Format conversion & export fidelity
Convert or export documents between Office, iWork, LibreOffice, WPS, PDF, CSV, images and video while preserving layout, content and features.
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- Pages · Word · LibreOffice · WPS
- 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.
R04.05.01.03-T01 Export a Pages report to Word keeping styles and footnotes
Original instruction
Export input/field_report.pages from Pages to output/field_report.docx so that paragraph styles keep their names (Heading 1/2, Body), the 14 footnotes stay real footnotes, and the 3 tables stay tables. Fix anything the export flattens before re-exporting.
- Platforms
mac· macOS- Applications
- Pages · Word
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
python-docx: style names on headings/body equal mapping; 14 w:footnoteReference with correct texts; 3 w:tbl with same cell text; body text equals source (normalized). Pages export only (osascript blocked); docx app.xml from Pages exporter.
Proposed partial credit
30% styles mapped; 35% footnotes real; 20% tables; 15% text fidelity.
R04.05.01.03-T02 Save a Word report as tagged PDF/A-2b with heading bookmarks
Original instruction
Save input/annual_review.docx as output/annual_review.pdf in PDF/A-2b with document structure tags, bookmarks created from Heading 1–2, and document properties (title, author) carried over. Fonts must be embedded.
- Platforms
mac· macOSwin· Windows- Applications
- Word
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & click- Work product
WP09· Files & folders- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
veraPDF PDF/A-2b validation passes; StructTreeRoot present; outline matches Heading 1–2 tree with correct destinations; all fonts embedded; XMP title/creator equal docx properties. VBA, COM and AppleScript automation disabled, no Python in the VM; file must be saved by Word (new rsid).
Proposed partial credit
35% PDF/A-2b valid; 20% tagged; 30% bookmarks; 15% metadata.
R04.05.01.03-T03 Export a Numbers budget to Excel with live formulas intact
Original instruction
Export input/household_budget.numbers to output/household_budget.xlsx so the 3 tables keep live formulas (not values). Four Numbers-only functions do not translate; after export, fix them in Excel with equivalent formulas so every computed cell equals the Numbers result.
- Platforms
mac· macOS- Applications
- Numbers · Excel
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicktype· Text & formsmulti· Cross-app- Work product
WP02· Records/table- Conditions
multiapp· Multi-app- Proposed difficulty
- 3 / 5
Proposed verifier
openpyxl: formula cells are formulas (≥95% of source formula cells), the 4 seeded cells use Excel equivalents; LibreOffice headless recalc values equal Numbers values exported by the grader (tol 0.005); sheet/table names preserved. VBA and osascript/JXA disabled, no Python in the VM; the .xlsx must be exported by Numbers and saved by Excel.
Proposed partial credit
35% formulas preserved; 35% four translations fixed; 20% values equal; 10% names.
R04.05.01.03-T04 Export a Keynote deck to PowerPoint and repair what breaks
Original instruction
Export input/launch.key to output/launch.pptx and fix the losses in PowerPoint: substituted fonts reset to the deck's fonts (installed), 5 builds that became static restored as equivalent PowerPoint animations, and the embedded video on slide 7 playing on click.
- Platforms
mac· macOS- Applications
- Keynote · PowerPoint
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickmulti· Cross-app- Work product
WP03· Visual layout- Conditions
multiapp· Multi-app- Proposed difficulty
- 3 / 5
Proposed verifier
pptx: theme/run fonts equal source font names; p:timing has entrance effects on the 5 seeded objects in Keynote build order; slide 7 video with click trigger; slide count and text equal source. VBA and osascript/JXA disabled, no Python in the VM; the .pptx must be exported by Keynote and saved by PowerPoint.
Proposed partial credit
25% fonts; 45% animations (pro-rata over 5); 20% video; 10% text/slide count.
R04.05.01.03-T05 Edit a redlined DOCX in LibreOffice without losing revisions or comments
Original instruction
Open input/nda_redline.docx (tracked changes by 2 authors, 9 comments) in LibreOffice Writer, accept only the changes in clause 4, reply to comment 3 with 'Done', and save back as .docx so the remaining revisions and comments keep their authors and dates when opened in Word.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- LibreOffice · Word
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicktype· Text & formsnav· Pan/zoom/scroll- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
docx: w:ins/w:del outside clause 4 equal input set with authors/dates; none left in clause 4; 9 original comments + 1 reply (threaded via commentsExtended) with authors; Word COM opens without repair. Macros, UNO and the soffice CLI are disabled for the agent.
Proposed partial credit
30% clause 4 accepted only; 30% other revisions preserved; 25% comments + reply; 15% opens cleanly in Word.
R04.05.01.03-T06 Convert a WPS document to DOCX preserving headers and tables
Original instruction
Open input/inspection_record.wps in WPS Writer and save it as output/inspection_record.docx so that odd/even headers, the 4 merged-cell tables, and the page border survive. Check in Word that page count is unchanged (12).
- Platforms
win· Windowsmac· macOSlin· Linux- Applications
- WPS · Word
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
docx: evenAndOddHeaders on with distinct header parts; 4 tables with gridSpan/vMerge matching source; pgBorders present; Word COM page count 12; render SSIM vs WPS PDF ≥0.9 per page. WPS macros are disabled and no conversion CLI (soffice, pandoc) is installed in the agent VM.
Proposed partial credit
25% headers; 35% tables; 15% page border; 25% page count and render fidelity.
R04.05.01.03-T07 Import a Word proposal into Pages and match its pagination
Original instruction
Open input/proposal.docx in Pages and fix import differences so the Pages version paginates like the Word PDF (input/proposal_word.pdf): same 18 page breaks, section headers, tab-aligned price lists. Save output/proposal.pages and export output/proposal_pages.pdf.
- Platforms
mac· macOS- Applications
- Pages · Word
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgmentdrag· Drag & drop- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
PDF: 18 pages; first line of each page equals the Word PDF's first line; header text per section correct; tab-aligned columns' x positions within 3 pt of reference; text equal. osascript is blocked for the agent; the file must be saved by Pages.
Proposed partial credit
40% page breaks match; 20% headers; 25% tab alignment; 15% text fidelity.
R04.05.01.03-T08 Convert ODT to DOCX keeping TOC, captions and cross-references live
Original instruction
Convert input/methods_manual.odt to output/methods_manual.docx with LibreOffice so the table of contents, 12 figure captions (numbered) and 20 cross-references remain fields in Word (update correctly after inserting a figure), not static text.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- LibreOffice · Word
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & click- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
docx: TOC field (fldChar TOC), SEQ Figure fields for 12 captions, REF fields for 20 cross-refs with bookmarks; grader inserts a figure via Word COM, updates fields and checks numbering shifts correctly. Macros, UNO and the soffice CLI are disabled for the agent.
Proposed partial credit
25% TOC field; 30% caption SEQ fields; 30% REF cross-refs; 15% update test.
R04.05.01.03-T09 Move an Excel workbook to ODS with formats and validation intact
Original instruction
Save input/intake_tracker.xlsx as output/intake_tracker.ods in LibreOffice Calc. Conditional formats (incl. an icon set and a data bar) and 6 data validation rules must work in Calc; replace any that do not convert with Calc equivalents producing the same visible result.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- LibreOffice Calc · Excel
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP02· Records/table- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
ODS content.xml: conditional formats present on the same ranges; headless render vs Excel render per range (color/icon sampling) matches; 6 content-validations behave the same on 12 probe inputs (grader UNO). Macros, UNO and the soffice CLI are disabled for the agent.
Proposed partial credit
45% conditional formats equivalent; 40% validations; 15% data unchanged.
R04.05.01.03-T10 Restore native charts and notes after a Google Slides round trip
Original instruction
input/sales_kickoff_roundtrip.pptx went PowerPoint → Google Slides → PowerPoint; its 6 charts became images and slide notes on 4 slides were lost. Using the original input/sales_kickoff_original.pptx as reference, restore native charts (same data/format) and the lost notes, keeping any newer slide text edits.
- Platforms
mac· macOSwin· Windows- Applications
- PowerPoint · Google Slides
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clickmulti· Cross-appvis· Visual judgment- Work product
WP03· Visual layout- Conditions
hidden· Hidden/stale state- Proposed difficulty
- 4 / 5
Proposed verifier
6 slides have chart parts (not pictures) with data equal original; notes on 4 slides equal original; slide text equals round-trip version (newer edits kept); render vs original SSIM ≥0.9 for chart slides. VBA, COM and AppleScript automation disabled, no Python in the VM; file must be saved by PowerPoint (app.xml).
Proposed partial credit
50% charts restored (pro-rata over 6); 25% notes; 25% newer text kept.
R04.05.01.03-T11 Modernize a Word 97 .doc without shifting its layout
Original instruction
Convert input/procedures_1999.doc to output/procedures.docx in Word with compatibility mode off, WordArt titles converted to modern shapes, and legacy form fields converted to content controls, while pages 1–24 keep the same line breaks as input/procedures_1999.pdf.
- Platforms
win· Windowsmac· macOS- Applications
- Word
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
docx compat setting compatibilityMode 15; no legacy WordArt (VML) objects; no FORMTEXT fields, equivalent w:sdt controls; Word PDF export: per-page line-break positions equal reference on ≥22 of 24 pages. VBA, COM and AppleScript automation disabled, no Python in the VM; file must be saved by Word (new rsid).
Proposed partial credit
20% compat mode; 20% WordArt; 25% form fields; 35% layout preservation.
R04.05.01.03-T12 Export a Pages guide to a reflowable EPUB that validates
Original instruction
Export input/garden_guide.pages to output/garden_guide.epub (reflowable) with a cover from input/cover.jpg, a TOC built from Heading styles, all 18 images carrying the alt text listed in input/alt_text.csv, and book metadata (title, author, language en). It must pass epubcheck.
- Platforms
mac· macOS- Applications
- Pages
- 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
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
epubcheck 0 errors; nav.xhtml entries equal heading tree; cover image hash; 18 img alt attributes equal CSV; OPF metadata exact; reflowable (no fixed-layout rendition). osascript is blocked for the agent; the file must be saved by Pages.
Proposed partial credit
25% epubcheck clean; 25% TOC; 30% alt text; 20% cover + metadata.
R04.05.01.03-T13 Convert Word equations into editable LibreOffice Math objects
Original instruction
Open input/physics_notes.docx (21 OMML equations) in LibreOffice Writer and save as output/physics_notes.odt so every equation is an editable Math object (not an image) with the same content, and inline equations stay inline.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- LibreOffice · Word
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
ODT: 21 embedded math objects (MathML) whose normalized trees equal the OMML→MathML reference; anchor type as-char for inline ones; no image fallbacks replacing them; text unchanged. Macros, UNO and the soffice CLI are disabled for the agent.
Proposed partial credit
60% equations converted correctly (pro-rata over 21); 25% inline anchoring; 15% no images.
R04.05.01.03-T14 Rebuild a 60-page Word manual in Pages with matching pagination
Original instruction
input/user_manual.docx (60 pages, Word) must be delivered as a Pages document whose PDF paginates like input/user_manual_word.pdf: same page breaks within one line, running headers by chapter, numbered lists continuing correctly, and figure placement on the same pages. Save output/user_manual.pages and PDF.
- Platforms
mac· macOS- Applications
- Pages · Word
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgmentdrag· Drag & drop- Work product
WP01· Document- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Exported PDF: 60 pages; per page first/last line within one line of reference; chapter headers correct; list numbering sequences equal; each figure on the same page as reference. osascript is blocked for the agent; the file must be saved by Pages.
Proposed partial credit
40% pagination; 20% headers; 20% list numbering; 20% figure pages.
R04.05.01.03-T15 Migrate an Excel model to Calc and report every lost feature
Original instruction
Move input/ops_model.xlsm (pivots, slicers, 9 charts, VBA, Power Query) to output/ops_model.ods in LibreOffice Calc: pivots become working pivot tables, charts keep data and type, and output/migration_report.csv lists every feature that did not survive with sheet and object name.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- LibreOffice Calc · Excel
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
HY· Hybrid- Interaction primitives
click· Point & clickvis· Visual judgmentmulti· Cross-app- Work product
WP02· Records/table- Conditions
long· Long-horizonmultiapp· Multi-app- Proposed difficulty
- 5 / 5
Proposed verifier
ODS: data-pilot tables reproduce pivot outputs (values equal); 9 charts with same type and ranges; migration_report precision/recall vs seeded lost-feature list (slicers, VBA modules, Power Query, 1 chart subtype) ≥0.9. Macros, UNO and the soffice CLI are disabled for the agent.
Proposed partial credit
35% pivots; 30% charts; 35% migration report accuracy.
R04.05.01.03-T16 Convert a WPS official-document template to DOCX keeping the line grid
Original instruction
Open input/notice_template.wps (Chinese official-document format: A4, 22 lines × 28 characters grid, specified margins, FangSong body, red header rule) in WPS and save as output/notice_template.docx so Word shows the same grid, fonts, margins and red rule, and the sample text wraps at the same characters.
- Platforms
win· Windowsmac· macOSlin· Linux- Applications
- WPS · Word
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 5 / 5
Proposed verifier
docx sectPr: docGrid type linesAndChars with linePitch/charSpace giving 22×28; margins equal spec (±0.5 mm); eastAsia fonts; red rule as border or shape #FF0000; Word PDF line-wrap positions equal WPS PDF on all lines. WPS macros are disabled and no conversion CLI (soffice, pandoc) is installed in the agent VM.
Proposed partial credit
35% document grid; 20% margins; 20% fonts and red rule; 25% identical wrapping.
R04.05.01.03-T17 Control pair of T08: convert the ODT with headless LibreOffice
Original instruction
Control pair of T08. From the terminal, convert input/methods_manual.odt to output/methods_manual_cli.docx with soffice --headless --convert-to, then verify with a script that TOC, SEQ and REF fields survived.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- LibreOffice
- Interfaces
term· Terminal- GUI need · task setting
CL· CLI-sufficient- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
code· Code & commands- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Same checker as T08; compared against the GUI variant.
Proposed partial credit
25% TOC; 30% captions; 30% cross-refs; 15% update test.
R04.05.01.03-T18 Bring a PowerPoint deck into Keynote and fix six conversion issues
Original instruction
Open input/investor_pitch.pptx in Keynote and fix the six issues Keynote reports on import (unsupported SmartArt rendered as image, missing font, WMF graphic, 2 unsupported animations, embedded audio codec). Rebuild them natively in Keynote and save output/investor_pitch.key.
- Platforms
mac· macOS- Applications
- Keynote · PowerPoint
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgmentdrag· Drag & drop- Work product
WP03· Visual layout- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Grader exports .key to .pptx: SmartArt slide has editable shapes with same text; fonts available; graphic now vector/PNG not WMF; 2 slides have equivalent builds; audio plays (AAC); render vs original SSIM ≥0.88 per slide. osascript is blocked for the agent; the file must be saved by Keynote.
Proposed partial credit
90% six conversion issues fixed (pro-rata); 10% overall visual fidelity.
R04.05.01.03-T19 Download a Google Doc as Word and repair numbering and comments
Original instruction
Download Google Doc 'Program handbook' as .docx and open it in Word. Fix what the conversion broke: restore automatic heading numbering (1, 1.1, 1.1.1) using a multilevel list linked to Heading styles, restore page numbers in the footer, and confirm all 7 comments came through. Save output/program_handbook.docx.
- Platforms
web· Any OS (web)mac· macOSwin· Windows- Applications
- Google Docs · Word
- Interfaces
brow· Browserdesk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickmulti· Cross-app- Work product
WP01· Document- Conditions
multiapp· Multi-app- Proposed difficulty
- 3 / 5
Proposed verifier
docx: Heading 1–3 linked to one abstractNum with lvlText '%1', '%1.%2', '%1.%2.%3'; manual typed numbers removed; footer PAGE field; 7 comments with original authors; text equals Google export. VBA, COM and AppleScript automation disabled, no Python in the VM; file must be saved by Word (new rsid).
Proposed partial credit
45% heading numbering; 20% manual numbers removed; 20% page numbers; 15% comments.
R04.05.01.03-T20 Batch-convert forty legacy documents to DOCX and verify each
Original instruction
Convert the 40 legacy files in input/archive/ (.wpd, .doc, .rtf, .wps) to .docx in output/archive_docx/ using LibreOffice's Document Converter wizard, then open the 6 files flagged in input/check_list.txt and fix any lost table borders or page breaks so their page counts match input/page_counts.csv.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- LibreOffice · WPS
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
HY· Hybrid- Interaction primitives
click· Point & clicknav· Pan/zoom/scrollvis· Visual judgment- Work product
WP09· Files & folders- Conditions
long· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
40 .docx files with matching basenames that open in Word COM without repair; text similarity to source ≥0.98 each; the 6 flagged files have table borders and page counts equal page_counts.csv. LibreOffice/WPS macros, UNO and the soffice --convert-to CLI are disabled for the agent.
Proposed partial credit
40% all 40 converted cleanly (1% each); 40% six flagged files fixed; 20% text fidelity.
Mapped research references
No benchmark reference is assigned to this class in the current mapping.