Redaction Supervision Demo flow

Use the Redaction Supervision Demo flow to find Personally Identifiable Information (PII) in documents, review the detected information in Supervision, and create redacted or masked PDF copies for sharing. This article explains the flow, its configuration options, and when to use them.

Understanding redaction and masking

Redaction replaces sensitive content in the rendered output with an opaque box. When searchable PDF output is enabled, text beneath the box is not included in the new searchable text layer.

Masking replaces sensitive text with synthetic values. Depending on the entity type, the replacement can preserve characteristics of the original value. For example:

• A masked Social Security number can preserve the number of digits and separators.

• A masked address can retain components such as a street number, street name, and postal code.

• A masked name can retain characteristics such as its length and given-name and surname structure.

The flow creates a redacted or masked derivative of the source document

It does not delete the original document or other processing data. Manage the retention and deletion of source data according to your organization's policies.

The Redaction Supervision Demo flow

The Redaction Supervision Demo flow is a maintained reference implementation. You can use it as a starting point for a redaction workflow and adapt its settings to your requirements. Contact your Hyperscience representative to confirm that you have the latest compatible flow and supporting artifacts.

The flow processes each submission in these steps:

1. Image Correction – The system rotates and de-skews the pages.

2. Entity Recognition – The system locates sensitive information with several detection mechanisms:

   • Named Entity Recognition Block (NER) finds person names, organization names, and addresses.

   • Custom Entity Recognition uses keywords and regular expressions to find values such as Social Security numbers, account numbers, phone numbers, email addresses, dates, and passport numbers. To learn more, see Entity Recognition Block.

   • Signature detection finds signatures on the page.

3. Supervision – When Manual Identification Supervision is enabled, a reviewer confirms, removes, or adds locations in a Field Identification Task.

4. Redaction – The system applies black boxes or synthetic data to the confirmed locations.

5. Redaction Rendering – The system creates downloadable PDF output. The PDF can optionally include a hidden searchable text layer.

The NER and signature-detection components are ready to use without customer training. Custom Entity Recognition uses the keyword and regular-expression rules configured in the flow.

v43.2 or later

The Redaction Rendering Subflow requires Hyperscience v43.2 or later.

Key features

Review detected information

Human review is commonly used in high-risk redaction processes. Manual Identification Supervision is enabled by default, but you can disable it for workflows that do not require review. Your organization is responsible for determining its review, audit, and regulatory requirements.

The flow includes settings that reduce the Average Handling Time (AHT) of redaction tasks:

• Use Single Task Supervision places all documents in a submission into one Supervision task. This can help reviewers examine case packets in which the same names, addresses, or other values appear across several documents.

• Use Single Field in Supervision displays all detected entities under one generic field. When the setting is off, fields are separated by entity type.

• The current keyboard shortcuts include an action that removes one field occurrence while keeping the reviewer on the same page. The shortcut is Alt/Option+Shift+X. Current shortcuts are enabled by default, and users can switch to the legacy shortcuts from the keyboard-shortcuts panel. To learn more, see Keyboard Shortcuts.

Control how content is hidden

• Text Replacement Mode – Select Black box to redact text, or Replace with synthetic data to mask it.

• Free-form entity replacement mode – Select Black box or Blurred overlay for non-text content, such as photos or logos that a reviewer marks with a free-form Bounding Box. This setting does not apply to signatures.

• Replacements – Set a fixed value for an entity type. For example, {"ssn": {"*": "111-11-1111"}} masks every Social Security number with the same value. The value adapts to the format of the original, such as its separators. Use this setting when readers must see that a value was masked.

• Skip Entity Types – List the entity types that the system must not redact or mask.

• Add Watermark – Marks each redacted document. By default, the watermark shows "Redacted at" and the date and time. You can enter your own text in Watermark Text.

A blurred overlay might not provide irreversible redaction.

Use an opaque box when your workflow requires sensitive pixels to be unreadable, and validate the rendered documents before releasing them.

Create searchable PDF output

When you turn on Searchable PDF Output, the system adds a hidden text layer to each rendered PDF. You can then search for visible text in tools such as Adobe Acrobat Reader and validate expected redactions.

The system generates this text layer from the final redacted page images instead of reusing text segments from the source pages. It runs Hyperscience Segmentation and Optical Intelligent Character Recognition (OICR) on those images before creating the PDF. As a result:

• Text replaced by an opaque box is not copied from the source page into the searchable text layer.

• The generated text geometry can be more precise than text recognition performed by a PDF viewer.

Searchable rendering reduces the risk of reintroducing original text through the PDF text layer.

It does not verify that the selected redaction or masking method has made the visible pixels unreadable. Review the rendered output before sharing it.

When to use redaction and masking

Best-fit scenarios

The flow can also support privacy workflows involving GDPR, legal discovery, HIPAA, or financial information. Using the flow does not by itself establish compliance with a law, regulation, or organizational policy.

Limitations and considerations

• Language support – Searchable PDF output supports documents that use Latin script. The system renders other documents without a searchable text layer. Learn more in Supported Languages.

• Processing time – Searchable PDF output runs text segmentation and recognition on every rendered page. This increases processing time, so enable it only when searchable output is required.

• OICR results – The searchable layer can contain synthetic replacement values and recognition errors. Validate the output before using it in a downstream process.

• Single-task output – When Use Single Task Supervision is enabled, the system creates one PDF containing all pages in the submission. Entity locations that span pages from different source files are not supported.

• Source-data retention – The flow creates a new rendered document; it does not delete the source document or all intermediate processing data.

Configuration recommendation

The supplied PII Regex Map and PII Keyword Map contain default entity types and detection rules. Extend these maps when you need additional entity types. Replace or remove default entries only after validating the change against representative documents.

Using redaction and masking

After you import the bundle, configure the flow, deploy it, and process a submission to create a redacted or masked PDF.

Step 1: Configure redaction and masking

1. Open the flow in Flow Studio

2. In Settings Type, select Redaction.

   

3. In Text Replacement Mode, select one of these options:

   • Black box – to redact text.

   • Replace with synthetic data – to mask text.

   

4. In Free-form entity replacement mode, select Black box or Blurred overlay.

   

5. (Optional) Set Replacements, Skip Entity Types, Add Watermark, and Watermark Text.

6. Click Save.

Step 2 (optional): Configure the PDF format

You can change the format of the rendered PDF in the Redaction Rendering Subflow.

1. Open the Redaction Rendering Subflow.

2. Select the Creates a printable file from a list of images block.

3. Update these settings as needed:

   • Format – The file format of the output, for example, PDF.

   • Page Height and Page Width – The page size, in millimeters. The default is 279.4 × 215.9 (US Letter).

   • Image Quality – A higher value keeps more detail but creates a larger file.

   • DPI – The resolution of the page images. The default is 200.

   • Use Native Dimensions – Keeps the original dimensions of each page.

   • Output Image Mode – Keeps the original colors, or converts the pages to grayscale or black and white. Grayscale and black and white create smaller files.

   

Step 3: Deploy the flow

Deploy the flow to apply your settings.

Step 4: Process a submission

1. Submit a document to the flow.

2. If Manual Identification Supervision is enabled, complete the Supervision task. Confirm, remove, or add locations as needed.

Step 5: Download the redacted document

1. Open the submission.

2. Click Actions, and then click View Transformed Output.

3. In Submission Transformed Output, find redacted_documents. It is the first item in the output.

4. Click the link under redacted_documents to download the PDF.

   

Result: The downloaded PDF shows black boxes or synthetic data in place of the detected information.

Example results

The following examples show the same page processed with each replacement mode and with a watermark.

Get the Redaction Supervision demo flow

The Redaction Supervision demo flow is available on demand. Hyperscience supplies it in a bundle containing the compatible flow, supporting artifacts, a preconfigured redaction layout, and the layout release.

The preconfigured layout contains fields for common types of information

The preconfigured layout contains fields, such as name, address, date, company name, email address. You can add or remove fields to meet your organization’s requirements. You can also use the Redaction catch-all field to mark any other content or page regions that you want to redact.

Before importing the bundle:

• Confirm that you are using Hyperscience v43.2 or later.

• Make sure that you have the latest compatible version of the flow and its supporting artifacts.

• Import the provided redaction layout and release.

For help with the current bundle or setup, file a Support ticket.

The following walkthrough shows how to configure the Redaction Supervision Demo flow, process a document, and download the redacted PDF: