Open-source PDF forensic tool
A PDF can look redacted while the original text is still stored inside the document. RedactionResearch checks whether supposedly hidden information remains technically present, then gives you a local interface for reviewing the evidence.
The problem
PDFs are collections of text, images, annotations, forms and document metadata. Drawing a dark rectangle above a name can make the page look safe while leaving the name selectable, searchable or extractable. The text may sit beneath another page object, inside a form field, or elsewhere in the PDF structure.
Secure redaction removes sensitive information from the document. Simply covering it visually may leave the original data intact. That does not mean every black rectangle is insecure; it means appearance alone cannot confirm that a PDF was properly redacted.
RedactionResearch checks the PDF itself for technical indicators of an incomplete or failed PDF redaction. Each indicator still requires human verification before it should be treated as an actual disclosure problem.
Local review workflow
Import locally, scan the document structure, and decide what the evidence actually means.
Create a project and choose a local folder. RedactionResearch imports the PDF files into the project workspace and keeps the collection organized for repeat scans and review.
The scanner examines document objects and contextual signals for possible redaction failures, hidden machine-readable text and sensitive information outside the visible page.
Open the actual PDF, jump to affected pages, inspect coordinates and reconstructed text, review metadata, evidence and risk score, then choose Accept or Skip.
The scanner does not automatically declare every technical signal to be a data leak. Human review is a deliberate part of the workflow.
Project overview
The project workflow is designed for reviewing more than a single suspicious file. Add PDFs, run the scanner across documents that have not already completed successfully, and track open findings from one local dashboard.
Accepted findings are collected separately. Skipped findings remain available, so a decision can be inspected again when context changes or a second reviewer takes over.
Detection coverage
The scanner combines direct evidence with contextual signals. The human-readable issue comes first; detector identifiers are included for technical traceability.
Flags genuine redaction annotations over live text, dark annotations, overlay objects, dark page content and machine-readable text that is visually hidden.
REDACTION_ANNOTATION_WITH_LIVE_TEXTDARK_ANNOTATION_OVER_LIVE_TEXTANNOTATION_OVERLAY_HIDES_LIVE_TEXTDARK_PAGE_CONTENT_HIDES_LIVE_TEXTLIVE_TEXT_NOT_VISIBLE_ON_WHITE_REGIONChecks PDF/XMP metadata, populated form fields, bookmark outlines, suspicious embedded attachments and incremental revisions. These channels can matter even when the visible page was correctly sanitized.
For detected email, IBAN and phone patterns, the scanner stores the category rather than persisting the matched value as review text.
Records optional content layers, embedded files, populated form fields and revision history as context. These signals can guide an investigation but are not necessarily evidence of a leak by themselves.
Risk score and severity help prioritize review; they do not replace examination of the document and its purpose.
Local-first by design
RedactionResearch runs as a local web application. PDFs are not uploaded to a hosted RedactionResearch service, and analysis results and review decisions remain on the computer where the application is running.
SQLite stores projects, findings, scan state and Accept or Skip decisions locally. Extracted PDF content and reconstructed text are displayed in the local interface. This makes the local PDF redaction checker useful when documents should not be sent to an external scanning service.
Human verification
PDF internals are messy. Legitimate annotations, form data or revision history can look suspicious without exposing sensitive information. RedactionResearch intentionally keeps technical detection separate from confirmation.
Choose Accept for a confirmed problem or Skip for a benign structure or false positive. Confirmed items appear under Found; skipped items stay under Skipped rather than disappearing from the audit trail.
Potential use cases
RedactionResearch can be useful for people who publish, receive or audit redacted documents.
A necessary limit
No.
If sensitive content was properly removed from the PDF, RedactionResearch cannot recover it. The tool finds cases where information is still technically present in the file.
It is not a magical utility for reading text behind every PDF redaction. It is a forensic check for information that a broken redaction, metadata field or other document structure failed to remove.
Installation
Install RedactionResearch globally with npm or Homebrew, then launch the local review interface from your terminal.
Install RedactionResearch globally from the npm registry:
npm install --global redaction-researchThen start the application:
RedactionResearchRequires Node.js 22.5 or newer.
For macOS and other Homebrew environments:
brew install oliverjessner/tap/redaction-researchThen start the application:
RedactionResearchHomebrew resolves the formula dependencies automatically.
http://127.0.0.1:3000/Homebrew offers a one-command setup for this PDF redaction checker for macOS and other Homebrew environments; npm is the direct package-registry option for supported Node.js environments.
FAQ
Practical answers about hidden text, local processing and what a forensic result means.
Do not rely on visual inspection alone. Check the file for remaining live text, annotation overlays, metadata, form values, attachments, bookmarks and revision history, then review every technical finding in context.
Yes. A rectangle or annotation can visually cover text while leaving the original text object selectable, searchable or extractable inside the PDF.
No. If the original information was genuinely removed from the PDF, there is nothing to recover. RedactionResearch only exposes or flags information that remains technically present in the file.
No hosted upload is part of the application. PDFs, analysis data and review decisions stay on the machine where RedactionResearch is running.
No. It also examines redaction and overlay annotations, visually hidden text, metadata, form fields, bookmark outlines, embedded attachments and incremental PDF revisions.
No. Findings are forensic indicators, not automatic breach declarations. A human should inspect the PDF, the affected region and the supporting evidence before accepting or skipping a finding.
Yes. Create a project, import a local folder of PDFs and run the forensic scan across documents that have not already been scanned successfully.
No. It is an open-source PDF redaction tool for detecting and reviewing potentially incomplete redactions, not for authoring or sanitizing redactions.
Open source · local review
Use an open-source PDF redaction tool to inspect the file, review the evidence and make an informed decision before publication.