> For the complete documentation index, see [llms.txt](https://docs.sail.money/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.sail.money/protocol/concepts/evaluation-guarantees.md).

# The four evaluation guarantees

Permissions are arbitrary, user-deployed Solidity. The kernel does not trust them — it *contains* them. Every permission evaluation, on every dispatch, is bounded by four structural guarantees. Together they mean the worst a buggy or malicious permission can do is **deny** calls, never widen authority or harm the kernel.

## 1. Static evaluation

Permissions are invoked via `staticcall`:

```solidity
(bool success, bytes memory ret) =
    permission.staticcall{gas: PERMISSION_GAS_CAP}(
        abi.encodeCall(IPermission.evaluate, (data, ctx))
    );
```

`staticcall` prohibits state mutation. A permission **cannot** modify any contract's storage during evaluation, and **reentrancy through the permission surface is structurally impossible** — there is no write path to re-enter.

**Protects against:** reentrancy, permissions that try to mutate state mid-evaluation, hidden side effects.

## 2. Gas isolation

Each permission is called with a fixed gas cap, `PERMISSION_GAS_CAP = 150_000`. A permission that exceeds its cap reverts and is treated as returning `false`. (Batch permissions get a larger `BATCH_EVAL_GAS_CAP = 1_000_000` because they must inspect every subcall.)

**Protects against:** a pathological permission denying service to the kernel or burning the manager's entire gas budget. The cap bounds the cost of evaluation no matter what the permission does.

## 3. Selective authorization

The manager's signature names **one** registered permission. The kernel evaluates that permission alone; no other registered permission is consulted. The named permission must be registered on the account — registration is the Permission Signer's trust anchor.

**Protects against:** unrelated permissions falsely denying each other, and lets one account hold many independent mandates. See [the mandate](/protocol/concepts/mandate.md).

## 4. Fail-closed

Any permission that returns `false`, reverts, runs out of gas, or returns malformed data causes the **entire dispatch to revert**. There is no partial-allow path, and an account with no registered permissions cannot dispatch at all (deny-by-default). The kernel decodes the return as a `uint256` and requires it to equal `1`, so a non-canonical boolean word is treated as denial, not acceptance.

**Protects against:** a buggy permission accidentally allowing a call. The default behavior of anything that goes wrong is to deny.

***

These four properties protect the kernel *from* the permission. Everything a permission expresses **inside** that envelope — which tokens, which venues, what amounts — is the permission author's responsibility. That division is what makes the kernel auditable in isolation and the permission set permissionlessly extensible. See [full expressiveness](/protocol/permissions/expressiveness.md).


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.sail.money/protocol/concepts/evaluation-guarantees.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.