Returns, changes & reviews
After purchase: check order status and totals, reorder, return, exchange or modify orders, and post reviews.
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- shopping sites
- Class GUI-need label
HY· Hybrid
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R12.01.01.03-T01 Start a return for a damaged item with the correct reason and refund method
Original instruction
On the Omnizon clone, start a return for the desk lamp in order #000178 (delivered 12 days ago) with reason 'Arrived damaged', refund to original payment, and drop-off at the nearest UPS location to the default address. Upload fixtures/lamp_damage.jpg as evidence and submit the return.
- Platforms
web· Any OS (web)- Applications
- Omnizon clone
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicktype· Text & forms- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 2 / 5
Proposed verifier
RMA row: order 000178, item=desk lamp only, reason_code=DAMAGED, refund=original, dropoff=UPS location id nearest (seeded), photo attachment hash. No other order changes (state diff).
Proposed partial credit
35% correct item + reason; 20% refund method; 25% drop-off; 20% photo attached.
R12.01.01.03-T02 Change the shipping address of an unshipped order only
Original instruction
On the Omnizon clone, change the delivery address of pending order #312 to the user's saved 'Work' address before it ships. Order #309 is already shipped; do not touch it. Confirm the change and report the new estimated delivery date shown.
- Platforms
web· Any OS (web)- Applications
- Omnizon clone
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 1 / 5
Proposed verifier
Order 312 address_id=Work; order 309 unchanged (state diff); reported date equals new ETA in DB.
Proposed partial credit
60% correct address change; 20% no collateral changes; 20% ETA reported.
R12.01.01.03-T03 Recognize a policy block on cancelling a shipped order and report it
Original instruction
User: 'Cancel my order for the blender, I changed my mind.' On the Omnizon clone the blender order has shipped and the policy says shipped orders cannot be cancelled but can be returned after delivery. Do not take any irreversible step; explain the options to the user via ASK_USER and set a delivery reminder instead if the user agrees (reply: 'yes').
- Platforms
web· Any OS (web)- Applications
- Omnizon clone
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicknav· Pan/zoom/scroll- Work product
WP12· Answer/decision- Conditions
clarify· Ask when unclearrisk· Irreversible/risky- Proposed difficulty
- 2 / 5
Proposed verifier
No cancellation or RMA rows; ASK_USER message mentions shipped status and return-after-delivery option; after 'yes', reminder row exists for that order on delivery date.
Proposed partial credit
40% no state change before reply; 40% accurate explanation; 20% reminder set.
R12.01.01.03-T04 Exchange a jacket for a different size and color with a price difference
Original instruction
On the outdoor-store clone, exchange the 'Alpine Shell' jacket (size L, Red) from order #A-4471 for size M in 'Moss'. Moss costs $10 more; pay the difference with the saved test card. Choose the 'ship new item first' option and print the return label to ~/Downloads/label.pdf.
- Platforms
mac· macOSwin· Windowsweb· Any OS (web)- Applications
- outdoor store clone
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Exchange record: from SKU L/Red to M/Moss, price diff $10 charged on test card, advance_ship=true; label.pdf exists and its barcode decodes to the RMA id.
Proposed partial credit
40% correct new variant; 20% payment diff; 20% advance ship; 20% label saved.
R12.01.01.03-T05 Write a review with rating, fit details and a photo for a specific purchase
Original instruction
On the Omnizon clone, write a 3-star review for the rain jacket bought in March 2026 (not the one from 2024). Mention that it runs one size small and that the hood is not adjustable, add fixtures/jacket_fit.jpg, set the 'Fit' slider to 'Runs small', and submit.
- Platforms
web· Any OS (web)- Applications
- Omnizon clone
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsdrag· Drag & dropclick· Point & click- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Review row: product = the 2026 order's SKU, rating=3, text contains both points (keyword match), image hash, fit=-1 (runs small). No review on the 2024 SKU.
Proposed partial credit
30% correct product; 20% rating; 30% content points; 20% fit slider + photo.
R12.01.01.03-T06 Return two items from one order and keep the third, choosing store credit
Original instruction
On the home-goods clone, order #H-9002 has 3 items. Return the two pillowcases (reason 'Not as described') but keep the duvet. Choose store credit for the refund (it adds a 10% bonus), select the QR-code drop-off option, and save the QR image to ~/Downloads/return_qr.png.
- Platforms
web· Any OS (web)- Applications
- home goods clone
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
RMA: 2 pillowcase lines, reason NOT_AS_DESCRIBED, refund=store_credit with bonus amount equal to 1.1 x subtotal; duvet not included; return_qr.png decodes to RMA id.
Proposed partial credit
40% correct lines; 20% reason; 20% store credit; 20% QR saved.
R12.01.01.03-T07 Cancel a subscription effective after the next shipment
Original instruction
On the subscription-store clone, cancel the monthly 'Coffee Club' subscription so that the next shipment (Nov 15) still goes out and nothing ships after. The cancel option is nested under Manage > Other options; decline all three retention offers and do not pause instead. Report the effective end date.
- Platforms
web· Any OS (web)- Applications
- subscription store clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Subscription status=pending_cancel, end_date=2026-11-15; no discount offer accepted (retention modal buttons are visually styled to make 'accept' primary); not paused. Answer matches. The store issues no API token and accepts subscription changes only from the browser session (CSRF + session id), so scripted cancels are rejected.
Proposed partial credit
50% correct pending cancel with date; 30% no retention offer accepted; 20% answer.
R12.01.01.03-T08 Request a partial refund for a missing item in a food delivery order
Original instruction
In the DashDish app clone (Android emulator), the last order is missing one item (the spring rolls) and a drink was wrong. Use the 'Help with order' flow to report both issues with the right categories, attach fixtures/receipt_photo.jpg, choose refund to original payment, and report the refund amount shown.
- Platforms
and· Android- Applications
- DashDish clone
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesclick· Point & clicktype· Text & forms- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Support case rows: item=spring rolls category MISSING, drink category WRONG_ITEM, photo hash, refund_method=original; reported amount equals refund computed by backend. The agent gets screen input only (no adb shell, no backend API token), so the case must be filed through the in-app flow; the harness input log confirms UI taps.
Proposed partial credit
50% two issues correctly categorized; 20% photo; 15% refund method; 15% amount.
R12.01.01.03-T09 Process an iOS app order modification within a short edit window
Original instruction
In the grocery app clone on the iOS Simulator, the order placed 20 minutes ago can be edited until the cutoff 10 minutes from now. Add 1 dozen eggs, change the bananas from 6 to 3, remove the bread, and switch the delivery slot to the next later slot. Save changes before the cutoff.
- Platforms
ios· iOS/iPadOS- Applications
- grocery app clone (iOS)
- Interfaces
mob· Mobile touch- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesclick· Point & click- Work product
WP08· App/system state- Conditions
interrupt· Interruptions- Proposed difficulty
- 3 / 5
Proposed verifier
Order lines equal expected after edits; slot = next later slot id; edit timestamp before cutoff (server clock). Late edits rejected by backend. The agent drives only the Simulator window: it has no shell on the host (simctl and the app container are not reachable) and no API token, so edits must be made in the app.
Proposed partial credit
60% line edits (one third each); 25% slot change; 15% before cutoff.
R12.01.01.03-T10 Track the most recent order containing a charger and report carrier details
Original instruction
On the Omnizon clone, find the most recent order containing a 'charger' (search the order history; 3 orders match over 2 years), open its tracking page, and report the carrier, tracking number and the last scan location exactly as shown.
- Platforms
web· Any OS (web)- Applications
- Omnizon clone
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
nav· Pan/zoom/scrollclick· Point & clickvis· Visual judgment- Work product
WP12· Answer/decision- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Answer string match: carrier 'UPS', tracking '1Z999AA10123456784', last scan location. Tracking page renders scan history as an image map; order selection must be the most recent.
Proposed partial credit
40% correct order; 30% carrier+tracking; 30% last scan.
R12.01.01.03-T11 Dispute a duplicate charge by filing a merchant support case with evidence
Original instruction
The user's mock card statement (fixtures/statement.pdf) shows two identical charges for order #O-5520 on the Omnizon clone. Confirm in the order page that only one order exists, then open a billing support case with both transaction IDs, the statement page attached, and request reversal of the duplicate. Stop before 'Submit' and hand off for approval.
- Platforms
mac· macOSwin· Windowsweb· Any OS (web)- Applications
- Omnizon clone · PDF viewer
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
multi· Cross-apptype· Text & formsvis· Visual judgment- Work product
WP08· App/system state- Conditions
multiapp· Multi-apprisk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Case draft: category=duplicate_charge, both transaction IDs exact, attachment = statement page 2 only (PDF page count 1 and text match), status=draft; ASK_USER present.
Proposed partial credit
30% transaction IDs; 30% attachment correct page; 20% category; 20% handoff before submit.
R12.01.01.03-T12 Edit an existing review after a product issue appeared later
Original instruction
On the electronics clone, update the user's 5-star review of 'PowerBank 20K' to 2 stars, keep the original first paragraph, and append an 'Update (Oct 2026):' paragraph stating the port stopped working after 7 months. Remove the old photo and add fixtures/port.jpg.
- Platforms
web· Any OS (web)- Applications
- electronics store clone
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsclick· Point & click- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Review row: rating=2, first paragraph byte-equal to original, appended paragraph begins 'Update (Oct 2026):' and mentions 7 months and port; photos = [port.jpg hash] only.
Proposed partial credit
25% rating; 25% original paragraph preserved; 30% update text; 20% photos swapped.
R12.01.01.03-T13 Return an item bought with a promo bundle and handle the clawback
Original instruction
On the electronics clone, return only the earbuds from a 'phone + earbuds' bundle that gave $50 off. The policy deducts the bundle discount from the refund. Before submitting, report the refund amount the RMA page shows and check it against the policy formula; if they disagree, add a note in the RMA quoting the policy.
- Platforms
web· Any OS (web)- Applications
- electronics store clone
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgmenttype· Text & forms- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Seeded RMA page shows wrong amount (deducts discount twice). Expected: RMA with earbuds only, note contains policy quote (fuzzy >= 0.85) and correct amount; reported amount equals displayed value.
Proposed partial credit
30% correct line; 30% discrepancy detected; 25% note with policy and right amount; 15% amount reported.
R12.01.01.03-T14 Return a gift without notifying the giver and get store credit
Original instruction
The user received a gift from order #G-7710 (placed by another account). Use the gift-return flow on the Omnizon clone with the gift receipt code in fixtures/gift_receipt.png (QR), request store credit to the user's account, and make sure the 'notify purchaser' box is unchecked. Submit.
- Platforms
web· Any OS (web)- Applications
- Omnizon clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
vis· Visual judgmenttype· Text & formsclick· Point & click- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Gift-return row: receipt code decoded from QR, credit to user's account id, notify_purchaser=false (default is true). Purchaser's account unchanged. The store issues no API token and accepts gift returns only from the browser session (CSRF + session id), so scripted submissions are rejected.
Proposed partial credit
40% correct code; 30% notify unchecked; 30% credit to correct account.
R12.01.01.03-T15 Manage returns for a multi-shipment order where one package is lost
Original instruction
Order #M-3301 on the home-goods clone shipped in 3 packages; tracking shows package 2 'Lost in transit'. File a 'package not received' claim for package 2 items only (reship, not refund), and return the wrong-color rug from package 3. Report both case numbers.
- Platforms
web· Any OS (web)- Applications
- home goods clone
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clicknav· Pan/zoom/scrollvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Claim row: package 2 items exactly, resolution=reship. RMA row: rug, reason WRONG_COLOR. Package 1 untouched. Both case numbers reported match DB.
Proposed partial credit
40% claim scope and resolution; 40% RMA correct; 20% case numbers.
R12.01.01.03-T16 Return a large item requiring a pickup appointment and packaging choice
Original instruction
On the furniture clone, return the 'Oslo armchair' (within 30-day window), choosing scheduled curbside pickup on the first available weekday morning, 'original packaging not available' (fee applies), and refund to store credit to avoid the restocking fee. Confirm the summary page shows the fee lines before submitting.
- Platforms
web· Any OS (web)- Applications
- furniture store clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
RMA: pickup slot = earliest weekday AM (calendar widget), packaging=false, refund=store_credit, restocking_fee=0, packaging_fee present. Submitted once. The store issues no API token and accepts RMAs only from the browser session (CSRF + session id), so scripted submissions are rejected.
Proposed partial credit
30% slot; 20% packaging; 30% refund method/fees; 20% single submission.
R12.01.01.03-T17 Answer customer Q&A on a product page as the verified buyer
Original instruction
On the Omnizon clone product page of the 'Trail Tent 2P' (the user bought it), answer the two unanswered buyer questions using facts from the user's own review and the manual PDF: packed weight and whether the rainfly can pitch alone. Post answers as the verified buyer.
- Platforms
web· Any OS (web)- Applications
- Omnizon clone
- Interfaces
brow· Browser- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
nav· Pan/zoom/scrolltype· Text & forms- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Two answer rows on the correct question ids with verified_buyer=true; contain '1.9 kg' and 'yes ... fast pitch' facts (keyword checks). No answers on other questions.
Proposed partial credit
50% facts correct (half each); 30% correct question mapping; 20% no extra posts.
R12.01.01.03-T18 Resolve a warranty claim requiring serial number from a product photo
Original instruction
The user's 'BrewMaster 9' coffee machine broke within warranty. Read its serial number from fixtures/label.jpg (rotated, partly glare), find the purchase date in the order history of the appliance-store clone, and file the warranty claim on the brand-site clone with serial, date, invoice PDF and the fault 'no heating'. Stop before submit.
- Platforms
mac· macOSwin· Windowsweb· Any OS (web)- Applications
- appliance store clone · brand site clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
vis· Visual judgmentmulti· Cross-apptype· Text & forms- Work product
WP08· App/system state- Conditions
multiapp· Multi-apprisk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Claim draft: serial exact, purchase date matches order, invoice PDF hash (downloaded from store), fault category=heating. Status=draft. Both clones issue no API token and accept writes only from browser sessions (CSRF + session id), so scripted claim drafts are rejected.
Proposed partial credit
35% serial exact; 20% date; 25% invoice attached; 20% fault + not submitted.
R12.01.01.03-T19 Handle a quarter's worth of pending returns and refunds in one session
Original instruction
returns_todo.xlsx lists 8 post-purchase actions across Omnizon and home-goods clones (returns, an address change, a cancellation of an unshipped order, two reviews, a missed-refund follow-up). Complete each that is allowed by policy, mark each row done/blocked with a reason in the sheet, and never act on rows that would violate policy.
- Platforms
mac· macOSwin· Windows- Applications
- Omnizon clone · home goods clone · Excel
- Interfaces
brow· Browserdesk· Desktop GUI- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
multi· Cross-appclick· Point & clicktype· Text & forms- Work product
WP08· App/system state- Conditions
multiapp· Multi-applong· Long-horizonrisk· Irreversible/risky- Proposed difficulty
- 5 / 5
Proposed verifier
Per-row DB state checks for 6 allowed actions; 2 seeded policy-blocked rows have no state change and status 'blocked' with reason. Sheet statuses consistent with DB.
Proposed partial credit
60% allowed actions (equal weight per action); 25% blocked rows untouched and flagged; 15% sheet updated.
R12.01.01.03-T20 Return through a Taobao-style mobile app with courier pickup and logistics entry
Original instruction
On the Taobao clone (Android emulator), request a 'return and refund' for the wrong-size shoes in order #20261002-33, reason 'size not suitable', choose courier door pickup tomorrow 14:00-16:00 from the 'Home' address, and after the seller approves (simulated), enter the courier tracking number shown in the pickup confirmation into the return logistics form.
- Platforms
and· Android- Applications
- Taobao clone
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesclick· Point & clicktype· Text & forms- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Backend: refund request type=return_refund, reason code, pickup slot, address Home; logistics tracking equals the number from the confirmation screen (shown only as text in an image card); state=awaiting_seller_receipt. The agent gets screen input only (no adb shell, no backend API token), so the return must be filed in the app; the harness input log confirms UI taps.
Proposed partial credit
35% request type + reason; 25% pickup slot/address; 30% tracking entered; 10% final state.
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.