How it works
The three checks
Every trade passes three independent gates before it reaches the exchange.
A trade Lloyd proposes is checked three times, by three systems that do not share code, before anything is signed. All three must pass. Any one can refuse.
This is the part of XIO that is hard to build and easy to skip, and it is the reason the limits in your plan behave like limits rather than like advice.
| Gate | Where it runs | What it enforces |
|---|---|---|
| 1. Your plan | XIO, in plain TypeScript | Size, leverage, market universe, exposure, when to stand aside |
| 2. Wallet policy | A secure enclave, with the keys | What the operator may sign at all, and its budget |
| 3. The exchange | Hyperliquid | Margin, tick and lot sizes, position limits, signer permissions |
1. Your plan
Section titled “1. Your plan”The first gate is your Trading Plan, compiled into a typed specification and evaluated by plain TypeScript. Position size, leverage, market universe, how much can be at risk at once, when to stand aside.
No model runs here. The same proposal and the same plan produce the same verdict every time, and the code is unit-tested like any other business logic. A proposal that breaches your plan stops at this gate.
2. The wallet policy
Section titled “2. The wallet policy”The second gate lives with the keys, inside a secure enclave, and it is default-deny: an action is refused unless it matches something explicitly permitted. It governs what the operator wallet is allowed to sign at all — which contracts, which kinds of action, within which budget.
Because it sits at the signing layer rather than in application code, a bug in XIO’s own backend cannot talk it into signing something outside its permissions.
3. The exchange
Section titled “3. The exchange”The third gate is Hyperliquid itself. Margin requirements, tick and lot sizes, position limits, and the permissions attached to the signer. The venue enforces its own rules regardless of what XIO believes.
Where the model sits
Section titled “Where the model sits”Nowhere in that chain.
Lloyd’s job ends when it emits a structured proposal. It can be wrong, it can be confused by a strange market, and it can propose something silly — and none of that reaches your funds, because the proposal is data that three deterministic systems then judge.
That is the failure this design exists to prevent.