---
url: /handbook/governance/rule-packs-and-sod.md
description: >-
  Agent rules, segregation-of-duties rules and mitigating controls, stored as
  versioned packs.
---

# Rule packs and SoD

DutyGraph ships neutral default packs, and a company can load its own. DutyGraph does not decide what a company's rules should be.

## The packs

| Pack | What it holds |
| --- | --- |
| Agent rules | Rules for non-human principals. The default pack has 14, including CEIL-01, CEIL-02 and AGT-12: "No agent approves work." |
| SoD rules | Risks: pairs of business functions one principal must not hold together. Each function is a set of scope patterns. The default pack has 12 risks. |
| Mitigating controls | Controls that allow a conflict to stand, with the principal types each control may cover. |
| System Catalog | See [System Catalog](system-catalog.md). |
| Holdings | See [Holdings](holdings.md). |
| Approval chain | The steps and the people who decide. See [Approval chain](approval-chain.md). |

Every pack is stored as an immutable version. A new load is a new version that supersedes the old one. Every rule is cited as `id@version`. Loading a pack writes a "pack sealed" entry to the [ledger](ledger.md).

## Which agent rules run

Not every rule in the pack is checked by the evaluator in the same way.

| Rule | How the code uses it |
| --- | --- |
| **CEIL-01** | Required. The evaluator removes every scope the owner does not hold. |
| **CEIL-02** | **Not enforced.** It says an agent's scopes never exceed what its Task Cards need. The evaluator does not apply it yet. See [The key join](the-key-join.md). |
| **AGT rules that forbid scopes** | A matching scope is removed, or escalated for a human decision, as the rule says. |
| **AGT-01 and AGT-10** | Lifetime rules. The gateway blocks calls after the end date or the event's end. |

## SoD runs on three sets

1. **The owner alone.** Do the owner's holdings conflict?
2. **The agent alone.** Do the agent's proposed scopes conflict?
3. **The owner plus all their agents.** Do the owner's holdings, the scopes of every active agent they own and the new agent conflict as one set?

The combined check finds a case that person-only SoD misses: a person is clean, and the agent they build creates a violation. An open conflict on the agent's own scopes is escalated.

## Mitigations do not transfer

A mitigating control is assigned to **one principal and one risk**. It does not pass to an agent the person builds. If the agent needs the conflicting scopes, Internal Controls must tag a control for the agent itself, and only a control whose principal types include agents.

A rule or SoD check removes or escalates a scope. Passing them does not show that a scope is needed. That is the approval chain's judgment today.

::: info
**Status.** **Built**: default packs, pack files, versioning, CEIL-01, the forbidding agent rules, the lifetime checks and the three SoD checks. **Simulated**: the sample rulesets, labelled "Saviynt-style, fictional columns" until a real export header is mapped. **Planned**: CEIL-02 enforcement and importing rulesets straight from an IGA product.
:::
