Configured product purchase
Choose variants, options and delivery to meet all constraints and place a sandboxed order.
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- Amazon clone · Taobao clone · shopping sites
- 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.
R12.01.01.01-T01 Order running shoes in an exact size, width and colorway with a coupon
Original instruction
On the Omnizon clone, order 2 pairs of 'Trail Runner X' in US 10.5 Wide, colorway 'Slate', apply coupon SAVE10, ship to saved address #2 with standard shipping, and place the sandbox order with the saved test card.
- 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
click· Point & clickvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 2 / 5
Proposed verifier
Orders table: exactly 1 new order, line SKU = Trail Runner X/10.5/W/Slate qty 2, coupon SAVE10, address id 2, shipping=standard. Swatch selection is image-only (no text labels), so colorway must be read visually. Extra orders or lines fail. The sandbox store issues no API token and accepts orders only from browser sessions (CSRF + session id), so scripted checkouts are rejected.
Proposed partial credit
40% correct variant SKU; 20% qty; 20% coupon; 20% address+shipping.
R12.01.01.01-T02 Configure a laptop build to spec and stop at the payment step
Original instruction
On the laptop configurator clone, configure model 'Aero 14' with 32 GB RAM, 1 TB SSD, matte display, US keyboard backlit, 3-year accidental-damage warranty, and no preinstalled Office. Add to cart and proceed to checkout, stopping on the payment page. Report the configured price.
- Platforms
web· Any OS (web)- Applications
- configurator 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
- 2 / 5
Proposed verifier
Cart line config JSON equals the 6 required options; reported price equals backend cart total; orders table empty. Options with dependent pricing (warranty only shows after display choice) must be set in order. No API token is issued; cart and checkout writes are accepted only from the browser session (CSRF + session id), so scripted requests are rejected.
Proposed partial credit
60% config options (equal weight per option); 25% price reported exactly; 15% stopped before payment.
R12.01.01.01-T03 Buy a sofa with fabric swatch, leg finish and white-glove delivery window
Original instruction
On the furniture-store clone, order the 'Harbor 3-seat sofa' in the fabric swatch that matches fixtures/swatch_photo.jpg (choose from 18 swatches), walnut legs, and white-glove delivery in the earliest Saturday window. Place the sandbox order.
- 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
vis· Visual judgmentclick· Point & click- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Order line fabric_id equals the swatch with min ΔE2000 to the photo's center patch (unique, margin > 3), legs=walnut, delivery=white_glove, slot = earliest Saturday slot in DB. One order only. The sandbox store issues no API token and accepts orders only from browser sessions (CSRF + session id); direct HTTP posts are rejected.
Proposed partial credit
45% correct swatch; 15% legs; 25% delivery type + slot; 15% single order.
R12.01.01.01-T04 Order a phone case compatible with the phone in an old order
Original instruction
In the Taobao-style Android app clone (emulator), find which phone model the user bought in order #20250812-77 and order one MagSafe-compatible clear case for exactly that model, color 'Frost', with the free gift-wrap option and the 'Home' address. Use the in-app 'confirm receipt' flow only if asked; it is not.
- Platforms
and· Android- Applications
- Taobao 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 & clickvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Backend: new order with case SKU whose compat list includes the phone model from order #20250812-77, color Frost, gift_wrap=true, address=Home. No receipt confirmation on the old order. adb input log confirms UI interaction (no backend API token on device).
Proposed partial credit
40% model compatibility; 20% color; 20% gift wrap; 20% address + old order untouched.
R12.01.01.01-T05 Customize engraved jewelry with exact text, font and ring size
Original instruction
On the jewelry clone, order a sterling silver band, ring size 7.5, engraving 'A&J · 2026.10.14' (with the middle dot) in font 'Script 2' on the inside, gift box, and preview the engraving before checkout. Place the sandbox order.
- Platforms
web· Any OS (web)- Applications
- jewelry store clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsclick· Point & clickvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 2 / 5
Proposed verifier
Order line: size=7.5, engraving string exact Unicode match (U+00B7), font=Script 2, position=inside, gift_box=true; preview event logged before checkout. Character-count limit warning must not truncate text.
Proposed partial credit
40% engraving exact; 20% font+position; 20% size; 20% preview + gift box.
R12.01.01.01-T06 Choose the cheapest bundle that satisfies a multi-constraint camera kit
Original instruction
On the electronics clone, buy a mirrorless camera kit with: full-frame sensor, a lens covering 24-70 mm, a 128 GB V60 or faster SD card, and an extra battery; total <= $2,300 using any combination of bundles and add-ons. Choose the cheapest valid combination and place the sandbox order.
- Platforms
web· Any OS (web)- Applications
- electronics 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 judgmentnav· Pan/zoom/scroll- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Constraint checker on order lines (sensor, focal coverage, card class/capacity, battery compat) and total == minimum feasible total computed from catalog. Non-minimal valid orders get partial credit. The sandbox store issues no API token and accepts orders only from browser sessions (CSRF + session id), so scripted checkouts are rejected.
Proposed partial credit
40% all constraints satisfied; 40% minimal total; 20% single order, no extra items.
R12.01.01.01-T07 Order custom-size blinds after measuring a window in a photo
Original instruction
fixtures/window.jpg shows a window with a 1 m reference ruler on the sill. On the blinds-store clone, order one roller blind (inside mount) sized to the window width and height rounded down to the nearest 0.5 cm, fabric 'Linen Blackout', chain on the right. Place the sandbox order.
- Platforms
mac· macOSwin· Windowsweb· Any OS (web)- Applications
- blinds store clone · Preview
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
vis· Visual judgmentgeom· Precise geometrytype· Text & forms- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Order line width/height within ±1.0 cm of ground-truth dimensions (photo is rectified; reference ruler known), mount=inside, fabric, chain=right. Form only accepts sizes via spin boxes with 0.5 cm steps. The store issues no API token; orders are accepted only from the browser session (CSRF + session id), so direct POSTs of sizes are rejected.
Proposed partial credit
50% dimensions within tolerance (half each for width and height); 30% fabric+mount; 20% chain side.
R12.01.01.01-T08 Buy a bike with frame size computed from rider measurements
Original instruction
On the bike-shop clone, buy the 'Gravel One' bike for a rider with inseam 81 cm and height 178 cm using the shop's sizing chart image on the product page, pick groupset 'GRX 2x', tubeless setup add-on, and in-store pickup at the 'Mission' store on the earliest date. Place the sandbox order.
- Platforms
web· Any OS (web)- Applications
- bike shop clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
vis· Visual judgmentclick· Point & click- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Order: frame size equals chart lookup (chart only as image), groupset GRX 2x, tubeless add-on, fulfilment pickup store=Mission, date = earliest available in DB. The sandbox store issues no API token and accepts orders only from browser sessions (CSRF + session id), so scripted checkouts are rejected.
Proposed partial credit
40% frame size; 20% groupset; 15% add-on; 25% pickup store+date.
R12.01.01.01-T09 Order prescription glasses from a sandbox optical store using an Rx PDF
Original instruction
On the optical-store clone, order frame 'Kite 52' in tortoise with prescription lenses from fixtures/rx.pdf (SPH, CYL, AXIS for both eyes, PD 63), 1.67 high-index, anti-reflective and blue-light coatings. The Rx form uses dropdowns and a PD slider. Place the sandbox order with the test card.
- Platforms
web· Any OS (web)- Applications
- optical store clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
type· Text & formsdrag· Drag & dropvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Order Rx JSON equals rx.pdf values exactly (8 fields), PD=63 (slider), index 1.67, two coatings, frame color. Values set by the slider are logged as UI events; direct form POST is blocked by CSRF token per page view.
Proposed partial credit
50% Rx fields (equal weight per field); 15% PD; 20% lens options; 15% frame color.
R12.01.01.01-T10 Order a printed photo book with chosen layout and cover image
Original instruction
On the photo-print clone, create a 20-page square photo book using the 40 images in photos/2026_trip/, two per page in chronological (EXIF) order, cover = photos/cover.jpg cropped to keep both faces fully visible, title 'Lisbon 2026' on the spine. Order 2 copies to the sandbox address.
- Platforms
mac· macOSwin· Windowsweb· Any OS (web)- Applications
- photo-print clone
- Interfaces
brow· Browser- 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
WP03· Visual layout- Conditions
risk· Irreversible/riskylong· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
Project JSON: 20 pages, 2 images each, order matches EXIF order (Kendall tau >= 0.95); cover crop rectangle contains both face boxes (precomputed); spine text exact; order qty 2. No API token is issued; project saves and orders are accepted only from the browser session (CSRF + session id), so scripted uploads are rejected.
Proposed partial credit
35% page layout and order; 25% cover crop; 15% spine; 25% order placed with qty 2.
R12.01.01.01-T11 Configure a made-to-order desk on iOS with dimensions and accessories
Original instruction
In the furniture app clone on the iOS Simulator, order a standing desk with top 160 x 80 cm, color 'Oak', black frame, 2 cable trays, and a monitor arm placed on the left. The 3D preview must show the arm on the left before you add to cart. Place the sandbox order.
- Platforms
ios· iOS/iPadOS- Applications
- furniture 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 gesturesgeom· Precise geometryvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Backend order config: top 160x80, oak, black frame, cable_trays=2, monitor_arm.side=left (set by drag in 3D preview; side field only settable through the preview). Single order. The agent has no host shell, simctl or backend API token, and the backend accepts monitor_arm.side only from the app's preview-drag event.
Proposed partial credit
40% dimensions and colors; 20% trays; 30% arm placement; 10% single order.
R12.01.01.01-T12 Pick a gift with a delivery-date constraint and gift message
Original instruction
On the Omnizon clone, buy a Bluetooth speaker rated >= 4.5 with >= 200 reviews, under $80, that the product page promises to deliver by Friday 2026-10-16 to saved address 'Mom', with a gift receipt and message 'Happy birthday, Mom! — Sam'. Place the sandbox order.
- 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
click· Point & clicktype· Text & formsvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Order: SKU in the feasible set (rating, reviews, price, delivery promise <= 10-16 for that address's ZIP), address=Mom, gift_receipt=true, message exact (em dash). Delivery estimates render only in the UI after address selection.
Proposed partial credit
50% SKU feasible; 20% address; 30% gift receipt + exact message.
R12.01.01.01-T13 Order a custom-printed T-shirt batch with per-size quantities and artwork placement
Original instruction
On the print-on-demand clone, order 30 shirts in 'Heather Navy': sizes S:6, M:10, L:9, XL:5. Upload art/logo.svg, place it centered on the chest at 25 cm width, 8 cm below the collar, using the design canvas. Choose standard production and place the sandbox order.
- Platforms
mac· macOSwin· Windowsweb· Any OS (web)- Applications
- print-on-demand clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
drag· Drag & dropgeom· Precise geometrytype· Text & forms- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Order JSON: size quantities exact; artwork hash = logo.svg; placement x-centered ±0.5 cm, width 25 ±0.5 cm, top offset 8 ±0.5 cm. Placement only editable on canvas (no numeric fields). No API token is issued; designs and orders are accepted only from the browser session (CSRF + session id), so direct JSON posts are rejected.
Proposed partial credit
30% quantities; 50% placement (center, width, offset); 20% color + production.
R12.01.01.01-T14 Assemble a PC parts order that passes the store's compatibility checker
Original instruction
On the PC-builder clone, build a system around CPU 'Ryzen 7 9700X' with a compatible B650 motherboard, 32 GB DDR5-6000 (2x16), a 1000 W 80+ Gold PSU, a mid-tower that fits a 360 mm radiator on top, and that radiator. Total <= $1,450. Resolve every compatibility warning and place the sandbox order.
- Platforms
web· Any OS (web)- Applications
- PC builder clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Order lines checked by an independent compatibility model (socket, RAM type/speed QVL, case radiator support top 360, PSU wattage); builder warnings list empty at checkout; total <= 1450. The sandbox store issues no API token and accepts orders only from browser sessions (CSRF + session id), so scripted checkouts are rejected.
Proposed partial credit
90% component classes correct (6 classes, equal weight); 10% budget respected.
R12.01.01.01-T15 Use store credit and split shipping to two addresses
Original instruction
On the Omnizon clone, order 3 items from wishlist 'Holiday': ship the board game to address 'Dad' and the two candles to 'Home'. Apply the $25 store credit first and pay the remainder with the test card ending 4242. Choose the cheapest shipping per package that still arrives before Dec 20. Place the sandbox order.
- 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
click· Point & clickvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Order has 2 shipments with correct item/address mapping; store credit applied = $25; card=4242; each shipment's method is the cheapest with ETA < 12-20 per DB; wishlist unchanged otherwise. The sandbox store issues no API token and accepts orders only from browser sessions (CSRF + session id), so scripted checkouts are rejected.
Proposed partial credit
35% split shipment mapping; 20% store credit; 35% shipping choices; 10% payment method.
R12.01.01.01-T16 Order a bundle on a mobile super-app mini-program with option sheets
Original instruction
In the WeChat-style host app clone (Android), open the 'TeaHouse' mini-program and order 3 drinks for pickup at 12:15: Oolong milk tea large, 30% sugar, less ice, add pearls; two Jasmine green tea medium, no sugar, hot. Use the member coupon and stop at the payment-password screen.
- Platforms
and· Android- Applications
- WeChat clone · mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesclick· Point & click- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Mini-program backend: draft order with 3 cups and exact option sets, pickup 12:15, coupon applied, status=awaiting_payment. Mini-program renders in a webview without accessibility tree; no payment password typed (input log).
Proposed partial credit
60% drink options (one third per drink); 15% pickup time; 15% coupon; 10% stop at payment.
R12.01.01.01-T17 Order a subscription box with a skip schedule and allergen exclusions
Original instruction
On the meal-kit clone, start a weekly plan for 2 people, 3 recipes per week, excluding shellfish and peanuts, select recipes for the first week whose cards show none of those allergen icons, skip the week of Nov 23, and set delivery to Wednesdays. Place the sandbox subscription.
- Platforms
web· Any OS (web)- Applications
- meal-kit 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
- 3 / 5
Proposed verifier
Subscription row: servings=2, recipes/week=3, exclusions={shellfish, peanut}, skip_weeks contains 2026-11-23, delivery_day=Wed; week-1 recipes have no excluded allergens in DB (icons shown only as images). No API token is issued; subscriptions are accepted only from the browser session (CSRF + session id), so scripted sign-ups are rejected.
Proposed partial credit
25% plan size; 25% exclusions; 25% week-1 recipes safe; 25% skip + delivery day.
R12.01.01.01-T18 Order wallpaper rolls from wall dimensions with pattern repeat
Original instruction
Order wallpaper 'Fern Trail' (roll 0.53 x 10 m, pattern repeat 64 cm) for a room in fixtures/room_plan.pdf: cover the 3 walls marked A, B, C, net of the door and window shown, plus 10% waste. Use the store's roll calculator if helpful and verify its answer; buy whole rolls from a single batch number with enough stock. Place the sandbox order.
- Platforms
mac· macOSwin· Windowsweb· Any OS (web)- Applications
- home-decor store clone · PDF viewer
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
vis· Visual judgmenttype· Text & formsmulti· Cross-app- Work product
WP08· App/system state- Conditions
risk· Irreversible/riskymultiapp· Multi-app- Proposed difficulty
- 5 / 5
Proposed verifier
Order quantity equals reference roll count from the drop-based computation (seeded calculator undercounts by 1 because it ignores repeat); all rolls same batch id with stock >= qty. The sandbox store issues no API token and accepts orders only from browser sessions (CSRF + session id), so scripted checkouts are rejected.
Proposed partial credit
40% wall areas read correctly (checked via quantity bracket); 40% exact roll count; 20% single batch.
R12.01.01.01-T19 Reorder a discontinued variant by finding its successor via spec matching
Original instruction
Order history shows printer toner 'TX-210 Black XL' (discontinued). On the office-supply clone, find the successor or compatible cartridge for the same printer model listed in that order, prefer the manufacturer's high-yield black, quantity 2, and use the business account's PO number 'PO-7781'. Place the sandbox order.
- Platforms
web· Any OS (web)- Applications
- office supply clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- 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
- 3 / 5
Proposed verifier
Order line SKU equals the OEM high-yield black successor per compatibility table (shown only as a product-page image), qty 2, po_number exact. Third-party compatible SKUs score partial. The sandbox store issues no API token and accepts orders only from browser sessions (CSRF + session id).
Proposed partial credit
50% correct successor (half credit if compatible third-party); 20% qty; 30% PO field.
R12.01.01.01-T20 Configure and order a custom keyboard with a per-key legend layout
Original instruction
On the keyboard-builder clone, configure a 65% board: gasket mount, hot-swap PCB, linear switches 45 g, keycaps 'Cream' with legends from fixtures/layout.json (Dvorak alphas, blank modifiers), and color the Escape and Enter keys 'Rust' in the visual key editor. Place the sandbox order.
- Platforms
mac· macOSwin· Windowsweb· Any OS (web)- Applications
- keyboard builder clone
- Interfaces
brow· Browser- GUI need · task setting
G2· GUI-essential- GUI need · ordinary access
G2· GUI-essential- Interaction primitives
click· Point & clickdrag· Drag & dropvis· Visual judgment- Work product
WP08· App/system state- Conditions
risk· Irreversible/riskylong· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Order config JSON: mount, PCB, switch type/weight, keycap set; per-key legend map equals layout.json for 68 keys; key colors: Esc and Enter Rust, all others Cream. Key editor only accepts per-key clicks/drag-paint (no import). No API token is issued; the order config is accepted only from the browser session (CSRF + session id), so posting a crafted config JSON is rejected.
Proposed partial credit
20% base options; 60% per-key legends (proportional); 20% accent colors.
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.