Rebuild a site from browsing it
Recreate a website's behavior after exploring it only through the browser (WebsiteBench-style).
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- Chrome · Node
- 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.
R02.04.01.02-T01 Rebuild a pet-adoption site with filters, favourites and an application form
Original instruction
Explore the reference site `http://ref-petadopt.local` only through the provided browser (view-source, DevTools and network inspection are disabled). Rebuild it in `/work/petadopt` (any stack, Docker Compose) with the same search filters, pet detail pages, favourites that persist across server restarts, and the adoption form's validation rules.
- Platforms
web· Any OS (web)- Applications
- Chrome · Node · Docker
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrollvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Hidden L1-L3 tests: ~150 Request/Browser/System journeys (Playwright) replayed on the rebuild; SSIM >= 0.75 at 10 checkpoints; favourites survive container restart. The reference is reachable only via a locked-down browser profile (no DevTools, no curl route).
Proposed partial credit
30% L1 page/request tests; 35% L2 journeys; 20% L3 persistence/edge; 15% visual checkpoints
R02.04.01.02-T02 Recreate only the checkout flow of a retail replica
Original instruction
Using only the locked-down browser, explore the checkout of `http://ref-retail.local` (cart, shipping options, address validation, promo codes, order review, confirmation page and order history). Rebuild just this flow in `/work/checkout` against the provided product seed, matching pricing, tax and validation behavior.
- Platforms
web· Any OS (web)- Applications
- Chrome · Node · Postgres
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrollvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
~120 hidden Playwright journeys: totals, taxes and shipping per address equal reference; promo stacking rules; validation messages exact; SSIM >= 0.75 at 6 checkpoints. Reference reachable only via the locked browser.
Proposed partial credit
40% totals/tax/shipping; 25% validation; 20% promos; 15% visuals
R02.04.01.02-T03 Reproduce RSVP and waitlist logic of an events site
Original instruction
Explore `http://ref-meetup.local` in the locked-down browser with 3 test accounts. Rebuild event pages, RSVP with capacity limits, waitlist promotion when someone cancels, host approval mode and the notification emails (to the sandbox inbox) in `/work/meetup-clone`.
- Platforms
web· Any OS (web)- Applications
- Chrome · Python/Node · MailHog
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrollvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
long· Long-horizon- Proposed difficulty
- 4 / 5
Proposed verifier
Hidden multi-user journeys (L3): capacity edge cases, waitlist ordering, cancellation promotes next, approval mode, email subjects/bodies in sandbox inbox match reference templates; persistence across restart.
Proposed partial credit
30% RSVP basics; 35% waitlist logic; 15% approval; 20% emails
R02.04.01.02-T04 Rebuild a six-page static brochure site from browsing
Original instruction
Browse `http://ref-bakery.local` (6 pages: home, menu, about, locations, catering, contact) in the locked browser and rebuild it as a static site in `/work/bakery` with the same navigation, content, images (download via the browser's Save Image), and responsive layout at 1280 and 390 px.
- Platforms
web· Any OS (web)- Applications
- Chrome · static site
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrollvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
For 6 pages × 2 widths: text similarity >= 0.95 and SSIM >= 0.8 vs reference renders; all nav links resolve; images present with matching perceptual hashes.
Proposed partial credit
40% content; 40% visual (scaled); 20% navigation
R02.04.01.02-T05 Replicate a recipe search with filters and sort behavior
Original instruction
Explore the search on `http://ref-recipes.local` in the locked browser: free-text search with typo tolerance, filters (diet, time, cuisine) with counts, sort orders and URL query parameters. Rebuild it in `/work/recipes` over the provided recipe dataset so results and counts match for any query.
- Platforms
web· Any OS (web)- Applications
- Chrome · Node · SQLite
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrollvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden 200 queries: top-20 result order Kendall tau >= 0.9 and facet counts exact vs reference; URLs round-trip state; typo-tolerance cases (edit distance 1) return the same top-5.
Proposed partial credit
40% result ordering; 25% facet counts; 20% typo tolerance; 15% URL state
R02.04.01.02-T06 Reproduce a loan calculator web app's numeric behavior
Original instruction
Use `http://ref-loancalc.local` in the locked browser to discover how it computes payments, amortization tables, extra-payment effects and rounding. Rebuild it in `/work/loancalc` so the same inputs produce identical outputs and the same validation messages.
- Platforms
web· Any OS (web)- Applications
- Chrome · browser
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
type· Text & formsvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Hidden 300 input sets entered via Playwright on both sites: every displayed number equal; validation messages exact for 20 invalid inputs; layout SSIM >= 0.75.
Proposed partial credit
60% numeric parity; 25% validation; 15% layout
R02.04.01.02-T07 Rebuild a cinema showtime and seat-booking site
Original instruction
Explore `http://ref-cinema.local` with the locked browser: theater/showtime selection, interactive seat map (accessible seats, pairs, seat-gap rule), ticket types and prices, 10-minute seat holds and booking confirmation emails. Rebuild in `/work/cinema` with concurrency-safe seat holds.
- Platforms
web· Any OS (web)- Applications
- Chrome · Node · Postgres
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrollvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Hidden journeys incl. 2 concurrent users racing for the same seats (exactly one wins), seat-gap rule enforced identically, hold expiry after 10 min (clock hook), totals equal, seat map SSIM >= 0.75.
Proposed partial credit
25% showtime browsing; 35% seat rules; 20% concurrency/holds; 20% booking + emails
R02.04.01.02-T08 Recreate a blogging platform's editor with drafts and comments
Original instruction
Explore `http://ref-blog.local` with two test accounts in the locked browser: the post editor (rich text, autosave drafts, scheduled publish), tags, public post pages, comments with moderation and author notifications. Rebuild in `/work/blog-clone`.
- Platforms
web· Any OS (web)- Applications
- Chrome · Node · SQLite
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrolltype· Text & formscode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Hidden L1-L3 journeys: autosave restores after reload, scheduled posts appear at time (clock hook), comment moderation visible to author only until approved, other-user permissions; content rendering SSIM >= 0.75.
Proposed partial credit
35% editor + drafts; 20% publishing; 30% comments/moderation; 15% visuals
R02.04.01.02-T09 Rebuild a Kanban board with drag-and-drop and WIP limits
Original instruction
Explore `http://ref-kanban.local` in the locked browser: boards, columns with WIP limits, card drag-and-drop between columns and reordering, card details with checklists, and activity history. Rebuild it in `/work/kanban` with the same drag behavior and limits.
- Platforms
web· Any OS (web)- Applications
- Chrome · React · Node
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
drag· Drag & dropvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Hidden Playwright drag journeys: card order after drags equals reference; WIP limit blocks drop with the same message; checklist progress; activity log entries; state persists after reload.
Proposed partial credit
40% drag/reorder parity; 20% WIP limits; 20% card details; 20% persistence + activity
R02.04.01.02-T10 Reproduce a hotel booking date picker and pricing rules
Original instruction
Use `http://ref-hotel.local` in the locked browser to learn its date-range picker (min stay, blackout dates, check-in days), nightly pricing (weekends, length-of-stay discounts) and occupancy rules. Rebuild the search and booking quote in `/work/hotel` over the provided room data.
- Platforms
web· Any OS (web)- Applications
- Chrome · Node
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
click· Point & clickvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden 150 date/occupancy queries: availability and total price equal reference; disabled dates in picker identical for 3 months (DOM aria-disabled); error messages exact.
Proposed partial credit
40% pricing parity; 30% availability/picker rules; 20% occupancy; 10% messages
R02.04.01.02-T11 Rebuild a forum with voting, ranking and nested replies
Original instruction
Explore `http://ref-forum.local` with 3 test accounts in the locked browser: posting, nested replies, up/down votes with self-vote rules, 'hot' and 'top' rankings, and karma display. Rebuild it in `/work/forum` with matching ranking behavior.
- Platforms
web· Any OS (web)- Applications
- Chrome · Python · Postgres
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrollvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden multi-user scenario (500 scripted actions): hot/top orderings equal reference (Kendall tau >= 0.95); vote rules; karma totals exact; reply nesting identical.
Proposed partial credit
30% posting/replies; 20% voting rules; 35% ranking parity; 15% karma
R02.04.01.02-T12 Recreate a weather dashboard's behavior over a mock API
Original instruction
Browse `http://ref-weather.local` in the locked browser (it calls a mock weather API you can also query at `http://mockwx.local`): city search with autocomplete, unit toggle, 7-day forecast, hourly chart and saved locations. Rebuild it in `/work/weather` against the same mock API.
- Platforms
web· Any OS (web)- Applications
- Chrome · Vite
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrollvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden Playwright tests: displayed values (with unit conversions and rounding) equal reference for 20 cities; autocomplete suggestions identical; saved locations persist (localStorage); layout SSIM >= 0.75.
Proposed partial credit
40% values + units; 20% autocomplete; 20% saved locations; 20% visuals
R02.04.01.02-T13 Replicate a multi-step signup with exact validation messages
Original instruction
Explore the signup wizard on `http://ref-signup.local` in the locked browser: 4 steps with field rules (password policy, username availability check, phone formats by country, age gate) and step navigation. Rebuild it in `/work/signup` with identical messages and step behavior.
- Platforms
web· Any OS (web)- Applications
- Chrome · Node
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
type· Text & formsvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden 120 input scenarios: validation messages and step transitions identical; created accounts in DB match reference schema; back/forward keeps entered data; SSIM >= 0.75 per step.
Proposed partial credit
45% validation parity; 25% navigation/state; 15% account creation; 15% visuals
R02.04.01.02-T14 Discover and reproduce a shop's coupon and bundle pricing rules
Original instruction
The cart at `http://ref-deals.local` applies undocumented coupon rules (stacking limits, category exclusions, bundle discounts, minimum spend, first-order codes). Discover them by experimenting in the locked browser with the test account and rebuild the cart pricing engine and UI in `/work/deals` so totals always match.
- Platforms
web· Any OS (web)- Applications
- Chrome · Node
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
click· Point & clickvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
hidden· Hidden/stale state- Proposed difficulty
- 4 / 5
Proposed verifier
Hidden 400 random cart+coupon combinations: line discounts and totals equal reference to the cent; rejection messages exact; coupon list in RULES.md checked for coverage of 8 hidden rules.
Proposed partial credit
60% totals parity (scaled); 20% messages; 20% rules documented
R02.04.01.02-T15 Rebuild a webmail client with threading, labels and search
Original instruction
Explore `http://ref-mail.local` with two test mailboxes in the locked browser: conversation threading, labels, archive/trash semantics, search operators (`from:`, `has:attachment`, `before:`), and sending between the two accounts. Rebuild it in `/work/mail-clone` with a seeded mailbox import.
- Platforms
web· Any OS (web)- Applications
- Chrome · Node · SQLite
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrollvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Hidden journeys: thread grouping identical on seed data; label/archive semantics; 40 search queries return identical message sets; sent mail appears in recipient inbox; SSIM >= 0.75 at 5 checkpoints.
Proposed partial credit
25% threading; 20% labels/archive; 35% search parity; 20% sending
R02.04.01.02-T16 Reproduce an admin CRUD app with role-based permissions
Original instruction
Explore `http://ref-admin.local` with three test accounts (admin, editor, viewer) in the locked browser: CRUD on products and categories, field-level permissions, audit log and soft delete with restore. Rebuild it in `/work/admin-clone` so each role sees and can do exactly what it can on the reference.
- Platforms
web· Any OS (web)- Applications
- Chrome · Django/Rails/Node
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrollvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 4 / 5
Proposed verifier
Hidden permission matrix (3 roles × 24 actions incl. direct URL access) identical to reference; audit log entries match fields; soft delete/restore semantics; SSIM >= 0.75.
Proposed partial credit
45% permission matrix; 25% CRUD; 15% audit log; 15% soft delete
R02.04.01.02-T17 Replicate a search autocomplete widget's keyboard and timing behavior
Original instruction
In the locked browser, study the header search on `http://ref-store.local`: debounce timing, suggestion grouping (products, categories, recent searches), highlight of matched text, keyboard navigation (arrows, Enter, Esc, Tab) and ARIA combobox semantics. Rebuild the widget in `/work/autocomplete` against the same suggestion API.
- Platforms
web· Any OS (web)- Applications
- Chrome · Vite
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
key· Shortcutsvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden Playwright tests: request count per typed string matches debounce (±1); suggestion order and grouping identical; keyboard behavior identical in 25 sequences; ARIA roles/states match.
Proposed partial credit
25% debounce; 30% suggestions; 30% keyboard; 15% ARIA
R02.04.01.02-T18 Recreate a map-based store locator from browsing it
Original instruction
Explore `http://ref-locator.local` in the locked browser: map with clustered markers, search by postcode, radius filter, open-now filter using store hours and time zones, and a list synced with the map. Rebuild it in `/work/locator` (Leaflet allowed) over the provided store data.
- Platforms
web· Any OS (web)- Applications
- Chrome · Leaflet
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrollvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden 60 searches: listed stores and order (by distance) identical; open-now results equal at 5 injected times; cluster counts at 3 zoom levels equal; clicking list item pans map to marker.
Proposed partial credit
40% search results; 25% open-now; 20% clustering; 15% list-map sync
R02.04.01.02-T19 Rebuild a branching quiz app with scoring rules
Original instruction
Explore `http://ref-quiz.local` in the locked browser: branching question paths depending on answers, timed questions, partial-credit scoring, retry rules and the results page with category breakdown. Rebuild it in `/work/quiz` so any answer path produces the same questions and scores.
- Platforms
web· Any OS (web)- Applications
- Chrome · Node
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
click· Point & clickvis· Visual judgmentcode· Code & commands- Work product
WP06· Code/engineering- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Hidden 200 random answer paths (with timing via clock hook): question sequence and final score breakdown identical; retry rules identical; results page SSIM >= 0.75.
Proposed partial credit
40% branching parity; 35% scoring; 15% timing/retry; 10% visuals
R02.04.01.02-T20 Rebuild a sandbox banking portal with transfers, limits and statements
Original instruction
Explore `http://ref-bank.local` with two test customers in the locked browser: accounts overview, internal and external transfers with daily limits and cut-off times, payee management with confirmation codes (sandbox inbox), scheduled transfers and monthly statements. Rebuild it in `/work/bank-clone` with matching rules.
- Platforms
web· Any OS (web)- Applications
- Chrome · Node/Python · Postgres
- Interfaces
brow· Browserterm· Terminal- GUI need · task setting
HY· Hybrid- GUI need · ordinary access
HY· Hybrid- Interaction primitives
nav· Pan/zoom/scrolltype· Text & formscode· Code & commands- Work product
WP06· Code/engineering- Conditions
long· Long-horizon- Proposed difficulty
- 5 / 5
Proposed verifier
Hidden L1-L3 journeys: limits and cut-off behavior (clock hook), payee confirmation flow via sandbox inbox, scheduled transfer execution, statement totals; ledger invariants; SSIM >= 0.75 at 6 checkpoints.
Proposed partial credit
25% overview/transfers; 25% limits/cut-offs; 20% payees; 30% scheduling and statements
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.