> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tythe.finance/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent Accountability

> Who answers for an agent's actions, what record every action leaves, what is published, and how responsibility travels down a delegation chain.

An agent can move real money. The question that makes that acceptable is not technical: it is who answers for it afterwards. On Tythe the answer is fixed before the first action, recorded on every action, and readable by anyone. This page is that answer.

### The principal answers

Under the Agent Authorization signed at binding, the principal appoints the agent and accepts responsibility for what it does within any mandate it is given. There is no ambiguity to resolve later.

| Question                                             | Answer                                                                                                    |
| ---------------------------------------------------- | --------------------------------------------------------------------------------------------------------- |
| Who owns the funds an agent spends?                  | The principal. They never leave the principal's account until the moment of payment                       |
| Who is the borrower on a Loan an agent draws?        | The principal                                                                                             |
| Who is liable for a payment made inside the mandate? | The principal. It was authorised in advance, in writing, on-chain                                         |
| Who is responsible for setting the ceiling?          | The principal                                                                                             |
| What is the agent's status in law?                   | Not a party. Its signature is evidence that the principal controls the key, not consent by a legal person |

An agent cannot be sued, cannot hold an obligation, and cannot be a counterparty. Everything it does resolves to a party that can be.

### What the agent answers for

Not liability, but its record. Every action an agent takes becomes a receipt, and the receipts become its Conduct Rating, which decides how much authority it can use in future and what other counterparties can see about it. An agent that behaves badly does not owe money. It loses authority, and the loss is visible.

### The record of an action

Every executed action emits a receipt naming the agent, the mandate it acted under, the rail, the counterparty, the amount, and whether an inflow counts as revenue. Rejections and escalations are recorded too. This is on-chain and permanent.

What that gives each party:

* **The principal** can see what each agent did, under which mandate, at any time, and export it.
* **The agent** can read its own history and the signals behind its rating.
* **Anyone** can verify from the chain that an action was within a mandate the principal had signed, because the mandate and its Delegation Instrument hash are on-chain too.

### The delegation chain

When an agent delegates to a sub-agent, the chain is recorded end to end. Every action a sub-agent takes resolves upward: to the mandate it acted under, to the parent mandate that authorised that one, to the principal that signed the first Agent Authorization.

* Responsibility does not dilute as it descends. The principal answers for the deepest sub-agent exactly as for the first.
* A sub-agent's conduct affects its parent's Conduct Rating as well as its own.
* Revoking a parent revokes everything beneath it in the same transaction.
* A sub-agent's ceiling can never exceed its parent's effective scope, and it spends from the parent's budget, so a chain can never create authority the principal did not grant.

### What is published

Tythe publishes a bounded summary of each agent's Conduct Rating to the ERC-8004 reputation registry, and mandate-compliance facts to the validation registry, so that a counterparty who has never dealt with an agent can still verify whether it is trusted and whether it has operated within its authority.

| Published                                           | Not published                                           |
| --------------------------------------------------- | ------------------------------------------------------- |
| A bounded Conduct Rating summary                    | Raw signals and model inputs                            |
| Mandate-compliance facts                            | The mandate's contents, beyond what is already on-chain |
| The agent's identity and its binding to a principal | The principal's identity, rating, limit, or balances    |

Publication is per agent and the principal can turn it off. Turning it off removes the external signal; it does not change anything inside Tythe.

### Why the record matters to an agent

An agent's history is the only asset it accumulates. It is what raises its effective scope inside a principal's ceiling, what a principal reads when deciding whether to widen that ceiling, and, through the public registries, what a counterparty reads when deciding whether to deal with it at all. An agent with a long clean record on Tythe carries something no new agent can buy.

### Disputes

A principal can dispute a revenue tag or a rating input from the console. A disputed inflow is excluded from capture until it resolves. An agent cannot raise a dispute itself; if an agent believes a rating signal is wrong, that is information for its principal, who can act on it.

<CardGroup cols={2}>
  <Card title="Agent underwriting" icon="gauge-high" href="/agents/underwriting">
    How the record becomes a rating, and the rating becomes authority.
  </Card>

  <Card title="Trust and transparency" icon="shield" href="/get-started/introduction/trust-transparency">
    What is on-chain across the whole protocol, and what can be verified.
  </Card>
</CardGroup>
