PDF editing, redaction & assembly
Redact, annotate, sign, reorder, merge and split PDFs correctly.
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- Acrobat · Preview · PDF-XChange · Foxit
- Class GUI-need label
G2· GUI-essential
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R04.05.01.01-T01 Reorder, rotate and trim a scanned lease in Preview
Original instruction
Open input/lease_scan.pdf (14 pages) in Preview. Put pages in the order given by the printed page numbers, rotate the two upside-down pages upright, delete the one blank page, and save as output/lease_ordered.pdf.
- Platforms
mac· macOS- Applications
- Preview
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Output has 13 pages; page image hashes in key order; /Rotate values make the two flipped pages upright (OCR orientation check); blank page absent; Producer is macOS Quartz PDFContext; the VM has no qpdf/PyMuPDF/PyObjC and Automator/Shortcuts PDF actions are disabled.
Proposed partial credit
45% correct order; 25% rotations; 20% blank removed; 10% saved by Preview.
R04.05.01.01-T02 Merge four reports into one PDF with a bookmark per file
Original instruction
Combine input/cover.pdf, input/summary.pdf, input/findings.pdf and input/appendix.pdf in that order into output/audit_pack.pdf with Acrobat's Combine Files, adding one top-level bookmark per source file named after it (without extension), each pointing to that file's first page.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
drag· Drag & dropclick· Point & click- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Page count equals sum; page content hashes in order; outline has 4 top-level items with exact titles and destinations at the right page indices; Producer string from Acrobat. Acrobat's JavaScript console is disabled and the VM has no PDF libraries.
Proposed partial credit
40% pages merged in order; 45% bookmarks titled and targeted; 15% nothing else altered.
R04.05.01.01-T03 Redact SSNs and three named people across a 40-page case file
Original instruction
In input/case_file.pdf (40 pages, synthetic data) redact every SSN-pattern number and every occurrence of the 3 names in input/redact_names.txt, apply the redactions, remove hidden information (metadata, comments, hidden text) and save as output/case_file_redacted.pdf. Keep the original file untouched.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicktype· Text & formsvis· Visual judgment- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Text extraction (pdfminer) and OCR of rendered pages find 0 SSN matches and 0 target names; black boxes cover each seeded location (IoU ≥0.8); no /Annots, XMP/Info cleared; non-target text intact (≥99% tokens). Acrobat's JavaScript console is disabled and the VM has no PDF libraries; Producer must be Acrobat.
Proposed partial credit
40% SSNs removed; 30% names removed; 15% no over-redaction; 15% hidden info removed.
R04.05.01.01-T04 Burn in redactions on an image-only scan with Preview
Original instruction
input/bank_letter_scan.pdf is image-only (no text layer). In Preview, use the Redact tool to permanently black out the account number, the routing number and the signature on page 1 and the account number again on page 2. Save as output/bank_letter_redacted.pdf.
- Platforms
mac· macOS- Applications
- Preview
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropvis· Visual judgmentgeom· Precise geometry- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Rendered page pixels inside each ground-truth box are ≥98% black; OCR finds no digits of the seeded numbers; no recoverable original image under the box (image XObjects re-encoded); boxes do not cover >20% extra area. The VM has no PDF libraries or PyObjC and Automator/Shortcuts PDF actions are disabled; Producer must be macOS Quartz.
Proposed partial credit
70% four regions redacted (pro-rata); 15% not recoverable; 15% minimal over-redaction.
R04.05.01.01-T05 Highlight one exact sentence and attach a reviewer note
Original instruction
In input/supply_contract.pdf highlight exactly the sentence beginning 'The Supplier shall indemnify' on page 4 (not the clause number) and attach a sticky note 'Check cap vs clause 12.3' anchored to it. Also underline the defined term 'Losses' in clause 9.1. Save in place.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat · Preview
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
drag· Drag & dropclick· Point & clicktype· Text & forms- Work product
WP10· Annotations- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Annots: one /Highlight whose QuadPoints cover exactly the sentence tokens (token-level F1 ≥0.98), a /Text or popup note with exact contents anchored on page 4, one /Underline over 'Losses' in 9.1; no other annotations. No PDF libraries or PyObjC in the VM; Acrobat JavaScript and Automator PDF actions disabled; Producer must be Acrobat or macOS Quartz.
Proposed partial credit
50% highlight span exact; 25% note text and anchor; 25% underline.
R04.05.01.01-T06 Split a statement bundle into per-customer PDFs by an index
Original instruction
input/statements_oct.pdf (120 pages) contains statements for 18 customers; input/index.csv lists customer ID and page range. With Foxit PDF Editor, split it into 18 files named output/stmt_<CustomerID>.pdf. Do not include the 2 cover pages in any output.
- Platforms
mac· macOSwin· Windows- Applications
- Foxit
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicktype· Text & forms- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
18 files with exact names; each file's pages equal the index range (page content hashes); cover pages absent; Producer indicates Foxit. Foxit's JavaScript console is disabled and the VM has no PDF libraries.
Proposed partial credit
30% file count/names; 55% correct page ranges (pro-rata); 15% covers excluded.
R04.05.01.01-T07 Insert, replace pages and set roman/arabic page labels
Original instruction
In input/thesis.pdf insert input/new_ch3_figs.pdf after page 52, replace page 12 with input/p12_corrected.pdf, then set page labels so front matter (title through TOC) is i, ii, iii… and the body starts at 1. Save output/thesis_final.pdf.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clickdrag· Drag & droptype· Text & forms- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Page sequence hashes equal key; /PageLabels number tree: lowercase roman from page 1 to end of TOC, decimal starting 1 at body start; total pages correct. Acrobat's JavaScript console is disabled and the VM has no PDF libraries; Producer must be Acrobat.
Proposed partial credit
30% insertion; 25% replacement; 45% page labels.
R04.05.01.01-T08 Apply continuous Bates numbers across three productions
Original instruction
Using PDF-XChange Editor, add Bates numbers 'ACME-000001' onward to input/prod_a.pdf, prod_b.pdf and prod_c.pdf in that order, continuing across files, 9 pt Arial, bottom-right 0.4 in from edges, and save to output/ with suffix _bates.
- Platforms
win· Windows- Applications
- PDF-XChange
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicktype· Text & forms- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Text extraction per page finds the expected Bates string, sequential across the 3 files with no gaps; position bbox within 0.05 in of spec; font Arial 9 pt; page content otherwise unchanged. PDF-XChange's JavaScript console is disabled and the VM has no PDF libraries; Producer must be PDF-XChange.
Proposed partial credit
50% correct continuous numbering; 25% position; 15% font; 10% filenames.
R04.05.01.01-T09 Sign a form with a test certificate and certify it against changes
Original instruction
Open input/vendor_agreement.pdf, sign the 'Supplier signature' field with the test digital ID input/test_id.p12 (password in input/pw.txt), then save as output/vendor_agreement_signed.pdf. Sign only this copy; the original must stay unsigned.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicktype· Text & forms- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro apprisk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
pyHanko validation: one valid signature in field 'Supplier signature' by the test cert, covering the whole document, intact; original file hash unchanged; no other field signed. Acrobat's JavaScript console is disabled and the VM has no PDF libraries; Producer must be Acrobat.
Proposed partial credit
50% valid signature in right field; 25% covers whole document; 25% original untouched.
R04.05.01.01-T10 Redact names inside a chart image and sanitize hidden content
Original instruction
input/hr_review.pdf mentions 4 employees (synthetic) in text and in a bar-chart image on page 6. Redact all mentions including the chart labels, then remove metadata, comments, attachments, hidden layers and document JavaScript with Sanitize, and save output/hr_review_clean.pdf.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropvis· Visual judgmentclick· Point & click- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro apphidden· Hidden/stale state- Proposed difficulty
- 4 / 5
Proposed verifier
Text extraction and OCR find none of the 4 names; chart label regions black (≥98% pixels); no /EmbeddedFiles, /OCProperties, /JavaScript, /Annots; metadata cleared; other text intact. Acrobat's JavaScript console is disabled and the VM has no PDF libraries; Producer must be Acrobat.
Proposed partial credit
30% text mentions; 30% chart labels; 30% sanitize items; 10% no over-redaction.
R04.05.01.01-T11 Assemble a visa application packet from PDFs and photos in Preview
Original instruction
Build output/visa_packet.pdf in Preview from input/packet/: application form PDF, passport scan JPG, 3 bank statement PDFs, photo JPG, in the order given in input/checklist.txt, with all pages A4 portrait and the JPGs fitted within margins. Use only Preview.
- Platforms
mac· macOS- Applications
- Preview
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
drag· Drag & dropclick· Point & clickvis· Visual judgment- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Page order by perceptual hash equals checklist; every MediaBox A4 portrait (±1 pt); images within 10 mm margins; page count exact; Producer is Quartz PDFContext. The VM has no PDF libraries or PyObjC and Automator/Shortcuts PDF actions are disabled.
Proposed partial credit
40% order; 30% A4 normalization; 20% images fitted; 10% made in Preview.
R04.05.01.01-T12 Compare two contract versions and comment every change
Original instruction
Compare input/msa_v1.pdf and input/msa_v2.pdf in Acrobat, then in a copy of v2 add one comment per substantive difference (text changes only; ignore formatting) anchored to the changed text, and export a comment summary to output/changes_summary.pdf.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicknav· Pan/zoom/scrolltype· Text & forms- Work product
WP10· Annotations- Conditions
licensed· Licensed pro app- Proposed difficulty
- 4 / 5
Proposed verifier
Annotations in output/msa_v2_reviewed.pdf: one per seeded change (14), anchor quads overlapping the changed tokens (precision/recall ≥0.9); no comments on formatting-only changes; summary PDF exists listing them. Acrobat's JavaScript console is disabled and the VM has no PDF libraries; Producer must be Acrobat.
Proposed partial credit
60% changes found and anchored; 20% no false positives; 20% summary exported.
R04.05.01.01-T13 Crop scanner borders off thirty pages without clipping content
Original instruction
input/minutes_scan.pdf has dark scanner borders of varying width on 30 pages. Set each page's crop box to remove the borders while keeping all text and stamps, using a consistent 5 mm margin around content. Save output/minutes_cropped.pdf.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropgeom· Precise geometryvis· Visual judgment- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Rendered pages: border pixels removed (dark edge fraction <1%); OCR text identical to original (no clipped tokens); content bbox margins 5 mm ±2 mm on each side. Acrobat's JavaScript console is disabled and the VM has no PDF libraries; Producer must be Acrobat.
Proposed partial credit
40% borders removed; 40% no content clipped; 20% margin consistency.
R04.05.01.01-T14 Produce a redacted, Bates-stamped discovery set from a privilege log
Original instruction
From input/email_export.pdf (200 pages) produce output/production_vol1.pdf: redact the 37 regions listed in input/privilege_log.csv (page, description, reason code), label each redaction with its reason code, apply Bates 'DEF-0000001' onward, and export output/redaction_log.csv with Bates page and reason per redaction.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropvis· Visual judgmenttype· Text & forms- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro applong· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Each logged region: ≥98% black with overlay text equal reason code; OCR finds no privileged keywords in those regions; Bates sequential on every page; redaction_log rows match regions (Bates page + code). Acrobat's JavaScript console is disabled and the VM has no PDF libraries; Producer must be Acrobat.
Proposed partial credit
45% regions redacted (pro-rata); 15% reason overlays; 20% Bates; 20% redaction log.
R04.05.01.01-T15 Correct five figures in a statement PDF with Edit PDF
Original instruction
input/fs_draft.pdf has 5 wrong numbers listed in input/corrections.csv (page, old, new). Replace each with Acrobat's Edit PDF tool so the new value uses the same font, size and right alignment as the column, and nothing else on the page moves.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicktype· Text & formsvis· Visual judgment- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro app- Proposed difficulty
- 5 / 5
Proposed verifier
Text extraction shows new values at the old locations; fonts equal column font; right edge aligned with column (±0.5 pt); render diff vs original is non-zero only inside the 5 value bboxes. Acrobat's JavaScript console is disabled and the VM has no PDF libraries; Producer must be Acrobat.
Proposed partial credit
50% values replaced (pro-rata); 25% font/alignment; 25% no collateral pixel changes.
R04.05.01.01-T16 Build a bookmarked exhibit binder with tab pages and linked index
Original instruction
With PDF-XChange Editor, assemble output/binder.pdf from the 25 exhibits in input/exhibits/ per input/exhibit_list.csv: a generated tab page 'Exhibit N' before each, hierarchical bookmarks (Part → Exhibit), and an index page whose 25 entries link to each tab page.
- Platforms
win· Windows- Applications
- PDF-XChange
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clickdrag· Drag & droptype· Text & forms- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro applong· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Page sequence = index + (tab + exhibit pages)×25 in list order; outline tree Part→Exhibit with correct destinations; 25 /Link annotations on the index page with destinations at tab pages; tab text exact. PDF-XChange's JavaScript console is disabled and the VM has no PDF libraries; Producer must be PDF-XChange.
Proposed partial credit
30% assembly order; 25% tab pages; 25% bookmarks; 20% index links.
R04.05.01.01-T17 Control pair of T03: redact the case file with a script
Original instruction
Control pair of T03. Using PyMuPDF or qpdf from the terminal, redact all SSN-pattern numbers and the 3 names from input/case_file.pdf, apply redactions, strip metadata and annotations, and save output/case_file_redacted_cli.pdf.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- PyMuPDF
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Same checker as T03; compared with the GUI variant to measure modality gap.
Proposed partial credit
40% SSNs; 30% names; 15% no over-redaction; 15% hidden info removed.
R04.05.01.01-T18 Shrink a 180 MB scanned PDF under 15 MB keeping OCR and legibility
Original instruction
Reduce input/site_survey_scan.pdf (180 MB, 96 pages) to ≤15 MB as output/site_survey_small.pdf: images downsampled to no less than 150 dpi, a searchable OCR text layer present on every page, and page order and size unchanged.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicktype· Text & forms- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro app- Proposed difficulty
- 4 / 5
Proposed verifier
File size ≤15 MB; effective image dpi ≥150 on all pages; text layer on every page with OCR word accuracy ≥95% vs ground truth; page count/sizes unchanged; render SSIM vs original ≥0.9. Acrobat's JavaScript console is disabled and the VM has no PDF libraries; Producer must be Acrobat.
Proposed partial credit
35% size target; 25% dpi floor; 25% OCR layer; 15% fidelity.
R04.05.01.01-T19 Add a diagonal DRAFT watermark and footer to selected pages
Original instruction
In input/policy_draft.pdf add a diagonal 'DRAFT' watermark (45°, 72 pt, 30% opacity, grey) only on pages 3–10, and a footer 'Policy v0.9 – page <n>' centered on all pages, then save output/policy_draft_marked.pdf.
- Platforms
mac· macOSwin· Windows- Applications
- Acrobat · Foxit
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicktype· Text & forms- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro app- Proposed difficulty
- 3 / 5
Proposed verifier
Watermark artifacts (or OCG) only on pages 3–10 with rotation 45°, opacity 0.3, size 72; footer text on every page with correct page numbers, centered (±2 mm); original text unchanged. Acrobat/Foxit JavaScript consoles are disabled and the VM has no PDF libraries; Producer must be Acrobat or Foxit.
Proposed partial credit
45% watermark on right pages with properties; 40% footer; 15% content unchanged.
R04.05.01.01-T20 Collate a mis-fed duplex scan into correct reading order
Original instruction
input/contract_duplex_scan.pdf (60 pages) was scanned as all fronts followed by all backs in reverse order, and several backs are rotated 180°. Reorder pages into reading order, fix rotations, and remove 3 blank backs. Save output/contract_collated.pdf using Foxit PDF Editor.
- Platforms
mac· macOSwin· Windows- Applications
- Foxit
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
drag· Drag & dropvis· Visual judgmentclick· Point & click- Work product
WP09· Files & folders- Conditions
licensed· Licensed pro app- Proposed difficulty
- 4 / 5
Proposed verifier
57 pages; page hashes in key order; OCR orientation upright on all pages; blank backs removed; page count of text pages unchanged. Foxit's JavaScript console is disabled and the VM has no PDF libraries; Producer must be Foxit.
Proposed partial credit
50% correct order; 25% rotations; 15% blanks removed; 10% nothing lost.
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.