-box fingerprint recorded at the moment of closing, before any result was opened, and the event log entry that timestamps it. | An auditor with access |
| I never voted, but I am shown as having voted | The Tokens screen records whether your token was used, and the event log timestamps every ballot as it arrives. If your token is marked used and you did not vote, that is a real finding, and the log narrows down when it happened. | An auditor with access |
| Ballots were added or removed | The board records every ballot with its tracking code as it arrives. The number of entries is checkable against the turnout and against the frozen roll. | An auditor with access |
| The event log was edited | Every entry carries the hash of the entry before it. Altering one breaks every entry that follows. | A technical auditor |
| The software was changed quietly during the election | The published version and source-code hash, frozen when the election was created. | A technical auditor |
Three mechanisms, and the first one does most of the work
-
The roll is frozen before any token is issued
The list of who may vote is locked and its fingerprint published before voting starts. That ordering is what answers "names were added later"; done afterwards, the same fingerprint would prove nothing.
-
The board is written as votes arrive
Every ballot is recorded with its tracking code at the moment it is cast, not assembled afterwards. A list produced after the fact proves nothing, because it could have been produced to order.
-
The event log is chained
Every entry carries the hash of the entry before it. Inserting, removing or altering one record breaks every record that follows, which is exactly the property a tamper attempt cannot route around.
What this system cannot prove
A voting system earns trust by being precise about where its guarantees stop. These are ours.
Stated plainly, before anyone has to discover them
- The operator holds the decryption key and could read the votes before closing. Nothing in the software prevents this; what prevents it is that no screen offers it and every action is logged. Key splitting is not available in this version.
- The system records which token was issued to which voter and keeps that record for the whole election. It is what lets an administrator see who has not voted and replace a lost token. It also means that somebody with direct database access could work out how a named person voted. This is the largest single thing the system asks you to trust, and it belongs in your election rules.
- Verification only works if people actually look. A board nobody reads proves nothing, which is why a rehearsal where every voter finds their own tracking code is a required step, not a nice extra.
- You can prove your ballot was recorded and was not altered. What you cannot prove by mathematics alone is that the code in your browser sealed the candidate you picked. A strict content security policy reduces this; it does not end it.
- In a small electorate a unanimous outcome reveals how everyone voted. That is arithmetic, not a flaw, and no system can prevent it. It belongs in the election rules.
Not our own scoring system
The design is mapped clause by clause against two external standards, including the clauses it does not fully meet.
Council of Europe CM/Rec(2017)5
The European recommendation on standards for e-voting: equal, free and secret suffrage; roll integrity; individual verifiability; transparency; accountability. Met except for the clauses we name: voting needs JavaScript, there is no automatic failover, cast-as-intended is not mathematical, the decryption key is held by the operator, and ballot secrecy holds against outsiders rather than against the operator.
OWASP ASVS v4
The application security verification standard: authentication, session management, access control, cryptography at rest and in transit, error handling and logging. The controls that are deliberately not implemented are written down rather than left blank.
Checked on the running system
A built-in security check inspects the live installation and reports findings, including the fact that the decryption key sits on the server. It is meant to be run before an election opens, and it does not grade on a curve.
What is checkable, and by whom
A voter can check their own ballot without help. Auditing the whole election is a different matter, and this version keeps the records without opening them to the public. Be clear about both halves when you set expectations.
A voter keeps their tracking code and looks it up themselves, on a page that needs no sign-in. Nobody has to be asked and nobody can refuse. The code reveals nothing about the choice, so producing it costs the voter nothing.
Beyond that, everyone sees one address: the meeting screen, which shows no figures at all until the result is opened. Auditing the board, the roll and the chained log requires access granted by the organisation running the election.