---
url: /handbook/governance/approval-provisioning-and-revalidation.md
description: >-
  What DutyGraph decides and records, what your identity systems provision, what
  runtimes enforce, and how an approval should be checked again when the company
  changes.
---

# Approval, provisioning and revalidation

An approval in DutyGraph is a recorded decision. It does not create an identity, grant access in a business system or prove what an agent did. This page separates those steps and explains how an approval should be checked again when the company changes.

## The integration stance

DutyGraph complements a company's existing identity infrastructure: directories, identity governance (IGA), IT service management and enterprise agent platforms.

* **DutyGraph** records the business justification and the authorization decision.
* **The company's IAM and provisioning systems** create the agent identity and grant its access.
* **Runtimes and target systems** enforce access when the agent acts.

DutyGraph has no live connection to any identity, IGA or IT service management product today. The optional [agent gateway](agent-gateway.md) is a narrow adapter for calls routed through it. It does not replace enterprise IAM.

## Five states

```mermaid
flowchart LR
    S1["1 Authorization requested<br/>DutyGraph: Built"] --> S2["2 Authorization approved<br/>DutyGraph: Built"]
    S2 --> S3["3 Provisioning requested<br/>For IAM to do card: Built<br/>Decision Export: Planned"]
    S3 --> S4["4 Provisioning confirmed<br/>your IAM<br/>Event Import: Planned"]
    S4 --> S5["5 Runtime activity observed<br/>gateway calls: Built<br/>other paths: Planned"]
    S5 -.->|"revalidation"| S2
```

| State | What it proves | What it does not prove |
| --- | --- | --- |
| 1 Authorization requested | Someone asked, for named Task Cards and scopes | That the access is needed |
| 2 Authorization approved | The approval chain decided, and the ledger recorded it | That any system granted the access |
| 3 Provisioning requested | IAM was told what to create | That it was created |
| 4 Provisioning confirmed | The company's IAM reports it created the access | That every action is checked |
| 5 Runtime activity observed | A recorded call was allowed or blocked | That every action was observed |

## Two contracts

::: info
**Planned.** Both contracts are written down, not built.
:::

**Decision Export** sends DutyGraph's decision to IAM or IT service management. It carries:

* the requesting principal and the accountable owner;
* the task justification: Task Cards and their versions;
* the validated scope requirements;
* the proposed and the approved permissions;
* the approvals, with deciders and times;
* the rules and mitigating controls applied;
* version and evidence references, including ledger entries;
* the effective lifetime.

**Event Import** brings events back from those systems and reconciles each with what DutyGraph approved:

* provisioning completed;
* entitlement changed;
* agent identity created;
* access review completed;
* credentials revoked;
* runtime action blocked;
* owner left.

## Revalidation: the living model

A living operating model means that a change to authoritative information starts a fresh check. Refreshing a file is not enough.

```mermaid
flowchart TB
    HC["Holdings change"] --> F["Find affected<br/>Agent Grants"]
    TC["Task Card changes<br/>materially"] --> F
    EX["Authorization expires"] --> B["Block at every<br/>enforcement point"]
    FU["Future events<br/>owner leaves or changes role,<br/>SoD rules change, mapping changes,<br/>agent or app changes,<br/>unmanaged agent found"] -.-> F
    F --> RC["Recompute against<br/>current evidence and rules"]
    RC --> D{"Still allowed?"}
    D -->|"Yes"| K["Keep, record the check"]
    D -->|"No"| N["Narrow, suspend or<br/>send for reassessment"]
    N --> LG["Record on the ledger"]
    B --> LG
    K --> LG
```

The priority events are:

1. **Holdings change.** Find the grants that depend on the owner's access, recompute them, and narrow or suspend unsafe authority under the company's policy.
2. **A Task Card changes materially.** Grants that depend on the old version become candidates for reassessment.
3. **Authorization expires.** Every supported enforcement point blocks the agent, and the expiry is recorded.

Later events: the owner leaves or changes role, SoD rules change, entitlement mappings change, the agent's implementation changes, an application's configuration changes, or an unmanaged agent is found.

### What exists today

| Mechanism | Status |
| --- | --- |
| Agent builder manifests show **Fresh**, **Out of date** or **Broken** when a cited file, a role holder or recorded access changes | **Built** |
| A pending decision is refused as stale when the access export or the rule pins change | **Built** |
| Each approving decision and the issue step recheck the owner's current holdings | **Built** |
| IAM or Internal Controls can revoke an agent; its credential stops working | **Built** |
| The gateway blocks calls after the agent's end date or event end | **Built** |
| Recompute issued agents when holdings or a Task Card change | **Planned** |
| Suspend an owner's agents and credentials when the owner leaves | **Planned**. Marking a person as left does not touch their agents today. |
| Pin Task Card versions on a request | **Planned**. A request records Task Card ids, not versions. |

New revalidation should reuse the existing audit events, record versions and ledger. It does not need a separate event system.

## The Agent Authority Record

::: info
**Planned.** A concept, not a built screen or table.
:::

The Agent Authority Record is one auditable view of an agent's authority:

* purpose, accountable owner and requester;
* Task Cards and their pinned versions;
* required business capabilities and entitlement mappings;
* the owner's relevant holdings and the scope calculation;
* SoD findings and mitigating controls;
* human decisions and the issued authority;
* the agent identity, its lifecycle state and provisioning status;
* its revalidation history.

It would be a view computed over records DutyGraph already keeps: requests, manifests, scope decisions, issued agents, the ledger and gateway activity. It is not new storage.

## Two agent authority models

| Model | Authority comes from | Status |
| --- | --- | --- |
| **Human-delegated agent** | One named owner. Scopes stay within the owner's holdings and the delegation rules. | **Built**. The current model. |
| **Service-owned agent** | A department or service, with an accountable human sponsor and a separately approved authorization. It need not match any person's entitlements. | **Planned**, as a future consideration. Not supported. |

::: info
**Status.** **Built**: authorization request and approval, the "For IAM to do" card, the gateway's activity record for routed calls, manifest freshness, stale-decision refusal, revocation and expiry checks at the gateway. **Simulated**: provisioning and identity systems in the sample companies. **Planned**: Decision Export, Event Import, provisioning confirmation, broad automatic revalidation, owner-departure suspension, the Agent Authority Record and service-owned agents.
:::
