---
title: "Document Processing Flow"
slug: "document-processing-flow"
description: "Configure Hyperscience's Document Processing flow. Use different subflows, such as ORCA Subflow or standard IDP subflow to process documents. "
updated: 2026-09-15T11:09:03Z
published: 2026-09-15T11:09:03Z
canonical: "help.hyperscience.ai/document-processing-flow"
---

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

# Document Processing Flow

Use the **Document Processing** flow to configure how Hyperscience processes submissions.

In v43, the editable Top-Level Flow calls **Document Processing Subflow V43** by default.

- To use ORCA, select **Document Processing with ORCA Subflow V43** from the **Flow Identifier** in the **Start Document Processing Subflow Block**.
- Custom flows may expose different settings, depending on the blocks they contain. For block-level settings, see [Flow Blocks](/v43/docs/flow-blocks).

## Understand the flow group

The v43 flow group contains the following flows:

- **Document Processing** — The editable top-level flow. It contains an Input Block, a block that calls the selected document-processing subflow, and an Output Block.
- **Document Processing Subflow V43** — The read-only standard subflow. It contains the standard document-processing blocks, except for the Input Block and Output Block.
- **Document Processing with ORCA Subflow V43** — The read-only alternative for processing different document types, such as semi-structured or unstructured, with an installed ORCA VLM. Configure the subflow through the **Start Document Processing Subflow Block** in the editable Document Processing top-level flow.
- **Submission State Notifications** — The subflow used to send document-processing state updates to downstream systems.

## Choose a document-processing subflow

Select the subflow that matches the documents and extraction method used by your flow.

| Subflow | Use this subflow for | Notes |
| --- | --- | --- |
| **Document Processing Subflow V43** | Structured and Semi-structured documents that use the standard Identification and Transcription process. | The subflow is read-only. Configure its exposed settings from the **Start Document Processing Subflow Block** in the top-level flow. |
| **Document Processing with ORCA Subflow V43** | Semi-structured documents that use ORCA VLM field extraction. | **The subflow is read-only.** Configure its exposed settings from the **Start Document Processing Subflow Block** in the top-level flow. - It requires GPU-enabled application machines, - an installed ORCA base model, - and a release that contains at least one Semi-structured Layout with fields. - It does not process Structured documents. For more information, contact your Hyperscience representative. - ORCA QA is not enabled by default. |

**Table extraction in v43.2 and later**

Use the separate **Table Extraction Flow** to extract tables as rows and columns with ORCA 2. This flow does not require a specialized model and is not selected from the **Flow Identifier** setting in the **Document Processing** flow. Learn more in [General Prompting Block](/v43/docs/using-the-general-prompting-block).

For ORCA installation and infrastructure requirements, see [Installing ORCA VLMs](/v43/docs/installing-orca-vlms).

## Configure the flow

Select and configure the document-processing subflow from the editable **Document Processing** top-level flow.

1. Go to **Flows**, find the **Document Processing** flow, and click its name.
2. Click **Edit Flows**.
3. In Flow Studio, open **Flow Settings** and select a release from the **Layout Release** drop-down. The flow cannot be enabled without a release.
4. On the Flow Studio canvas, click the **Start Document Processing Subflow Block**.
5. In the right-hand panel, under **Block Details**, open the **Flow Identifier** drop-down list.
6. Select the subflow:
  - Select **Document Processing Subflow V43** to use the standard document-processing flow.
  - Select **Document Processing with ORCA Subflow V43** to use ORCA for field extraction.
7. From **Settings Type**, select a setting group.
8. Update the settings. Repeat the previous step for each setting group you need to configure.
9. Click **Save**.

**Duplicate the flow**

We recommend duplicating and renaming the flow.

**Flow Identifier** determines which document-processing subflow the top-level flow calls. **Settings Type** determines which settings from the selected subflow appear in the panel.

## Configure the top-level blocks

The editable Document Processing flow contains an **Input Block**, the selected Document Processing subflow, and an **Output Block**. Configure the top-level blocks separately from the subflow settings.

### Input Block

The Input Block displays the following submission methods. In the current v43.2 implementation, both fields are read-only and always enabled.

| Field | Behavior |
| --- | --- |
| **Allow API submissions** | All flows can be started through the Hyperscience API. This field is displayed as enabled and cannot be changed in Flow Studio. |
| **Allow manual submissions** | Submissions can be uploaded manually from the Submissions page. This field is displayed as enabled and cannot be changed in Flow Studio. |

**Start the flow from an email inbox:** Add an [Email Listener](/v43/docs/email-listener) input connection to monitor an inbox and submit incoming email bodies and attachments to the flow.

### Output Block

Use an Output Block to send completed submission data to a downstream destination. For the available output options, see [Output Blocks](/v43/docs/flow-blocks#output-blocks).

## Find the settings you need

The selected subflow determines which setting groups appear.

| Available in | Setting groups |
| --- | --- |
| Both subflows | **File Filter**, **Submission Bootstrap**, **Collation**, **Classification**, **Address Parsing**, **Flexible Extraction**, **Document Rendering**, and **Reprocessing** |
| Standard subflow only | **Identification**, **General Transcription**, **Structured Document Transcription**, and **Semi-Structured Document Transcription** |
| ORCA subflow only | **ORCA** |

**Address Parsing** is available in v43.2 and later. Other v43.2 changes are identified in the applicable setting descriptions.

## Settings shared by both subflows

The following setting groups are available with both **Document Processing Subflow V43** and **Document Processing with ORCA Subflow V43**.

### File Filter

Use these settings to remove files from processing based on file type, image dimensions, file size, or extension.

**Placement and output**

The File Filter Block is already placed in the standard Document Processing subflows. In a custom flow, place it after the Input Block and before the Submission Bootstrap Block. Its output identifies each excluded file and the filters that excluded it. For Email Listener inputs, the output also includes the file name.

| Setting | Description | Default value |
| --- | --- | --- |
| **All Files or Images Only** | Choose whether the remaining filters apply to all files or only to files with an image MIME type. | **Apply to all files** |
| **Minimum Image Width (px)** | Minimum image width, in pixels. The filter removes smaller images and does not affect non-image files. | Blank |
| **Minimum Image Height (px)** | Minimum image height, in pixels. The filter removes smaller images and does not affect non-image files. | Blank |
| **Minimum File Size (KB)** | Minimum file size, in kilobytes. The filter removes smaller files. | Blank |
| **File Extension Action** | Choose whether to skip extension filtering, allow only selected extensions, or deny selected extensions. | **Do not filter files by extension** |
| **File Extensions** | Select the extensions to allow or deny, based on **File Extension Action**. If you select **zip**, the filter treats each ZIP file as one file and does not decompress its contents. Select **other** to enter extensions that are not listed. This setting appears when **Allow only these file extensions** or **Deny files with these extensions** is selected. | (Does not appear) |
| **Other File Extensions** | Enter a comma-separated list of extensions that are not available in **File Extensions**. This setting appears when **Other** is selected. | (Does not appear) |

### Submission Bootstrap

Use these settings to configure how the flow retrieves files and initializes submissions. Configure only the retrieval store and authentication method used by your flow.

**Attached email files in v43.2 and later**

The Email Listener passes an attached `EML` or `MSG` file to the flow as a single file. The Submission Bootstrap Block then extracts the nested email body and its supported attachments. The standard flow processes one nested email level. To process emails nested more deeply, add a Custom Code Block before the Submission Bootstrap Block. See [How do I access an email stored by the Email Listener?](https://flows-sdk.hyperscience.ai/pages/examples.html#how-do-i-access-an-email-stored-by-the-email-listener) in the Flows SDK documentation.

After you change a retrieval-store configuration, click **Save**, and then click **Save & Deploy**.

#### AWS S3

Use the following settings when files are stored in Amazon S3.

| Setting | When to use it | Description |
| --- | --- | --- |
| **S3 Submission Retrieval Store Configuration** | You use an S3 bucket and do not authenticate through IAM roles. | Click **Edit value**, enter the access key ID and secret access key as JSON, and click **Done**: ```json { "aws_access_key_id": "<your_access_key_id>", "aws_secret_access_key": "<your_secret_key>" } ``` To authenticate with AWS Signature Version 2, also add: ```json "s3_signature_version": "s3" ``` |
| **S3 Submission Retrieval Endpoint URL** | Your S3 retrieval store is not in the public cloud, such as a government cloud or an S3-compatible internal deployment. | Enter the endpoint URL. Leave this setting blank for buckets in the public cloud. You do not need to edit the “.env” file. |

For more information about AWS credentials, see Amazon’s [Understanding and getting your AWS credentials](https://docs.aws.amazon.com/general/latest/gr/aws-sec-cred-types.html).

#### OCS

Use these settings only when files are stored in OCS.

**Deprecated in v43.1:** Oracle Content Server (OCS) support is planned for removal in v46.0. Replace `ocs://` paths with Amazon S3, Azure Blob Storage, GCS, or Generic Web Storage (HTTP/HTTPS).

| Name | Required? | Description |
| --- | --- | --- |
| **OCS Host URL** | Yes, when you use an OCS submission retrieval store | Host URL for the OCS submission retrieval store. |
| **OCS Username** | Yes, when you use an OCS submission retrieval store | Username used to authenticate with the OCS submission retrieval store. |
| **OCS Password** | Yes, when you use an OCS submission retrieval store | Password used to authenticate with the OCS submission retrieval store. |
| **OCS SSL Certificate** | Yes, when you use an OCS submission retrieval store | CA bundle filename in the `$HS_PATH/certs` directory. Enter `SKIP` to disable certificate validation. |

#### Generic Web Storage (HTTP/HTTPS)

Use Basic Authentication or OAuth client credentials to retrieve protected files over HTTP or HTTPS.

| Name | Required? | Description |
| --- | --- | --- |
| **Generic Web Storage (HTTP/HTTPS) Configuration** | No | Credentials and certificate information for Basic Authentication. Enter the values as JSON: ```json { "username": "<your_username>", "password": "<your_password>", "ssl_cert": "<CA_bundle_filename_OR_SKIP>" } ``` The `ssl_cert` value is the CA bundle filename in the `$HS_PATH/certs` directory. Enter `SKIP` to disable certificate validation. |
| **HTTP/HTTPS OAuth Token URL** | No | OAuth token endpoint used for client-credentials authentication. |
| **HTTP/HTTPS OAuth Client ID** | No | Client ID used for OAuth-authenticated HTTP/HTTPS file downloads. |
| **HTTP/HTTPS OAuth Client Secret** | No | Client secret used for OAuth-authenticated HTTP/HTTPS file downloads. |
| **HTTP/HTTPS OAuth Scope** | No | Optional OAuth scope sent with the client-credentials token request. |

#### Azure Blob Storage

From **Azure Blob Storage Authentication Type**, select **SAS Token Only**, **Service Principal**, **Managed Identity**, or **Account Key**. The settings for the selected authentication type then appear.

| Authentication type | Setting | Required? | Description |
| --- | --- | --- | --- |
| SAS Token Only | **Azure Blob Storage Account URL** | Yes | Storage-account URL, such as *https://<account_name>.blob.core.windows.net*. |
| Service Principal | **Azure Blob Storage Account URL** | Yes | Storage-account URL, such as *https://<account_name>.blob.core.windows.net*. |
| Service Principal | **Azure Blob Storage Tenant ID** | No | Tenant ID of the service principal. |
| Service Principal | **Azure Blob Storage Client ID** | No | Client ID of the service principal. If more than one client ID exists and you leave this setting blank, Hyperscience uses the default client ID. |
| Service Principal | **Azure Blob Storage Client Secret** | No | Client secret of the service principal. |
| Service Principal | **Azure Blob Storage Authority Host** | No | Microsoft Entra authority host for the storage account. If you leave this setting blank, Hyperscience uses the Azure Public Cloud authority, *login.microsoftonline.com*. For valid values, see Microsoft’s [azure.identity.AzureAuthorityHosts class](https://learn.microsoft.com/en-us/python/api/azure-identity/azure.identity.azureauthorityhosts?view=azure-python). |
| Managed Identity | **Azure Blob Storage Account URL** | Yes | Storage-account URL, such as *https://<account_name>.blob.core.windows.net*. |
| Managed Identity | **Azure Blob Storage Client ID** | No | Client ID of the managed identity. If more than one client ID exists and you leave this setting blank, Hyperscience uses the default client ID. |
| Account Key | **Azure Blob Storage Account URL** | Yes | Storage-account URL, such as *https://<account_name>.blob.core.windows.net*. |
| Account Key | **Azure Blob Storage Account Key** | No | Access key for the storage account. |
| Account Key | **Azure Blob Storage Account Name** | No | Name of the storage account. |

If the authentication information is incorrect, file ingestion fails. The flow-run output contains the error returned by Azure. For help reviewing the error, see [Testing and Debugging Flows](/v43/docs/testing-and-debugging-flows).

#### GCS Storage

Use one of the following authentication methods when files are stored in a GCS bucket.

| Name | Required? | Description |
| --- | --- | --- |
| **Use GCP Workload Identity** | Enabled by default. Turn it off if you provide **GCP Service Account JSON**. | Uses credentials obtained through Workload Identity Federation. This option applies to Hyperscience installations in GKE clusters. |
| **GCP Service Account JSON** | Yes, when **Use GCP Workload Identity** is turned off | Service-account JSON credential used to access the retrieval-store bucket. Click **Edit value**, enter valid JSON, and click **Done**. |

#### Other settings

Use these settings to configure submission-initialization notifications and the per-file page limit.

| Setting | Description | Default value |
| --- | --- | --- |
| **Submission Initialization Notification Flow** | Notification flow called during submission initialization. | **Submission State Notifications** |
| **Enable file page-limit check** | Turn on this setting to fail a submission when any file exceeds the configured page limit. | **Disabled** |
| **Maximum pages allowed per file** | Maximum number of pages allowed in each file. This setting appears when **Enable file page-limit check** is turned on. | 100 |

### Collation

Use these settings to control how files update cases and how long cases are retained.

| Setting | Description | Default value |
| --- | --- | --- |
| **Replace Case Data From Duplicate File Names** | Turn on this setting to keep case data from the most recently submitted file when a file added to a case has the same name as a file already in the case. The existing file’s data is not deleted; it is removed only from the case. This setting applies only when files are added to cases. | Disabled |
| **Retention Period** | Number of days to retain a case after this block updates it. Leave the setting blank to keep the case’s current deletion date. | (Blank) |
| **Refresh Retention Period** | Turn on this setting to apply **Retention Period** to every case processed by the block. When it is off, the block updates only cases that do not already have a deletion date. | Enabled |

### Classification

Use these settings to configure machine and manual classification, Classification QA, and image-processing options.

**v43.2 and later**

Faster PDF Transcription supports additional file types, such as `.docx`, `.xlsx`, `.eml`, `.msg`, `.html`, and `.xps`. It can now be enabled together with Image Correction. For files containing digital and scanned pages, embedded text is used where available and Image correction is applied to the remaining pages.

| **Setting** | Description | Default value | Applies to |
| --- | --- | --- | --- |
| **Structured Layout Match Threshold** | Minimum confidence required to match a Structured page to a layout variation automatically. | 0.6 | Both |
| **Semi-Structured Classification** | Turn on this setting to classify Semi-structured and Additional documents automatically. | **Enabled** | Both |
| **Manual Classification Supervision** | Turn on this setting to create a Classification Supervision task when machine classification cannot complete the task. | **Disabled** | Both |
| **Semi-Structured Classification Target Accuracy** | Target accuracy for machine classification of Semi-structured documents. | 99 | Both |
| **Semi-Structured Classification Grouping Logic** | Controls how consecutive pages matched to a layout variation are grouped. | **Consecutive pages as a document** | Both |
| **Semi-Structured Classification QA Sample Rate** | Percentage of documents selected at random for Classification QA. | 5 | Both |
| **Image correction** | Detects and corrects page orientation. | **Enabled** | Both |
| **Mobile processing** | Improves machine readability for documents captured with a mobile device. | **Disabled** | Both |
| **Faster PDF transcription** | Uses embedded text to reduce transcription time. In v43.2 and later, this setting supports DOCX, XLSX, EML, MSG, HTML, and XPS files and can be used with **Image correction**. For files that contain both digital and scanned pages, Hyperscience uses embedded text where it is available and applies **Image correction** to the remaining pages. | **Disabled** | Both |
| **Validate classification using LayoutID** | Compares the document’s Form ID with the layout’s Layout ID to validate a Structured-document match. | **Disabled** | Both |
| **Bypass validation if LayoutID missing** | Turn on this setting to skip LayoutID validation when the matched layout page does not have a Layout ID. | **Disabled** | Both |
| **Manual Classification Notification Flow** | Notification flow called when a Classification Supervision task is required. | **Submission State Notifications** | Both |
| **Default Task Restrictions** | Controls which users can access Classification Supervision tasks. | None selected | Both |

### Address Parsing

In v43.2 and later, use these settings to parse address transcriptions into structured components and standardize their output.

**Available in v43.2 and later**

Address parsing converts address transcriptions into consistent components, such as house number, street, city, state, postal code, and other address elements. In the standard subflow, parsed addresses appear in Machine Transcription, Manual Transcription, and Flexible Extraction outputs. In the ORCA subflow, the settings apply to Flexible Extraction output.

| **Setting** | Description | Default value |
| --- | --- | --- |
| **Address Parsing Enabled** | Turn on this setting to parse and standardize address-typed fields after transcription. When it is off, address fields remain plain transcription strings. Address parsing may increase memory usage and require additional infrastructure resources. | **Disabled** |
| **Address Parsing Provider** | Select the local US and international parsers or a supported external provider: **Google Address Validation API**, **Smarty US Extract API**, or **Geoapify Geocoding API**. All supported parsers return address components in a consistent format. | **Local only (no external API)** |
| **API Key** | Credential for the selected external provider. This setting is hidden when the local parser is selected. | (Blank) |
| **Provider Options** | Provider-specific request options entered as a JSON object. The options use the parameter names defined by the selected provider. | `{}` |

### Flexible Extraction

Use these settings to configure Flexible Extraction tasks, access, and notifications.

**v43.2 and later**

Users can search for text segments in Flexible Extraction tasks and select one or more segments to populate a field. Unsupported characters are filtered according to the field’s data type, and selecting checkbox or signature segments updates those fields automatically.

| **Setting** | Description | Default value |
| --- | --- | --- |
| **Flexible Extraction Notification Flow** | Notification flow called when a Flexible Extraction task is required. | **Submission State Notifications** |
| **Flexible Extraction Transcription Masking** | Prevents users from entering characters that are not valid for the field’s data type. | **Enabled** |
| **Default Task Restrictions** | Controls which users can access Flexible Extraction tasks. | None selected |
| **Flexible Extraction Show Machine Predictions** | Pre-populates low-confidence machine transcriptions in Flexible Extraction tasks for human review. | **Enabled** |
| **Enable Text Segment Search** | In v43.2 and later, turn on this setting to search for text in Flexible Extraction tasks. Matching text is highlighted in the document. | **Disabled** |

### Document Rendering

Use these settings to create and format a PDF as part of the completed submission output.

These settings are available when the Document Renderer feature is enabled for your instance.

**Rendering limits**

The maximum supported page size is 600 mm × 600 mm. The **Quality** setting applies when **Output Image Mode** keeps the original colors or converts pages to grayscale; it does not apply to black-and-white output.

| **Setting** | Description | Default value |
| --- | --- | --- |
| **Document Rendering Enabled** | Creates a downloadable PDF for the submission and includes its download URL in the completed submission output. | **Disabled** |
| **Document Measurement Unit** | Unit used for **Document Width** and **Document Height**. | **Inches** |
| **Document Width** | Width of each page in the rendered PDF. | 8.5 |
| **Document Height** | Height of each page in the rendered PDF. | 11 |
| **Quality** | Image quality used for color or grayscale output. Higher values produce larger files. | 50 |
| **Output image mode** | Choose whether to keep the original colors or convert images to grayscale or black and white. | **Keep original colors** |
| **Searchable PDF Output** | In v43.2 and later, adds a searchable text layer to the rendered PDF. If text segments are unavailable, the Document Renderer Block runs Segmentation and OCR to generate them. Searchable PDF output is supported only for documents that use a Latin-script language family. For other language families, Hyperscience creates an image-only PDF without a searchable text layer. Search highlights may be approximate because their placement is based on OCR-derived word bounding boxes and estimated text positioning. | **Disabled** |

#### Download a rendered document

When document rendering is enabled, each rendered document includes a `download_url` in the completed submission output.

1. Open the completed submission from the **Submissions** page.
2. From the submission card, click **Actions > View JSON Output**. This action is available to users with API access.
3. Search the output for `download_url`.
4. Open the URL. If it is a relative path beginning with `/api/`, append it directly to the base URL of your Hyperscience instance.

### Reprocessing

Use this setting to return an incorrectly matched document to Document Classification.

| Setting | Description | Default value |
| --- | --- | --- |
| **Reprocessing Enabled** | Turn on this setting to let keyers select **Mark Layout Variation Incorrect** during a Flexible Extraction task. The document then returns to Document Classification so that a keyer can match it to the correct layout variation. | Enabled |

## Standard subflow settings

The following setting groups are available when **Document Processing Subflow V43** is selected.

### Identification

Use these settings to configure Identification target accuracy, Supervision, QA, notifications, and task access.

| **Setting** | **Description** | **Default value** |
| --- | --- | --- |
| **Identification Target Accuracy (Entry Level)** | Set target accuracy for individual fields and table columns in the flow’s release. These individual values take the place of **Field Identification Target Accuracy** or **Table Identification Target Accuracy** for the selected entries. Click the pencil icon, and then enter a value in the **Target Accuracy** cell for each entry. Click outside the cell to save the value. Use the **Layout Name**, **Entry Name**, and **Entry Type** filters to narrow a long list. When you test one document at several target accuracies, an entry sent to Supervision at one accuracy may not be sent at every higher accuracy. Identification selects combinations of confidence thresholds independently for each accuracy level. Across varied submissions over time, a higher target accuracy results in lower automation and more Supervision tasks for that entry. For more information, see [Identification Settings](/v43/docs/identification-settings). | The value set in **Field Identification Target Accuracy** or **Table Identification Target Accuracy** |
| **Field Identification Target Accuracy** | Target accuracy for field identification. When a prediction falls below the value, Hyperscience sends the field and all its occurrences to Field Identification Supervision. | 95 |
| **Table Identification Target Accuracy** | Target accuracy for table identification. When a prediction falls below the value, Hyperscience sends the table to Table Identification Supervision. | 96 |
| **Manual Identification Supervision** | Turn on this setting to enable Field Identification Supervision and Table Identification Supervision. | Enabled |
| **Field Identification Quality Assurance** | Turn on this setting to enable Field Identification QA. When it is off, Hyperscience does not collect the QA data used to retrain existing Semi-structured models or automate Field Identification for new Semi-structured layouts. | Enabled |
| **Field Identification QA Sample Rate** | Percentage of documents selected at random for Field Identification QA. This setting appears when **Field Identification Quality Assurance** is turned on. | 5 |
| **Table Identification Quality Assurance** | Turn on this setting to enable Table Identification QA. When it is off, Hyperscience does not collect the QA data used to retrain existing Semi-structured models with tables or automate Table Identification for new Semi-structured layouts. | Enabled |
| **Table Identification QA Sample Rate** | Percentage of documents selected at random for Table Identification QA. This setting appears when **Table Identification Quality Assurance** is turned on. | 5 |
| **Manual Identification Notification Flow** | Notification flow used to send Manual Identification updates to downstream systems. | Submission State Notifications |
| **Default Task Restrictions** | Controls which users can access Supervision tasks created by the Manual Identification Block. For more information, see [Task Restrictions Overview](/v43/docs/task-restrictions-overview). | None |

### General Transcription

Use these settings to configure Transcription target accuracy, Supervision, QA, and model-training behavior.

| **Setting** | **Description** | **Default value** |
| --- | --- | --- |
| **Transcription Target Accuracy (Entry Level)** | Set flow-level target accuracy for individual fields in Structured layouts and for fields or table columns in Semi-structured layouts. For a Structured field, a value entered here overrides its Field Dictionary value. Click **Transcription Target Accuracy (Entry Level)** to open **Customize Transcription Target Accuracy**. Enter a value in the entry’s **Target Accuracy** cell, and click outside the cell to save it. Use the entry and layout filters to narrow a long list. A target accuracy does not guarantee that accuracy for every processed value. It applies a higher threshold to the selected entry than the applicable general text target accuracy. For more information, see [Transcription Automation and Accuracy](/v43/docs/transcription-accuracy-and-automation). | - For fields in Structured documents: - The value set in **Structured Text Target Accuracy**, or the **Transcription Target Accuracy** value set in the Field Dictionary (if any) - For fields or table columns in Semi-structured documents: - The value set in **Semi-structured Text Target Accuracy** |
| **Manual Transcription Supervision** | Turn on this setting to enable Transcription Supervision. | Enabled |
| **Transcription Quality Assurance** | Turn on this setting to enable Transcription QA. When it is off, Hyperscience does not collect the QA data used to determine transcription accuracy. | Enabled |
| **Automatic QA Sample Rate** | Turn on this setting to calculate QA sample rates automatically from existing QA records for Structured text, Structured checkboxes, Structured signatures, and Semi-structured fields. The calculated rates override the four corresponding Transcription QA sample-rate settings. | Disabled |
| **Structured Text Transcription QA Sample Rate** | Percentage of text fields in Structured documents selected at random for Transcription QA. | 5 |
| **Structured Checkbox Transcription QA Sample Rate** | Percentage of checkbox fields in Structured documents selected at random for Transcription QA. | 5 |
| **Structured Signature Transcription QA Sample Rate** | Percentage of signature fields in Structured documents selected at random for Transcription QA. | 5 |
| **Semi-structured Transcription QA Sample Rate** | Percentage of fields in Semi-structured documents selected at random for Transcription QA. | 5 |
| **Table Transcription Quality Assurance** | Turn on this setting to enable Table Transcription QA. When it is off, Hyperscience does not collect the QA data used to determine table-cell transcription accuracy. | **Enabled** |
| **Table Transcription QA Sample Rate** | Percentage of table cells selected at random for Transcription QA. | 5 |
| **Finetuning Only For Trained Layouts** | When this setting is on, Transcription Automation uses only layouts included when its model was trained. A newly added layout uses manually entered thresholds until a new model is trained with that layout. | Enabled |
| **Force Normalization Errors To Supervision** | Turn on this setting to send fields with normalization errors to Supervision for human review. | Disabled |
| **Force Missing/Blank Fields To Supervision** | Turn on this setting to send required fields that are blank to Transcription Supervision and required fields that are missing to Identification Supervision. | Disabled |

#### Task handling and notifications

These settings control Transcription Supervision tasks, notifications, and task access.

| **Setting** | Description | Default value |
| --- | --- | --- |
| **Manual Transcription Notification Flow** | Notification flow called when a Transcription Supervision task is required. | **Submission State Notifications** |
| **Default Task Restrictions** | Controls which users can access Transcription Supervision tasks. | None selected |
| **Supervision Transcription Masking** | Prevents users from entering characters that are not valid for the field’s data type. | **Enabled** |
| **Create Manual Transcription Task For Tables With Blank Cells** | Sends blank table cells to Manual Transcription Supervision regardless of their confidence scores. | **Enabled** |
| **Table Output Manual Review** | Turn on this setting to create a table-transcription task whenever a layout contains a table. | **Disabled** |
| **Exclude Auto-Transcribe Fields From QA** | Excludes fields with a Supervision Autotranscribe override from Transcription QA. | **Disabled** |

### Structured Document Transcription

Use these settings to configure Transcription Automation for text, checkbox, and signature fields in Structured documents.

| **Setting** | **Description** | **Default value** |
| --- | --- | --- |
| **Transcription Automation Training** | Uses Transcription QA data to calculate the combination of keyer and machine transcriptions needed to reach the target accuracy with the least keyer effort. Click **See Projections** to open the Transcription section of **Administration > System Settings**. | Disabled |
| **Period of Records to Use** | How far back Hyperscience looks for training data. This setting appears when **Transcription Automation Training** is turned on. | 30 days |
| **Improved Threshold Accuracy** | Helps ensure that the target accuracy is met across processed fields. | Enabled |
| **Structured Text Target Accuracy** | Target accuracy for text fields in Structured documents. When **Transcription Automation Training** is on and enough Transcription QA data is available, Hyperscience uses this value to calculate **Structured Text Automation**. | 95 |
| **Structured Text Automation** | Expected automation while Hyperscience works toward **Structured Text Target Accuracy**. Hyperscience calculates this value after enough Transcription QA data is available. | N/A |
| **Structured Text Threshold** | Minimum confidence required to process a text field automatically. A field below the threshold is sent to Transcription Supervision. When **Transcription Automation Training** is on, the calculated threshold replaces a value entered manually. | 0.52 |
| **Structured Text Minimum Legibility Threshold** | Minimum confidence required for a Structured text field to be treated as legible. A field below the value is marked as illegible. This setting applies only when Manual Transcription Supervision is off, the field has a Supervision Autotranscribe override, or the submission has a Machine-only override. | 0.1 |
| **Structured Checkbox Target Accuracy** | Target accuracy for checkboxes in Structured documents. When **Transcription Automation Training** is on, and enough Transcription QA data is available, Hyperscience uses this value to calculate **Structured Checkbox Automation**. | 95 |
| **Structured Checkbox Automation** | Expected automation while Hyperscience works toward **Structured Checkbox Target Accuracy**. Hyperscience calculates this value after enough Transcription QA data is available. | N/A |
| **Structured Checkbox Threshold** | Minimum confidence required to process a checkbox automatically. A checkbox below the threshold is sent to Transcription Supervision. When **Transcription Automation Training** is on, the calculated threshold replaces a value entered manually. | 0.56 |
| **Structured Checkbox Minimum Legibility Threshold** | Minimum confidence required for a Structured checkbox to be treated as legible. A checkbox below the value is marked as illegible. This setting applies only when Manual Transcription Supervision is off, the field has a Supervision Autotranscribe override, or the submission has a Machine-only override. | 0.5 |
| **Structured Signature Target Accuracy** | Target accuracy for signatures in Structured documents. When **Transcription Automation Training** is on and enough Transcription QA data is available, Hyperscience uses this value to calculate **Structured Signature Automation**. | 95 |
| **Structured Signature Automation** | Expected automation while Hyperscience works toward **Structured Signature Target Accuracy**. Hyperscience calculates this value after enough Transcription QA data is available. | N/A |
| **Structured Signature Threshold** | Minimum confidence required to process a signature automatically. A signature below the threshold is sent to Transcription Supervision. When **Transcription Automation Training** is on, the calculated threshold replaces a value entered manually. | 0.99 |
| **Structured Signature Minimum Legibility Threshold** | Minimum confidence required for a Structured signature to be treated as legible. A signature below the value is marked as illegible. This setting applies only when Manual Transcription Supervision is off, the field has a Supervision Autotranscribe override, or the submission has a Machine-only override. | 0.5 |

### Semi-Structured Document Transcription

Use these settings to configure Transcription Automation for text fields, table cells, and checkboxes in Semi-structured documents.

| **Setting** | **Description** | **Default value** |
| --- | --- | --- |
| **Transcription Automation Training** | Uses Transcription QA data to calculate the combination of keyer and machine transcriptions needed to reach the target accuracy with the least keyer effort. Click **See Projections** to open the Transcription section of **Administration > System Settings**. | Disabled |
| **Period of Records to Use** | How far back Hyperscience looks for training data. This setting appears when **Transcription Automation Training** is turned on. | 30 days |
| **Improved Threshold Accuracy** | Helps ensure that the target accuracy is met across processed fields. | Enabled |
| **Semi-structured Text Target Accuracy** | Target accuracy for text fields in Semi-structured documents. When **Transcription Automation Training** is on and enough Transcription QA data is available, Hyperscience uses this value to calculate **Semi-structured Text Automation**. | 95 |
| **Semi-structured Text Automation** | Expected automation while Hyperscience works toward **Semi-structured Text Target Accuracy**. Hyperscience calculates this value after enough Transcription QA data is available. | N/A |
| **Semi-structured Text Threshold** | Minimum confidence required to process a text field automatically. A field below the threshold is sent to Transcription Supervision. When **Transcription Automation Training** is on, the calculated threshold replaces a value entered manually. | 0.52 |
| **Semi-structured Text Minimum Legibility Threshold** | Minimum confidence required for a Semi-structured text field to be treated as legible. A field below the value is marked as illegible. This setting applies only when Manual Transcription Supervision is off, the field has a Supervision Autotranscribe override, or the submission has a Machine-only override. | 0.1 |
| **Table Target Accuracy** | Target accuracy for table-cell transcriptions in Semi-structured documents. When **Transcription Automation Training** is on, and enough Transcription QA data is available, Hyperscience uses this value to calculate **Table Automation**. | 95 |
| **Table Automation** | Expected automation while Hyperscience works toward **Table Target Accuracy**. Hyperscience calculates this value after enough Transcription QA data is available. | N/A |
| **Table Threshold** | Minimum confidence required to process a table cell automatically. A table cell below the threshold is sent to Transcription Supervision. When **Transcription Automation Training** is on, the calculated threshold replaces a value entered manually. | 0.52 |
| **Table Minimum Legibility Threshold** | Minimum confidence required for a Semi-structured table cell to be treated as legible. A table cell below the value is marked as illegible. This setting applies only when Manual Transcription Supervision is off, or the submission has a Machine-only override. | 0.1 |
| **Semi-structured Checkbox Target Accuracy** | Target accuracy for checkboxes in Semi-structured documents. When **Transcription Automation Training** is on, and enough Transcription QA data is available, Hyperscience uses this value to calculate **Semi-structured Checkbox Automation**. | 95 |
| **Semi-structured Checkbox Automation** | Expected automation while Hyperscience works toward **Semi-structured Checkbox Target Accuracy**. Hyperscience calculates this value after enough Transcription QA data is available. | N/A |
| **Semi-structured Checkbox Threshold** | Minimum confidence required to process a checkbox automatically. A checkbox below the threshold is sent to Transcription Supervision. When **Transcription Automation Training** is on, the calculated threshold replaces a value entered manually. | 0.56 |
| **Semi-structured Checkbox Minimum Legibility Threshold** | Minimum confidence required for a Semi-structured checkbox to be treated as legible. A checkbox below the value is marked as illegible. This setting applies only when Manual Transcription Supervision is off, the field has a Supervision Autotranscribe override, or the submission has a Machine-only override. | 0.5 |

## ORCA subflow settings

The following settings are available when **Document Processing with ORCA Subflow V43** is selected.

**When evaluating ORCA**

- To review ORCA output through Supervision, turn on **Flexible Extraction Show Machine Predictions**.
- ORCA predictions then appear in Flexible Extraction tasks for human review.
- To evaluate ORCA accuracy through Quality Assurance, select **ORCA Quality Assurance Flow** and set **ORCA QA Sample Rate** to 100%. This configuration generates QA tasks and provides the data used for accuracy reporting.

### ORCA

Use these settings to select the ORCA model, configure field-extraction accuracy, and generate ORCA QA tasks.

**ORCA 2 in v43.2**

ORCA 2 is available for field extraction. ORCA 1.0 and its specialized models remain supported during the transition, but ORCA 1.0 is planned for deprecation in a future version. The base models available under **Administration > Assets** depend on your license.

| Setting | Description | Default value |
| --- | --- | --- |
| **ORCA Base Model** | Select the installed base model used to process submissions when **Use Live Model** is off. The selected model must be installed. | None selected |
| **Use Live Model** | Controls whether processing uses the live specialized model associated with the layout or the ORCA base model selected in the flow. - When it is on, Hyperscience uses the live VLM Field Extraction model associated with the layout. If no live model is available, processing stops and returns an error. The base model selected in the flow does not control processing. - When it is off, the selected ORCA base model controls processing. A specialized model is used only when it was trained with the selected base model. Models trained with a different base model are ignored. | **Disabled** |
| **ORCA Target Accuracy** | Target accuracy for field extraction. When a prediction falls below the value, Hyperscience sends the field and all its occurrences to Flexible Extraction Supervision. This target applies only after a specialized model definition has been created for the flow from the base model. | 95 |
| **Sliding Window Size** | Number of pages the model processes at once. Larger values can improve throughput but increase GPU-memory usage. Hidden by default. | 10 |
| **Max New Tokens** | Maximum number of tokens the model can generate for extracted values. Scale this value for the number and expected length of fields. Hidden by default. | 1024 |
| **Max Image Tokens** | Maximum image-token budget per page. Larger values can improve accuracy for dense or difficult-to-read pages but increase GPU-memory usage. Hidden by default. | 1280 |
| **Max Number of Bounding Boxes** | Maximum number of candidate bounding boxes for each field occurrence. Larger values can improve localization but may increase false positives. Hidden by default. | 3 |
| **Quality Assurance Flow** | Subflow used to generate VLM QA tasks. To generate QA tasks and measure accuracy, select **ORCA Quality Assurance Flow**. | None |
| **Task Restrictions** | Controls which users can access ORCA QA tasks. For more information, see [Task Restrictions Overview](/v43/docs/task-restrictions-overview). | None selected |
| **ORCA QA Sample Rate** | Percentage of documents selected at random for VLM QA. This setting applies only when a flow is selected in **Quality Assurance Flow**. | 5 |

**VLM Table extraction in v43.2 and later**

Use the separate **Table Extraction Flow** for out-of-the-box ORCA 2 table extraction. It extracts tables as rows and columns without requiring a specialized model. Contact your Hyperscience representative for more information.

## Upgrade the standard subflow

When a newer standard document-processing subflow is available and supports your business case, select it from **Flow Identifier** in the **Start Document Processing Subflow Block**. The top-level Input Block, Output Block, and their connections remain in place.

If you create a custom top-level flow, configure it so that its document-processing subflow can also be replaced through **Flow Identifier**.

The main flow that manages the end-to-end document processing, coordinating with subflows to handle specific components of the process. Learn more in [Flows](https://help.hyperscience.ai/latest/docs/flows).

A foundational model that provides core general-purpose capabilities and is not directly trained on customer-specific examples. Use-case specialization is achieved through additional training on top of the base model using customer-specific data.

A configuration within Hyperscience designed to process documents where fields and table cells are present, but their positions can vary among documents. Unlike Structured layouts, Semi-structured layouts do not rely on fixed field locations. Instead, a model is trained to find the fields and cells based on provided training examples.

The process of grouping related files, documents, or pages into a single case using a unique identifier called a Case ID. For example, if you submit multiple documents for a loan application, collation ensures that all these documents are grouped together under one case for streamlined processing and review.

A setting in the Machine Classification Block that identifies and corrects the orientation of page images by automatically rotating them.

**Manual Transcription** is a Supervision task used to review and correct extracted data when the system lacks confidence. It is commonly used for handwritten text, low-quality scans, or unclear characters, ensuring accurate results.

A task in Hyperscience that involves human intervention to validate or correct data extraction for Structured documents. This task is used when automatic extraction isn’t fully reliable, allowing you to transcribe or adjust specific fields to ensure accuracy.

A manual task that is created when the system’s confidence in a prediction is below the confidence threshold. Supervision allows a human to review and correct the output, ensuring data accuracy through human-in-the-loop input.

## Related

- [Flows Overview](/flows-overview.md)
- [Flow Blocks](/flow-blocks.md)
- [Editing Flows in Flow Studio](/editing-flows-in-flow-studio.md)
- [Testing and Debugging Flows](/testing-and-debugging-flows.md)
- [On-Error Flows](/on-error-flows.md)
- [Flow Runs Page](/flow-runs-page.md)