1. LandOffer
  2. Blog
  3. All articles
  4. Redact a Portfolio PDF: Remove the Data Behind the Black Boxes

Job search

Redact a Portfolio PDF: Remove the Data Behind the Black Boxes

Compare an actual synthetic PDF overlay with applied redactions and hidden-carrier cleanup, then verify the exact local file and proposed package.

9 min readLandOffer team

A black rectangle can hide a line on screen while leaving the underlying PDF text available to copy, search, or extract. Check the exact saved PDF before attaching it; the appearance of a black box does not establish what the file still contains. Visible redaction, hidden-content cleanup, readability, and permission to share are separate decisions.

This guide actually created and processed an original two-page PDF containing an invented identifier. The negative examples and cleaned result are saved with their scripts, extracted text, object inspections, and rendered pages. No employer document, real secret, Acrobat account, upload, or recipient was involved.

A paper conservator compares a black ribbon over intact writing with a removed strip and separate hidden envelopes.

Use LandOffer’s recently posted jobs to choose which authorized work sample belongs in an application. Keep the reviewed candidate in a separate package from the original and the diagnostic files used to inspect it.

Establish what the final PDF is allowed to contain

Start with the specific artifact and sharing scope. A redaction tool can change content; it cannot grant an employer's permission, resolve a confidentiality obligation, or approve a new audience. If the proposed use is not authorized, technical cleanup does not turn it into authorized disclosure.

For an allowed portfolio excerpt, identify the pages, text, figures, names, identifiers, comments, attachments, and metadata that must be excluded. Consider the whole package, including the filename and related downloads. Review the attachment list as well as the PDF: an added appendix may exceed the permitted scope even when the selected PDF passes its technical checks.

Work on a copy while preserving the restricted original in its appropriate location. The original can support an authorized review of the changes. Exclude it from the proposed recipient package rather than placing both versions in the same attachment folder. Our originals are harmless synthetic fixtures; that is not a reason to bundle source versions with a real application.

Keep this task distinct from requesting permission. The earlier approval process determines what may be shared. Here, the technical question is whether the proposed file actually excludes the specified content and remains readable. The sample below does not send a permission request or make a real disclosure decision.

Apply redactions instead of painting over content

Adobe's current Acrobat Pro redaction instructions distinguish selecting content, applying redactions, choosing whether to sanitize hidden information, and saving the result. Marking an area for redaction is a step before applying it. An ordinary drawing rectangle is another operation entirely.

If using Acrobat, follow the current documented Redact a PDF workflow and inspect the resulting saved copy. This article did not run that UI or test an Acrobat subscription. Its executed comparison uses PyMuPDF 1.28.2 locally, with ReportLab generating the original PDF.

A visible black area is therefore not the acceptance criterion. The output should also be reopened and checked for the targeted data through extraction and search, along with the hidden carriers relevant to the file. Keep the test method specific enough that another reviewer can reproduce it.

Do not use a screenshot of the black boxes as proof of removal. A screenshot captures pixels from one rendering; it does not describe the objects or attachments remaining in the PDF. Conversely, a zero-result text search cannot establish that every possible image, encoding, or hidden data carrier has been inspected.

Asset one: the executed overlay and removal comparison

The original fixture uses the invented marker SYNTHETIC-ONLY-COPPER-742 in two text positions, one on each page. A separate preparation step places that same marker in standard document metadata, XMP metadata, two text comments, and one embedded text attachment. All content is newly authored for this experiment.

The PDF uses ordinary Helvetica and Courier text without scanned images, form fields, or optional-content layers. Its public headings, explanatory paragraphs, and page footers remain outside the targeted rectangles. That gives us both a removal target and content that should survive the operation.

The first negative control draws opaque black rectangles over the two visible marker locations. After saving and reopening it, both PyMuPDF and pypdf still extract the marker, and page search finds two hits. Its metadata, comments, and attachment also remain. The black boxes change the appearance without deleting the text behind them.

The second version creates actual redaction annotations and applies them to the page text. It is saved as a new file with garbage collection. Both extractors then omit the marker and page search returns zero hits, but the intentionally retained metadata, XMP, comments, and attachment still contain it.

The final version explicitly removes those tested hidden carriers, performs additional configured cleanup, and saves a new complete PDF with garbage=4, content cleaning, compression, and incremental=False. It is then reopened for the checks below.

Saved version Page text extraction and search Tested hidden carriers Completed decision
Black-box overlay Marker extracted by both tools; two search hits Metadata, XMP, two comments, one attachment remain Reject as a removal result
Applied redactions only Marker absent from both extractions; zero search hits The same hidden carriers remain Reject as the cleaned sharing candidate
Final cleaned copy Marker absent from both extractions; zero search hits No marker in tested metadata/XMP; zero comments and embedded files Retain as a local candidate within the experiment's limits

The final checks also inspect every PDF cross-reference object and decoded stream available through PyMuPDF, plus raw bytes and specified marker encodings. None contains the known marker after cleanup. The saved report lists the object and decoded-stream references where the negative controls retained the marker, so the zero-hit final result can be compared with actual failing versions.

Fifteen assertions pass. Both pages remain present, the allowed public context remains extractable, and the final file reports a single saved version. The enriched source file's hash remains unchanged during the overlay and cleanup work. These are observed results for this fixture, not a universal guarantee that arbitrary PDF secrets are unrecoverable.

Check hidden content separately from page text

Adobe's sanitization guidance, updated October 22, 2026, describes hidden information such as metadata, comments, and layers, with remove-all and selective-removal paths. This supports reviewing more than what appears on the page. It does not make our local cleanup an Acrobat experiment.

In our test, page text was already removed in the intermediate version while the hidden carriers still held the identifier. That concrete counterexample is why the final review lists extraction, metadata, comments, and attachments separately. A single successful search would have missed the remaining issue.

PyMuPDF's page-redaction reference documents text removal and save-related conditions, while also describing separate treatment of images and graphics. The executed fixture selects text removal and leaves image and graphics handling out of the claim. No image-redaction behavior was tested.

Its document cleanup and save reference describes configurable cleanup and garbage collection; incremental saving is a different mode. Our script writes new files with explicit options. That does not erase the separately preserved originals, filesystem copies, backups, or anything previously sent to another person.

Treat low-level checks as bounded evidence. Searching one known marker in objects and streams is useful here because we authored its locations. It checks this known marker in the tested representations. Unknown secrets, untested encodings, and other structures require their own review; a zero-hit result here cannot clear them.

Reopen and inspect every final page

We rendered all five saved PDFs, each with two pages, using Poppler. The ten page renders and full-page contact sheets were actually inspected. The two final pages were also viewed individually. Headings, public paragraphs, and footers remained legible; the targeted locations showed black areas, and the final copy no longer showed the comment icons.

That visual inspection serves a different purpose from extraction. It checks that redaction did not clip surrounding material, remove an important explanation, or leave the portfolio unreadable. The extracted results establish a different boundary: the selected identifier no longer appears through the tested text and object paths.

The fixture is deliberately limited. It does not test scanned or raster text, OCR layers, rotated text, non-Latin or subset fonts, forms or XFA, JavaScript, optional-content layers, digital signatures, encryption, malformed objects, or pre-existing incremental revisions. It also does not perform cross-tool forensic recovery. Do not carry its clean result over to those untested situations.

If a real document includes such content, identify it before choosing the method and verify the actual output with suitable tools and authorized review. Do not call an untested carrier cleared merely because its name appears among a library's cleanup options.

Asset two: a completed scope and final-package decision

A separate original decision memo, SCOPE-B-042, asks what belongs in the proposed attachment package after technical checking. Its scope is stipulated fiction: exactly the two-page cleaned synthetic PDF, public summary text only, with no appendices, source versions, or diagnostic dumps. It is not an actual employer approval, and nothing is uploaded or sent.

The initial package includes the cleaned PDF plus a newly authored separate appendix containing the invented marker SYNTHETIC-ONLY-FERN-19. The appendix is outside the stipulated scope even though it contains no real confidential material. The completed package evaluator holds that combination rather than treating PDF cleanup as approval for the extra file.

The final package contains only final-clean.pdf. The decision memo records its full SHA-256 hash, with prefix f755d8c6ec3c, its two-page count, the tested residue report, and the excluded files. A counterexample with a different expected review hash is held, preserving the distinction between the reviewed version and a later change.

Review item Actual local result Final disposition
Added appendix Outside declared package; its separate marker is present Exclude the appendix and hold the initial package
PDF residue Known marker absent through tested paths; comments and embedded files absent Preserve technical evidence without expanding it into a permission decision
Version identity Final file's hash matches the exercise's recorded candidate Keep that exact local file as the candidate
Changed version Deliberately mismatched expected hash Hold until the new version is reviewed
Disclosure authority Only a fictional scope exists; no real approval or send Real sharing decision remains unperformed

Five package checks pass. This second asset is a complete decision record with a real file identity and actual package counterexamples, not a blank checklist or an unsent permission-email template. Its evaluator checks the declared file list and known appendix marker; it is not a general disclosure-policy engine.

A clean two-sheet folio remains separate from excluded drafts, an appendix, and diagnostic review materials.

Keep the evidence bundle separate from the attachment

The diagnostic bundle contains the original PDFs, negative controls, extracted text, object dumps, scripts, results, and renders. In this experiment those are all synthetic. In a real workflow, those files may reproduce exactly the information excluded from the portfolio and should stay within the authorized review environment.

Do not attach the whole working folder because the final PDF passed a check. Open the exact candidate again, confirm its filename and version, inspect the pages and relevant hidden carriers, and compare the proposed package with the approved scope. Any further edit changes what needs review.

A safe resume description of this actual personal exercise would identify it as an original synthetic PDF experiment: compared overlays with applied redactions, checked selected hidden carriers, preserved readable context, and documented a bounded final-package decision. It would not claim Acrobat production experience, legal clearance, or guaranteed irreversible removal for every PDF.

Keep the selected work sample and its permitted review notes beside relevant applications in LandOffer.ai. Before sharing, match the saved candidate and attachment list to the technical results and the separate authorization covering that exact version, audience, and route.

Sources and Further Reading