Precise in-place text editing & selection
Select exact spans and edit rich text in place without collateral formatting damage.
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- Word · Google Docs · Pages
- Class GUI-need label
G3· GUI-only
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R04.02.01.04-T01 Fix six typos located by rendered page and line in a Word report
Original instruction
Open input/annual_report.docx in Word. input/typos.txt lists 6 errors only by rendered position (e.g. 'p.3 line 12: recieve'). Correct each word in place, leaving every other character, run format and paragraph untouched. Save in place as .docx.
- Platforms
mac· macOSwin· Windows- Applications
- 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
- 2 / 5
Proposed verifier
Word-level diff vs reference: exactly the 6 target tokens changed; run properties (rPr) of all other runs byte-identical. Positions exist only in the rendered layout. File must carry a new Word rsid session and Word app.xml; zip rewrites by python-docx (lost rsids, reordered parts) fail. VBA, COM and AppleScript automation disabled, no Python in the VM.
Proposed partial credit
15% file saved by Word; 70% six typos fixed at the right location (equal shares); 15% zero collateral character/format changes.
R04.02.01.04-T02 Suggest 'shall'→'must' edits only in clauses 4.1–4.3 of a Google Doc
Original instruction
In the shared Google Doc 'MSA draft v7' (test account), switch to Suggesting mode and change 'shall' to 'must' only inside clauses 4.1, 4.2 and 4.3 (5 occurrences). Occurrences elsewhere, including 'shall' inside defined-term quotes in 4.2, stay unchanged. Do not accept your suggestions.
- Platforms
web· Any OS (web)- Applications
- Google Docs
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- 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
Grader reads the doc with a read-only Docs API token: exactly 5 suggestedDeletion/suggestedInsertion pairs on the expected ranges, authored by the test user; body text otherwise identical. Agent has no API credentials and Apps Script is disabled for the test account; edits must appear in revision history as editor UI edits.
Proposed partial credit
20% Suggesting mode used (no direct edits); 60% five correct suggestions (equal shares); 20% defined-term quote and out-of-scope occurrences untouched.
R04.02.01.04-T03 Recolor every body-text 'WARNING' in a Pages manual, not captions
Original instruction
Open input/pump_manual.pages in Pages. Make every occurrence of the uppercase word 'WARNING' in body text bold and color #C00000. Occurrences of 'Warning' in figure captions and the header must keep their current style. Save the document.
- Platforms
mac· macOS- Applications
- Pages
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
click· Point & clicktype· Text & forms- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Grader exports the saved .pages to .docx via its own AppleScript after the session and checks run color and bold on all 9 body occurrences; caption/header runs unchanged. osascript is disabled in the agent session; file must be saved by Pages (IWA version metadata).
Proposed partial credit
60% body occurrences recolored and bold (pro-rata); 25% captions/header untouched; 15% no other text changed.
R04.02.01.04-T04 Track-change the governing-law swap and one clause deletion in Word
Original instruction
In input/supply_contract.docx set the Word user name to 'Reviewer A', turn on Track Changes, replace 'New York' with 'Delaware' in its 3 governing-law places (not in the notice address) and delete only the second sentence of clause 9.2. Save as output/supply_contract_ra.docx.
- Platforms
mac· macOSwin· Windows- Applications
- 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 & formskey· Shortcuts- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Parse w:ins/w:del: exactly 3 replace pairs and 1 deletion, all author 'Reviewer A', spans equal reference ranges; accepted-view text equals reference; notice address still 'New York'. Hand-written revision XML is detected by missing rsid/date consistency. VBA, COM and AppleScript automation disabled, no Python in the VM.
Proposed partial credit
15% Track Changes with correct author; 45% three replacements (equal thirds); 25% exact sentence deletion; 15% no stray edits.
R04.02.01.04-T05 Bold one phrase and add one link in a WordPress post editor
Original instruction
In the sandbox WordPress site, open draft post 'Autumn promo' in the Classic (TinyMCE) editor. Bold exactly the phrase 'limited time' in paragraph 2 and make 'pricing page' in paragraph 3 a link to /pricing. Save the draft without publishing.
- Platforms
web· Any OS (web)- Applications
- WordPress
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicktype· Text & forms- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Saved post_content HTML diff vs reference: exactly one <strong> around 'limited time' and one <a href='/pricing'> around 'pricing page'; Levenshtein outside target spans = 0; post_status still 'draft'. REST API and wp-cli are disabled for the agent.
Proposed partial credit
40% bold span exact; 40% link span and href exact; 20% no other markup changed and post not published.
R04.02.01.04-T06 Swap two sentences while keeping the footnote reference attached
Original instruction
In input/policy_brief.docx, paragraph 3 on page 2 has two sentences in the wrong order; the first carries footnote 4. Swap the two sentences so the footnote reference stays at the end of its original sentence and footnote text is unchanged. Keep spacing to one space after each period. Save in place.
- 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
drag· Drag & dropclick· Point & clickkey· Shortcuts- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Paragraph text equals reference order; w:footnoteReference id 4 sits immediately after the moved sentence; footnotes.xml unchanged; footnote count unchanged (no duplicate created by copy-paste). Document saved by Word (new rsid). VBA, COM and AppleScript automation disabled, no Python in the VM.
Proposed partial credit
40% sentence order correct; 35% footnote reference attached and not duplicated; 25% spacing and rest of document unchanged.
R04.02.01.04-T07 Strike through four discontinued SKUs inside a comma list in Google Docs
Original instruction
In the Google Doc 'Spring catalogue notes', paragraph 'Available colours' lists 14 SKUs separated by ', '. fixtures/discontinued.txt names 4 of them. Apply strikethrough to exactly those 4 SKU codes, not to the commas or spaces around them, and leave all other formatting as is.
- Platforms
web· Any OS (web)- Applications
- Google Docs
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickdrag· Drag & drop- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Read-only Docs API: textRun ranges with strikethrough=true equal exactly the 4 SKU character ranges; no strikethrough on separators; no other textStyle changes. Agent has no API token and Apps Script is disabled; revision history must show UI edits by the test user.
Proposed partial credit
80% four SKUs struck (equal shares, exact boundaries required); 20% no separators or other runs changed.
R04.02.01.04-T08 Block-select and delete line-number columns from a code listing in Word
Original instruction
input/firmware_spec.docx contains a monospaced code listing on page 5 whose lines 10–40 start with a 4-character line-number prefix ('0010'). Using column (block) selection, remove those 4 characters from lines 10–40 only, keeping indentation after the prefix and all other lines intact. Save in place.
- 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
drag· Drag & dropkey· Shortcutsgeom· Precise geometry- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Lines 10–40 of the listing equal reference (prefix removed, indentation preserved); lines 1–9 and 41+ byte-identical; listing style unchanged. Word session rsid required; listing positions are given by page, so text extraction scripts cannot locate them without rendering. VBA, COM and AppleScript automation disabled, no Python in the VM.
Proposed partial credit
20% correct line range targeted; 60% prefixes removed on all 31 lines (pro-rata); 20% indentation and surrounding lines intact.
R04.02.01.04-T09 Apply small caps to capitalized defined terms in clauses 1–5 only
Original instruction
In input/licence_agreement.docx apply Small Caps character formatting to every capitalized use of the defined terms 'Agreement', 'Licensor' and 'Licensee' in clauses 1–5 body text. Leave headings, lowercase uses ('the licensee's staff'), and clauses 6+ untouched. Do not change the font or size. Save in place.
- 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 & clicktype· Text & forms- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
w:smallCaps set on exactly the 23 expected runs (ranges match reference); headings, lowercase uses and clauses 6+ have no smallCaps; font/size unchanged. Run splitting by Word expected; file must be saved by Word (rsid). VBA, COM and AppleScript automation disabled, no Python in the VM.
Proposed partial credit
70% target runs set (pro-rata over 23); 20% no false positives in headings/lowercase/out-of-scope clauses; 10% font and size unchanged.
R04.02.01.04-T10 Paste clause text from an Outlook email using destination formatting
Original instruction
Open input/vendor_reply.msg in Outlook and copy the 3 paragraphs under 'Revised clause 7'. Replace the current clause 7 body in input/msa.docx with them so they adopt the existing 'Clause Body' style: no imported styles, fonts or colours from the email. Keep the clause 7 heading and numbering. Save in place.
- Platforms
mac· macOSwin· Windows- Applications
- Word · Outlook
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
CL· CLI-sufficient- Interaction primitives
multi· Cross-appclick· Point & clickkey· Shortcuts- Work product
WP01· Document- Conditions
multiapp· Multi-app- Proposed difficulty
- 3 / 5
Proposed verifier
Clause 7 text equals the email paragraphs (normalized whitespace); each paragraph style 'Clause Body'; no new styles in styles.xml; no direct rFonts/color in the pasted runs; numbering ids unchanged. Clipboard paste from Outlook required; file saved by Word. VBA, COM and AppleScript automation disabled, no Python in the VM.
Proposed partial credit
35% correct text in place; 35% destination style with no foreign styles/fonts; 15% heading and numbering intact; 15% rest unchanged.
R04.02.01.04-T11 Convert straight quotes to curly in Pages body text but not in code
Original instruction
input/api_tutorial.pages mixes straight quotes in prose and in code samples styled 'Code'. Convert all straight double and single quotes in prose to typographic (curly) quotes, apostrophes included, and leave every quote inside 'Code' paragraphs and inline 'Code' spans straight. Save.
- Platforms
mac· macOS- Applications
- Pages
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicktype· Text & formskey· Shortcuts- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Grader-side AppleScript export to .docx: all prose quote characters are U+2018/2019/201C/201D with correct opening/closing direction; all 'Code' style runs keep U+0022/U+0027; no other characters changed. osascript blocked in agent session.
Proposed partial credit
50% prose quotes converted with correct direction; 35% code quotes untouched; 15% no other edits.
R04.02.01.04-T12 Edit an exponent and subscript inside a Google Docs equation
Original instruction
In the Google Doc 'Heat transfer notes', equation 3 reads q = k A ΔT / L^2 and should be q_x = k A ΔT / L. Using the built-in equation editor, change the left side to q with subscript x and remove the exponent on L, leaving equations 1, 2 and 4 untouched. Do not replace the equation with an image or plain text.
- Platforms
web· Any OS (web)- Applications
- Google Docs
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicktype· Text & formskey· Shortcuts- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Grader exports the doc to .docx and parses OMML: equation 3 math tree equals reference (m:sSub for q_x, no m:sSup on L); still an equation object, not an image/text run; other equations identical. Docs API cannot author equations; agent has no API token.
Proposed partial credit
40% subscript q_x correct; 40% exponent removed with fraction intact; 20% remains a native equation and others unchanged.
R04.02.01.04-T13 Remove only yellow 'Legal' highlights that start mid-word
Original instruction
input/handbook_review.docx has yellow highlights from Legal and green highlights from HR; several yellow spans begin or end mid-word and some overlap green ones. Remove every yellow highlight and keep all green highlights exactly as they are, without changing text or other formatting. Save in place.
- 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 & clickvis· Visual judgment- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
No w:highlight val='yellow' remains; set of green-highlighted character ranges equals reference exactly; text and other rPr unchanged. Saved by Word (rsid). VBA, COM and AppleScript automation disabled, no Python in the VM.
Proposed partial credit
45% all yellow removed; 40% green ranges preserved exactly; 15% no text/format collateral.
R04.02.01.04-T14 Resolve a three-reviewer redline in a 120-page tender by rule
Original instruction
input/tender_v5.docx (120 pages) has tracked changes by Mary Chen, Raj Patel and Ana Ruiz plus 31 comments. Accept all of Mary's formatting changes, reject Raj's deletions in sections 7–9 only, accept everything else in section 7, leave all other revisions pending, reply 'Agreed' to Ana's two comments in 8.4 and resolve them. Save in place.
- Platforms
mac· macOSwin· Windows- Applications
- Word
- Interfaces
desk· Desktop GUI- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicknav· Pan/zoom/scrollhover· Hover & context- Work product
WP01· Document- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Remaining w:ins/w:del/w:rPrChange set equals reference by id, author and range; accepted text in sections 7–9 equals reference; commentsExtended shows 2 replies 'Agreed' and done=1 on Ana's 8.4 comments; other 29 comments untouched. Saved by Word. VBA, COM and AppleScript automation disabled, no Python in the VM.
Proposed partial credit
25% Mary formatting accepted; 25% Raj deletions in 7–9 rejected; 35% rest of section 7 accepted and other revisions still pending; 15% replies + resolve.
R04.02.01.04-T15 Apply 15 comment-requested micro-edits as suggestions and resolve them
Original instruction
The 40-page Google Doc 'Grant narrative v3' has 15 open comments, each asking for one small edit at its anchored text ('change 2023 to 2024', 'drop this comma'). In Suggesting mode, make each requested edit exactly at the anchored range, then resolve that comment. Leave the 4 comments that ask questions open and unanswered.
- Platforms
web· Any OS (web)- Applications
- Google Docs
- Interfaces
brow· Browser- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicktype· Text & formsnav· Pan/zoom/scroll- Work product
WP01· Document- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Read-only Docs/Drive API: 15 suggestions whose ranges overlap the comment anchors and whose content matches the requested edit; those 15 comments resolved; 4 question comments open with no replies; no other body changes. Agent has no API token and Apps Script is disabled; revision history must show UI edits.
Proposed partial credit
60% correct suggestions (pro-rata over 15); 20% the 15 comments resolved; 10% question comments untouched; 10% no stray edits.
R04.02.01.04-T16 Edit copy in linked Pages text boxes without causing overflow
Original instruction
input/clinic_brochure.pages has a story flowing through 3 linked text boxes. Replace paragraph 2 with the text in input/new_para2.txt and shorten paragraph 4 by removing its last sentence, so the story still ends inside box 3 with no overflow indicator and no text moved to a new box. Keep all styles. Save.
- Platforms
mac· macOS- Applications
- Pages
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicktype· Text & formsvis· Visual judgment- Work product
WP03· Visual layout- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Grader exports PDF and .docx via its own AppleScript: story text equals reference; the PDF shows the final word inside box 3 bounds; box count and link chain unchanged; paragraph styles unchanged. osascript blocked in agent session.
Proposed partial credit
35% paragraph 2 replaced exactly; 25% paragraph 4 sentence removed; 25% no overflow and chain intact; 15% styles unchanged.
R04.02.01.04-T17 Update text inside content controls without breaking fields
Original instruction
input/offer_letter_template.docx has 6 rich-text content controls (Candidate, Role, Salary, Start, Manager, Location) and DATE and REF fields. Fill the controls with values from input/candidate.json by typing inside each control, keeping the controls themselves, their tags and the fields as live fields. Save as output/offer_rivera.docx.
- 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 & clicktype· Text & forms- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
w:sdt count stays 6 with unchanged w:tag values and sdtContent text equal to candidate.json; fldSimple/fldChar for DATE and REF still present (not converted to text); body outside controls unchanged. File must be saved by Word (rsid). VBA, COM and AppleScript automation disabled, no Python in the VM.
Proposed partial credit
50% six controls filled correctly (pro-rata); 25% controls and tags preserved; 25% fields intact and no outside edits.
R04.02.01.04-T18 Use iPad touch selection handles to restyle one sentence in Pages
Original instruction
On iPad, open 'Field Notes.pages' in Pages. Using touch selection handles, select exactly the second sentence of paragraph 4 and apply the character style 'Emphasis'. Then drag-move the word 'tentatively' from paragraph 5 to just before 'confirmed' in paragraph 6. Keep everything else the same.
- Platforms
ios· iOS/iPadOS- Applications
- Pages
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
touch· Touch gesturesdrag· Drag & drop- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Document synced via iCloud and exported by the grader to .docx: 'Emphasis' style applied to exactly the sentence range; 'tentatively' absent from paragraph 5 and present once before 'confirmed' with single spaces; all other text identical. Shortcuts is disabled on the agent's iPad and it has no Mac or iCloud web access; the file must be saved by Pages for iPadOS (device in app metadata).
Proposed partial credit
45% exact sentence styled; 40% word moved with correct spacing; 15% no other changes.
R04.02.01.04-T19 Edit a mixed Arabic–English clause without scrambling bidi order
Original instruction
input/bilingual_lease.docx has right-to-left Arabic clauses with embedded English names and numbers. In clause 5 replace the Arabic amount phrase with the one in input/clause5_fix.txt and change the English tenant name 'Omar Haddad' to 'Omar Al-Haddad' inside the same RTL paragraph. Keep paragraph direction RTL and all other text. Save.
- 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 & clicktype· Text & formskey· Shortcutsvis· Visual judgment- Work product
WP01· Document- Conditions
- None declared
- Proposed difficulty
- 5 / 5
Proposed verifier
Clause 5 logical-order text equals reference (Unicode string compare); w:bidi on the paragraph and w:rtl on Arabic runs preserved; no stray LRM/RLM marks added; other clauses byte-identical. Visual check: grader renders PDF and compares clause 5 line images to reference (SSIM ≥0.95). VBA, COM and AppleScript automation disabled, no Python in the VM; file must be saved by Word (new rsid).
Proposed partial credit
35% Arabic phrase replaced; 30% English name changed in place; 20% direction properties intact; 15% rendered order matches reference.
R04.02.01.04-T20 Control pair of T01: fix the same typos by script from exact strings
Original instruction
Control pair of T01. Given input/annual_report.docx and input/typos_exact.txt (6 lines 'wrong→right' with paragraph index), write and run a script (python-docx or similar) that corrects the 6 words in place without changing run formatting, and saves output/annual_report_fixed.docx.
- Platforms
mac· macOSwin· Windowslin· Linux- Applications
- Word · python-docx
- 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 T01 minus the Word-session requirement: exactly 6 tokens changed, all other runs' rPr identical. Used to contrast CLI success with the GUI variant that only gives rendered page/line positions.
Proposed partial credit
70% six typos fixed (pro-rata); 30% no collateral run or text changes.
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.