Protocol
Security and upgrades
Updated 30 October 20266 min read
Redacted's security model comes down to one rule, which is that nobody can spend what your proofs didn't authorize. Everything below applies that rule.
No backdoors by construction
The contracts were designed so that there is nothing privileged to abuse.
- There is no admin sweep, no owner override and no switch that stops an exit home. The Reserve's message surface is exactly the user actions, a fee-governance path and the node rules, and no admin message can touch users' funds. The node network's brakes act on everyone at once and never touch an exit to your home or rescue address or your own money coming back, and an alarm can slow withdrawals that rest on a proof for at most about 3 days (see How the node network works).
- Bōheki, the DAO's 2-of-3 safety multisig, can pause deposits, switch on incident mode or raise an alarm once, for at most about 3 days. Only the DAO can re-arm it, and it can never touch an exit home. See Brakes that keep your way out open.
- After a break of the proofs or of elliptic-curve signatures, the contract switches into a safe mode for good, and governance can then hold back up to half of each exit, once per coin, and pay it out about 30 days later. It is the one way a vote reaches an exit home, and it never redirects one.
- The core contributors, the DAO and Bōheki can't turn away or send back anyone's money. Only the nodes can, by their own checks and votes, and a waiting deposit only ever goes back to its sender. See How the node network works.
- Every private account is bound to its home wallet. Nobody can redirect a payment home, and in an incident money finds its way home (see Your money can always go home).
- Spending accounts are immutable from the start. They are created without any migration admin and obey only the Reserve acting on your proofs, and nobody, including us, can change their code.
- Token holders decide the fees (the DAO until the token relaunch), but rates are hard-capped at 10% on-chain, and every change needs an on-chain notice of at least a day before it can take effect, so a change can't happen silently.
- The one power that could change these rules is a code upgrade, which waits at least 3 days and is visible in advance, apart from a rescue upgrade that the emergency lane's co-signers run at once. See Upgrades, in the open.
- The relayer carries sealed envelopes and holds no power over them. See Relayer.
Double-spend protection and solvency checks
Every spent note is retired by a nullifier that the contract records forever, so a note can be spent only once. On every outgoing transfer the Reserve also re-checks its own solvency, comparing its recorded obligations with its actual holdings, before it releases anything, so it can't pay out what it doesn't provably hold.
Your side of the model
The protocol keeps your funds safe from everyone else, and two habits keep them safe from bad luck.
- Keep your 25 words, or your backup file and its passphrase, separate and private, because together they are your account, and guard the wallet that opens a wallet account as you always have. See Backup and recovery.
- Use the official links. As with all of DeFi, the genuine frontend matters because your browser is where your keys live, so bookmark the app you reach from redacted.gg and verify links through official channels.
Upgrades, in the open
Every release publishes its contract identities and administration status, so what you verify is what you use. Upgrades never silently move funds, and moving to a new Reserve is always an explicit step that you take yourself.
The contract's code can be replaced only by the Redacted DAO (token holders after the relaunch), and always in public.
Token holders decide upgrades after the token relaunch. Until then the DAO holds their role, as a 2-of-3 group of one core contributor and two independent members, and the handover to token holders is itself a change that waits the same notice.
The DAO schedules an upgrade on an upgrade timelock, which holds it for at least 3 days, with the exact notice published with each release. Every scheduled upgrade also needs hash-based co-signatures from a majority of a published group of co-signers. The timelock lists every scheduled upgrade with the checksum of the new code, so anyone can see it coming, and anyone who disagrees can withdraw during the notice. Until it runs, the DAO can cancel it.
A larger share of the co-signers, set with each release, can run the emergency lane, and it is logged in public. It moves the contract at once to a rescue build that was itself scheduled in public at least one notice period earlier. The emergency lane never chooses the code and can't change who controls upgrades or how long they wait, and replacing the co-signers waits the same notice. It is the one way to skip the notice.
Audits and reviews
An earlier version of the contracts was independently audited by FailSafe. The audit was a full review of the proxy contract, the sub-wallet contracts and the proof-verification package, and it concluded that "the on-chain components align with their intended security model", with every raised issue resolved or consciously acknowledged.
The contracts in use today add the hash-based proofs and signatures, the node network and the upgrade timelock. They went through internal security reviews and an outside review whose findings were each answered.
The on-chain core doesn't change without public notice. If a contract upgrade is ever proposed, it is reviewed again and waits in public before it reaches the chain.
The parts that change more often, such as the interface, the clients and the infrastructure, are under standing adversarial review by purpose-built security models such as FailSafe's GlassBreak and by the latest publicly available frontier AI models, and the tooling is upgraded as the models improve. The aim is that Redacted adds no new attack vector to your life. An audit is a snapshot, and here it covers the part that cannot change without public notice.
Every private action is proved with a STARK, a zero-knowledge proof whose only cryptographic building block is a hash, with no trusted setup at all, and the contract checks it on-chain. Nodes and Spending accounts sign with SLH-DSA, the hash-based signature standard NIST published as FIPS 205. This hash-based design is the same for every account, and bunker mode extends it to your account itself, which is what makes an account quantum-safe.
Verify it yourself
Every security claim in these docs is built to be checked with public data and open tools.
- You can rebuild the contracts. The source is published, so the exact on-chain bytecode can be reproduced deterministically on independent systems and you can compare the hashes yourself.
- You can check the app you open. Every release of the app is pinned with a fixed content hash and published on IPFS, so you can confirm that the app in your browser is exactly the one that was published (see Open the real app).
- You can check a node release. Every node release is published with its checksum, and the node refuses to start if any of its files has changed (see Run a node, step by step).
- You can verify every action. Each proof is checked on-chain, in public, every time, with no privileged verifier and no off-chain judgment call.
- You can verify your own account by rebuilding it from chain data alone, in any browser. If the claim that there is no database were false, this wouldn't work.
For the privacy model, see How privacy works.