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

# Human in the Loop (HIL)

> Human review for biometric cases the agent cannot resolve with certainty.

Human in the Loop (HIL) adds a human operator to the biometric process: it reviews cases where the agent doesn't have enough certainty to approve or reject automatically, before the verification is resolved.

> [!NOTE]
>
> HIL is available **only for Enterprise accounts**. It is not a standard product feature.

## How to activate it

Activate it with the **Enable human review** toggle, in the **Document verification** tab of the Webview configuration. Since HIL is only available for Enterprise accounts and has an additional cost, coordinate its use with your Account Manager.

## What an HIL operator does

Human operators assigned by your company get access to a review platform where they can **approve or reject** cases the biometric agent flags as uncertain, stating the reason, or **reject and send to blacklist** immediately when the case warrants it.

### Coordinated fraud detection

HIL identifies when the same document is linked to multiple phone numbers, with an automatic risk level:

* **High:** three or more clients linked to the same document.
* **Medium:** two clients linked.

The relationship is displayed in a network graph.

### Blacklist management

* Bulk block or removal of up to one hundred users in a single action.
* Support for temporary blocks with a scheduled duration.
* Historical log filterable by company, date, status, and operator.

### Traceability and roles

HIL records every case with who reviewed it, when, and under what criteria. You manage access by role (admin / operator), with segmentation by company and/or country.

> [!NOTE]
>
> HIL can increase the approval rate by **five to fifteen percent** compared to pure automatic rejection, by resolving ambiguous cases that would otherwise be rejected.

## When is a request sent to HIL?

HIL receives a case when the biometric agent detects a signal of doubt that is not enough to reject the process outright.

### By document verification level

In Webview, the main trigger is an error in any parameter set as **required** for the document verification level you configure on the Biometrics node. See the [document verification level](https://docs.jelou.ai/en/guides/integrations/identity/document-analysis-level.md) to review and adjust these parameters.

| Level | Required parameters that route to HIL |
| - | - |
| **Standard** | Document type and origin, expiry date, and physical security elements. |
| **Advanced** | Everything in Standard, plus the document's overall result and, when enabled, the automatic reading zone and document data consistency. |
| **Custom** | Depends entirely on the parameters you set as required for your company. Any required parameter that errors out routes the case to HIL. |

### Other signals that trigger HIL

These signals route a case to human review regardless of the document verification level configured:

* Too many document retries with HIL active (maximum three attempts; liveness has a separate retry limit).
* Screenshot detection — the selfie comes from a screen instead of a live capture.
* Document tampering detection — a substituted photo or modified data.

> [!TIP]
>
> Only parameters marked as **required** trigger an HIL derivation when they fail. Review this configuration on the Biometrics node before enabling HIL to avoid derivations you don't expect.

> [!WARNING]
>
> The system always rejects some cases outright, without going through HIL: failed liveness, facematch below the threshold, underage users, a document that doesn't match the accepted type, or a failed government validation.

## Rejection causes and referral to HIL

### Cases that go to human review

| Failure step | Failure reason | What it means | When it happens |
| - | - | - | - |
| `document_ocr` | `max_attempts` / `validation_failed` | The system couldn't read the document data properly. | Blurry photo, glare, or an uncommon document. |
| `document_recapture` | `max_attempts` / `recapture_detected` | The photo was detected as coming from a screen or a copy. | False positive: laminated document, plastic cover, or strong glare. |
| `document_tampering` | `max_attempts` / `manipulation_detected` | Signs of tampering were detected in the document. | A person confirms its legitimacy before rejecting. |
| `document_government` | `gov_validation_failed` | The government entity didn't validate the data. | Single-attempt query. |
| `document_government` | `gov_unavailable` | The government entity didn't respond. | The official source is down or slow; it is reviewed once retries are exhausted. |
| `document_ocr` | `dni_expired` | The document is expired. | Only if your company requests it. It is enabled case by case and also catches other failures if the document is expired. |

### Cases rejected outright, without going through HIL

* Liveness check (video or gestures).
* Comparison between the selfie and the document photo.
* Expired sessions.
* Business-rule rejections: underage users, a document that isn't allowed, a country that isn't enabled, and inconsistent front and back.

## What HIL does not cover

* Deduplication or identification of synthetic or recycled biometric identities via a 1:N search against the client's full facial database.
* System-level role permission controls (read/write by profile) — currently under evaluation.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.