- id: cascader-antd-T01 name: 'Select office location: Zhejiang / Hangzhou / West Lake' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: hierarchical_path_select secondary_template: null browsergym_goal: In the Office location field, select Zhejiang / Hangzhou / West Lake. The task will finish automatically when done. ui_copy: 'Office location: select Zhejiang / Hangzhou / West Lake. The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Component: a single AntD Cascader input labeled “Office location” with placeholder “Please select”. Behavior: clicking the input opens a 3-column dropdown (province → city → landmark). Selection is committed immediately when the final (leaf) item is clicked. Options shown (representative): - Zhejiang → Hangzhou → West Lake - Jiangsu → Nanjing → Zhong Hua Men Initial state: no selection (placeholder visible). Distractors: none; only helper text under the field describing what a cascader is. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: easy tier: L0 axes_ratings: precision_requirement: 2 target_acquisition: 2 density_choice_interference: 1 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 1 disambiguation_load: 1 justification: Small option tree with clear labels and a single instance; the main challenge is completing the full 3-level path selection. success_trigger: human_readable: - The only Cascader on the page has selected path labels exactly equal to [Zhejiang, Hangzhou, West Lake]. - The selected path values exactly equal ['zhejiang','hangzhou','xihu']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Zhejiang - Hangzhou - West Lake path_values: - zhejiang - hangzhou - xihu tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting Zhejiang or Zhejiang / Hangzhou but not selecting the final leaf (West Lake). - Selecting the wrong province/city/landmark path (e.g., Jiangsu / Nanjing / Zhong Hua Men). - Typing text elsewhere or interacting with non-target UI (there is none) without setting the cascader value. expected_interaction_path: - Click the Office location cascader input to open the dropdown. - Click Zhejiang in the first column. - Click Hangzhou in the second column. - Click West Lake in the third column to commit the selection. notes: |- Instrumentation: add data-testid='office-location-cascader' on the Cascader root. Checker: read the committed value array from the Cascader control (value path), not just the visible text. - id: cascader-antd-T02 name: Change preferred warehouse to Europe / Germany / Berlin canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: open_and_select secondary_template: null browsergym_goal: Change the Preferred warehouse selection to Europe / Germany / Berlin. The task will finish automatically when done. ui_copy: 'Preferred warehouse: set to Europe / Germany / Berlin. The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Preferred warehouse”. Initial state: pre-selected value shown in the input as “Americas / USA / New York”. Options: a 3-level geography tree (Region → Country → City) with a few entries: - Americas → USA → New York, Chicago - Americas → Canada → Toronto - Europe → Germany → Berlin Behavior: selecting a new leaf replaces the existing selection immediately. Distractors: a short paragraph explaining that this selection affects shipping estimates. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: easy tier: L0 axes_ratings: precision_requirement: 2 target_acquisition: 2 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 1 disambiguation_load: 1 justification: There is a single field and a small tree; the only extra step is replacing an existing default value. success_trigger: human_readable: - The Preferred warehouse Cascader value path equals Europe → Germany → Berlin. - The committed value array equals ['europe','de','berlin']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Europe - Germany - Berlin path_values: - europe - de - berlin tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Leaving the default selection unchanged (Americas / USA / New York). - Selecting Europe / Germany but not selecting Berlin (partial path). - Selecting the wrong city (e.g., Toronto or Chicago). expected_interaction_path: - Click the Preferred warehouse cascader input to open the dropdown. - Click Europe → Germany → Berlin to commit the new selection. notes: |- Instrumentation: data-testid='preferred-warehouse-cascader'. Checker should assert the underlying value path, not just the rendered string. - id: cascader-antd-T03 name: Clear the Product category selection canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: clear_reset secondary_template: null browsergym_goal: Clear the selection in the Product category field so it is blank. The task will finish automatically when done. ui_copy: 'Product category: clear the selection (leave it blank). The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Product category” with allowClear enabled. Initial state: the cascader currently shows “Electronics / Computers / Laptops”. UI details: when the input is hovered, a small clear (×) icon appears on the right side of the input. Behavior: clicking the clear icon removes the selection and returns the input to the placeholder state. Distractors: none. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: easy tier: L0 axes_ratings: precision_requirement: 1 target_acquisition: 2 density_choice_interference: 1 depth_layering: 1 feedback_dynamics: 1 semantic_observability: 1 disambiguation_load: 1 justification: Single-field reset using the built-in clear affordance; no navigation through options is needed. success_trigger: human_readable: - Product category Cascader has no selected value (empty / null value path). - The placeholder is visible and no selection tags/text are present. canonical_predicate: predicate_type: equals target_state: value: null tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Opening the dropdown but leaving the existing selection intact. - Selecting a different category instead of clearing. - Clearing the wrong field (there is only one cascader on the page). expected_interaction_path: - Hover the Product category input if needed to reveal the clear icon. - Click the clear (×) icon to reset the cascader to blank. notes: |- Instrumentation: data-testid='product-category-cascader'. Checker: treat both null and empty-array value as 'no selection' depending on the adapter. - id: cascader-antd-T04 name: 'Inline panel: select Engineering / Frontend / Web Platform' canonical_type: cascader implementation_source: antd implementation_variant: panel implementation_component: 'AntD: Cascader.Panel' task_template: hierarchical_path_select secondary_template: null browsergym_goal: In the Team routing panel, select Engineering / Frontend / Web Platform. The task will finish automatically when done. ui_copy: 'Team routing: select Engineering / Frontend / Web Platform. The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Component: an inline AntD Cascader.Panel labeled “Team routing” (always visible; no trigger input). Panel layout: three adjacent columns are shown at once (Department → Team → Subteam). Options (representative): - Engineering → Frontend → Web Platform, Design Systems - Engineering → Backend → Payments, Data - Operations → IT → Helpdesk Initial state: nothing selected in the panel. Distractors: none; the panel is the only interactive component in the card. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: easy tier: L0 axes_ratings: precision_requirement: 2 target_acquisition: 2 density_choice_interference: 2 depth_layering: 1 feedback_dynamics: 1 semantic_observability: 1 disambiguation_load: 1 justification: Inline panel avoids overlay management; the agent only needs to click through a short 3-level hierarchy. success_trigger: human_readable: - The Cascader.Panel selected path labels equal [Engineering, Frontend, Web Platform]. - The selected path values equal ['eng','fe','web-platform']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Engineering - Frontend - Web Platform path_values: - eng - fe - web-platform tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting a different subteam (e.g., Engineering / Frontend / Design Systems). - Stopping at a non-leaf level (e.g., selecting Engineering only). - Interacting with the page without changing the panel selection. expected_interaction_path: - In the first column, click Engineering. - In the second column, click Frontend. - In the third column, click Web Platform. notes: |- Instrumentation: data-testid='team-routing-cascader-panel'. Checker: read the committed selection from the panel state; selection occurs on leaf click. - id: cascader-antd-T05 name: 'Dark theme: pick Europe / France / Paris' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: hierarchical_path_select secondary_template: null browsergym_goal: In the Travel destination field, select Europe / France / Paris. The task will finish automatically when done. ui_copy: 'Travel destination: choose Europe / France / Paris. The task will finish automatically when done.' setup_description: |- Theme: dark. Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Travel destination”. Options: Region → Country → City with a small set: - Europe → France → Paris - Europe → Spain → Madrid - Asia → Japan → Tokyo Initial state: empty (placeholder visible). Distractors: none. scene_context: theme: dark spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: easy tier: L0 axes_ratings: precision_requirement: 2 target_acquisition: 2 density_choice_interference: 1 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 2 disambiguation_load: 1 justification: Same simple 3-level selection as the baseline, but in dark theme which can slightly reduce visual salience. success_trigger: human_readable: - Travel destination cascader selected path labels equal [Europe, France, Paris]. - Selected path values equal ['europe','fr','paris']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Europe - France - Paris path_values: - europe - fr - paris tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting the wrong city (Madrid or Tokyo). - Stopping before the leaf city is selected. - Leaving the cascader blank. expected_interaction_path: - Click the Travel destination input to open the dropdown. - Click Europe → France → Paris. notes: 'Instrumentation: data-testid=''travel-destination-cascader''.' - id: cascader-antd-T06 name: 'Two instances: set Primary office to North America / Canada / Vancouver' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: hierarchical_path_select secondary_template: null browsergym_goal: Set the Primary office field to North America / Canada / Vancouver. The task will finish automatically when done. ui_copy: 'Primary office: North America / Canada / Vancouver. The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Components: two AntD Cascader inputs using the same options tree. 1) “Primary office” (target) 2) “Backup office” (distractor) Options: Continent → Country → City: - North America → USA → Seattle, Austin - North America → Canada → Vancouver, Montreal - Europe → UK → London Initial state: Primary office is blank; Backup office is preselected to “Europe / UK / London”. Distractors: the two fields are visually similar; only the label distinguishes them. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 2 guidance: text clutter: none difficulty: difficulty_bucket: easy tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 2 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 1 disambiguation_load: 3 justification: Selection is straightforward, but there are two identical cascaders so the agent must target the correct labeled instance. success_trigger: human_readable: - The cascader labeled “Primary office” has selected path labels [North America, Canada, Vancouver]. - The cascader labeled “Backup office” may remain unchanged. - Primary office selected path values equal ['na','ca','vancouver']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - North America - Canada - Vancouver path_values: - na - ca - vancouver tolerance: null require_confirm: false confirm_control: null require_correct_instance: true target_instance_label_or_id: primary-office terminal_condition: task ends when predicate holds negative_cases: - Setting the Backup office instead of Primary office. - Selecting North America / Canada but choosing Montreal instead of Vancouver. - Leaving Primary office blank. expected_interaction_path: - Locate the cascader input labeled Primary office. - Open it and select North America → Canada → Vancouver. notes: |- Instrumentation: data-testid='cascader-primary-office' and data-testid='cascader-backup-office'. Checker: ensure correct-instance by binding to label or testid, not by DOM order. - id: cascader-antd-T07 name: Match the Target breadcrumb for Selected service canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (show full path display)' task_template: match_reference secondary_template: null browsergym_goal: Set Selected service to match the breadcrumb shown in the Target path box. The task will finish automatically when done. ui_copy: Match Selected service to the Target path shown. The task will finish automatically when done. setup_description: |- Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Selected service”. Guidance element: a non-interactive “Target path” preview box above the cascader showing a styled breadcrumb with the desired path (e.g., Services › Consulting › Security Review). Options tree: Service line → Offering → Package: - Services → Consulting → Security Review (target) - Services → Consulting → Architecture Review - Services → Support → Priority Support Initial state: Selected service is blank. Distractors: an extra note that explains the breadcrumb is the target reference. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: visual clutter: none difficulty: difficulty_bucket: easy tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 2 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 3 disambiguation_load: 1 justification: Requires mapping a visual breadcrumb reference to the cascader selection, but the option set is small and there is only one field. success_trigger: human_readable: - Selected service cascader selected path equals the Target path breadcrumb shown on the page. - The selected path labels equal [Services, Consulting, Security Review]. - The selected path values equal ['services','consulting','security-review']. canonical_predicate: predicate_type: matches_reference target_state: reference_element_id: target-path-breadcrumb path_labels: - Services - Consulting - Security Review path_values: - services - consulting - security-review tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting a different consulting package (e.g., Architecture Review). - Selecting only part of the path (stopping before the leaf). - Leaving the cascader blank while the Target breadcrumb remains visible. expected_interaction_path: - Read the Target path breadcrumb in the reference box. - Open Selected service cascader and click Services → Consulting → Security Review. notes: |- Instrumentation: data-testid='selected-service-cascader' and data-testid='target-path-breadcrumb'. Checker: for matches_reference, compare cascader value path to the reference breadcrumb’s canonical path. - id: cascader-antd-T08 name: Open the Support topic cascader dropdown (leave it open) canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: open_overlay secondary_template: null browsergym_goal: Open the Support topic cascader so the dropdown options are visible, and leave it open. The task will finish automatically when done. ui_copy: 'Support topic: open the dropdown (do not select anything). The task will finish automatically when done.' setup_description: |- Layout: isolated card anchored near the top-right of the viewport. Component: one AntD Cascader input labeled “Support topic”, initially blank. Options: Topic → Subtopic → Detail (examples): - Account → Billing → Refunds - Account → Login → Password reset - Product → Mobile → Crashes Behavior: clicking the input opens a floating multi-column dropdown. Clicking outside closes it. Goal of this task: only the open/closed state matters; no selection should be made. scene_context: theme: light spacing: comfortable layout: isolated_card placement: top_right scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: easy tier: L0 axes_ratings: precision_requirement: 1 target_acquisition: 2 density_choice_interference: 1 depth_layering: 1 feedback_dynamics: 1 semantic_observability: 2 disambiguation_load: 1 justification: Tests basic ability to open an overlay without making any selection; otherwise a simple single-instance setup. success_trigger: human_readable: - The Support topic cascader dropdown is open/visible (popup rendered). - The cascader still has no committed selection (value is null/empty). canonical_predicate: predicate_type: equals target_state: open: true value: null tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting any option in the cascader (even if the dropdown is open). - Opening the dropdown and then closing it before the checker runs (open=false). - Interacting with other UI (none) while leaving the cascader closed. expected_interaction_path: - Click the Support topic cascader input to open the dropdown. - Do not click any option; keep the dropdown open. notes: |- Instrumentation: data-testid='support-topic-cascader'. Checker: determine open state via presence of the popup element bound to this cascader instance. - id: cascader-antd-T09 name: 'Compact spacing: select America / New York' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (size=middle)' task_template: hierarchical_path_select secondary_template: null browsergym_goal: In the Timezone field, select America / New York. The task will finish automatically when done. ui_copy: 'Timezone: America / New York. The task will finish automatically when done.' setup_description: |- Spacing: compact mode (reduced padding between labels and controls). Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Timezone”. Options: Region → City (2 levels): - America → New York, Los Angeles - Europe → London Initial state: blank. Distractors: none. scene_context: theme: light spacing: compact layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: easy tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 3 density_choice_interference: 1 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 1 disambiguation_load: 1 justification: Compact spacing reduces click target size slightly, but the hierarchy is shallow and the options are few. success_trigger: human_readable: - Timezone cascader selected path labels equal [America, New York]. - Selected path values equal ['america','new-york']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - America - New York path_values: - america - new-york tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting America / Los Angeles. - Selecting Europe / London. - Leaving the field blank. expected_interaction_path: - Open the Timezone cascader. - Click America, then click New York. notes: 'Instrumentation: data-testid=''timezone-cascader''.' - id: cascader-antd-T10 name: 'Search and select: South Campus / Art Studio (top-left placement)' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (showSearch=true)' task_template: search_and_select secondary_template: null browsergym_goal: Use the search in the Campus building selector to choose South Campus / Art Studio. The task will finish automatically when done. ui_copy: 'Campus building: search and pick South Campus / Art Studio. The task will finish automatically when done.' setup_description: |- Layout: isolated card anchored near the top-left of the viewport. Component: one AntD Cascader labeled “Campus building” with showSearch enabled. UI behavior: opening the cascader reveals a search input; typing filters the available paths. Options: Campus → Building: - North Campus → Science Hall, Library - South Campus → Gym, Art Studio (target) Initial state: blank. Distractors: the dropdown still supports manual column navigation, but the intent is to use search. scene_context: theme: light spacing: comfortable layout: isolated_card placement: top_left scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: easy tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 2 density_choice_interference: 2 depth_layering: 3 feedback_dynamics: 2 semantic_observability: 2 disambiguation_load: 1 justification: Adds a search box interaction step, but the filtered results are few and the target is unique. success_trigger: human_readable: - Campus building cascader selected path labels equal [South Campus, Art Studio]. - Selected path values equal ['south','art-studio']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - South Campus - Art Studio path_values: - south - art-studio tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Typing in the search box but not selecting a result (no committed value). - Selecting South Campus / Gym. - Selecting a North Campus building. expected_interaction_path: - Open the Campus building cascader dropdown. - Type a query like “Art” in the search input. - Click the result path for South Campus / Art Studio. notes: |- Instrumentation: data-testid='campus-building-cascader'. Checker: verify committed value path; do not accept merely having search text present. - id: cascader-antd-T11 name: 'Two instances: set Billing region to United States / California / San Jose' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: hierarchical_path_select secondary_template: null browsergym_goal: Set the Billing region field to United States / California / San Jose. The task will finish automatically when done. ui_copy: 'Billing region: United States / California / San Jose. The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Components: two similar AntD Cascader inputs stacked vertically: 1) “Shipping region” (distractor) 2) “Billing region” (target) Options: Country → State/Territory → City, with several similar city names: - United States → California → San Jose (target), San Diego - United States → Texas → San Antonio - Puerto Rico → (territory) → San Juan Initial state: Shipping region is set to “Puerto Rico / San Juan”; Billing region is blank. Distractors: multiple cities start with “San …”, so reading the full path is required. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 2 guidance: text clutter: none difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 3 target_acquisition: 2 density_choice_interference: 3 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 2 disambiguation_load: 3 justification: Requires disambiguating between similar city names and choosing the correct labeled field among two instances. success_trigger: human_readable: - The cascader labeled “Billing region” has selected path labels [United States, California, San Jose]. - Billing region selected path values equal ['us','ca','san-jose']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - United States - California - San Jose path_values: - us - ca - san-jose tolerance: null require_confirm: false confirm_control: null require_correct_instance: true target_instance_label_or_id: billing-region terminal_condition: task ends when predicate holds negative_cases: - Setting Shipping region instead of Billing region. - Selecting a different “San …” city (San Diego, San Antonio, or San Juan). - Stopping at United States / California without selecting the city leaf. expected_interaction_path: - Locate the Billing region cascader (not Shipping region). - Open it and select United States → California → San Jose. notes: |- Instrumentation: data-testid='shipping-region-cascader' and data-testid='billing-region-cascader'. Checker: enforce correct instance by testid/label binding. - id: cascader-antd-T12 name: 'Mixed guidance: match Target breadcrumb for Cost center' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: match_reference secondary_template: null browsergym_goal: Set Cost center to match the Target breadcrumb shown on the page (it starts with “Finance”). The task will finish automatically when done. ui_copy: 'Cost center: match the Target breadcrumb (starts with Finance). The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Cost center”. Guidance elements: - A “Target” breadcrumb preview directly above the field showing the full desired path as styled crumb chips. - A small text hint under the preview: “Target starts with Finance”. Options tree: Division → Department → Team: - Finance → Accounting → Accounts Payable (target) - Finance → Accounting → Accounts Receivable - Finance → FP&A → Forecasting - Operations → Procurement → Vendor Management Initial state: blank. Distractors: Finance has multiple similar teams under Accounting. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: mixed clutter: none difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 2 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 3 disambiguation_load: 1 justification: The target is given via a reference breadcrumb plus a hint; difficulty comes from comparing the reference and selecting among similar Finance teams. success_trigger: human_readable: - Cost center cascader selected path equals the Target breadcrumb reference. - Selected path labels equal [Finance, Accounting, Accounts Payable]. - Selected path values equal ['finance','acct','ap']. canonical_predicate: predicate_type: matches_reference target_state: reference_element_id: target-cost-center-breadcrumb path_labels: - Finance - Accounting - Accounts Payable path_values: - finance - acct - ap tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting Accounts Receivable instead of Accounts Payable. - Selecting a Finance path that does not match the Target breadcrumb. - Leaving the field blank. expected_interaction_path: - Read the Target breadcrumb (and optional hint). - Open Cost center cascader and select Finance → Accounting → Accounts Payable. notes: |- Instrumentation: data-testid='cost-center-cascader' and data-testid='target-cost-center-breadcrumb'. Checker: compare cascader value path to the reference breadcrumb’s canonical path. - id: cascader-antd-T13 name: 'Form section: set Department to Engineering / Platform / Infrastructure' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: hierarchical_path_select secondary_template: null browsergym_goal: In the Project intake form, set Department to Engineering / Platform / Infrastructure. The task will finish automatically when done. ui_copy: 'Project intake: Department = Engineering / Platform / Infrastructure. The task will finish automatically when done.' setup_description: |- Layout: form section (a realistic intake form) centered on the page. Clutter: low — there are several non-required fields. Components on the form: - Text input: “Project name” (pre-filled) - Select: “Priority” (default set) - Cascader (target): “Department” Department options: Division → Group → Team: - Engineering → Platform → Infrastructure (target), Developer Experience - Engineering → Product → Web, Mobile - Sales → Enterprise → North America Initial state: Department is blank. Behavior: Department cascader commits on leaf click; no Save/Submit is required for this task. scene_context: theme: light spacing: comfortable layout: form_section placement: center scale: default instances: 1 guidance: text clutter: low difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 2 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 1 disambiguation_load: 2 justification: Additional form fields add mild distraction, but the cascader options are modest and clearly labeled. success_trigger: human_readable: - Department cascader selected path labels equal [Engineering, Platform, Infrastructure]. - Selected path values equal ['eng','platform','infra']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Engineering - Platform - Infrastructure path_values: - eng - platform - infra tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting Engineering / Platform / Developer Experience. - Selecting Engineering / Product / Web or another branch. - Changing other form fields while leaving Department unset. expected_interaction_path: - Locate the Department cascader in the form. - Open it and select Engineering → Platform → Infrastructure. notes: 'Instrumentation: data-testid=''department-cascader''.' - id: cascader-antd-T14 name: 'Dashboard filter: set Product taxonomy to Home / Kitchen / Coffee Makers' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: open_and_select secondary_template: null browsergym_goal: On the analytics dashboard, set the Product taxonomy filter to Home / Kitchen / Coffee Makers. The task will finish automatically when done. ui_copy: 'Dashboard filter: Product taxonomy = Home / Kitchen / Coffee Makers. The task will finish automatically when done.' setup_description: |- Layout: dashboard with a top filter bar. Clutter: medium — multiple filters and charts are present. Target component: an AntD Cascader in the filter bar labeled “Product taxonomy”. Other UI (distractors): a date range picker, a keyword search box, a region select, and several summary charts. Taxonomy options: Category → Subcategory → Product type: - Home → Kitchen → Coffee Makers (target), Cookware, Cutlery - Home → Lighting → Lamps - Electronics → Audio → Headphones Initial state: Product taxonomy is blank (interpreted as “All”). scene_context: theme: light spacing: comfortable layout: dashboard placement: center scale: default instances: 1 guidance: text clutter: medium difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 2 density_choice_interference: 3 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 1 disambiguation_load: 3 justification: The dashboard context adds competing interactive elements and a moderately dense taxonomy, but the target leaf is still easy to identify. success_trigger: human_readable: - Product taxonomy cascader selected path labels equal [Home, Kitchen, Coffee Makers]. - Selected path values equal ['home','kitchen','coffee-makers']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Home - Kitchen - Coffee Makers path_values: - home - kitchen - coffee-makers tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting Home / Kitchen / Cookware (neighbor leaf). - Interacting with other filters without setting Product taxonomy. - Leaving the filter at blank/all. expected_interaction_path: - Find the Product taxonomy cascader in the dashboard filter bar. - Open it and choose Home → Kitchen → Coffee Makers. notes: |- Instrumentation: data-testid='product-taxonomy-cascader'. Robustness: ensure charts are non-blocking and do not overlap the dropdown. - id: cascader-antd-T15 name: 'Cancel out: open Delivery zone and close without changing' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: confirm_cancel secondary_template: null browsergym_goal: Open the Delivery zone selector, then close it without changing the current selection (keep EMEA / UK / London). The task will finish automatically when done. ui_copy: 'Delivery zone: open then cancel (keep EMEA / UK / London). The task will finish automatically when done.' setup_description: |- Layout: isolated card anchored near the bottom-left of the viewport. Component: one AntD Cascader labeled “Delivery zone”. Initial state: already selected to “EMEA / UK / London”. Options: Region → Country → City: - EMEA → UK → London (current) - EMEA → UAE → Dubai - Americas → USA → New York Behavior: opening the dropdown shows columns; clicking outside (or pressing Escape) closes the dropdown. Selection only changes if a new leaf is clicked. Goal: ensure the dropdown ends closed and the committed selection remains unchanged. scene_context: theme: light spacing: comfortable layout: isolated_card placement: bottom_left scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 2 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 3 disambiguation_load: 1 justification: Requires managing overlay open/close while preserving the existing selection; success depends on both value and popup visibility. success_trigger: human_readable: - Delivery zone cascader dropdown is closed (open=false). - The committed selection remains exactly EMEA / UK / London with values ['emea','uk','london']. canonical_predicate: predicate_type: equals target_state: open: false path_labels: - EMEA - UK - London path_values: - emea - uk - london tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Changing the selection to a different city (e.g., Dubai). - Leaving the dropdown open at the end. - Clearing the value or making the field blank. expected_interaction_path: - Click the Delivery zone input to open the dropdown. - Close the dropdown by clicking outside or pressing Escape, without selecting a new option. notes: |- Instrumentation: data-testid='delivery-zone-cascader'. Checker: must verify both selection value and open state to distinguish 'cancel' from 'do nothing'. - id: cascader-antd-T16 name: 'Small scale: select Hardware / Networking / Routers' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (size=small)' task_template: hierarchical_path_select secondary_template: null browsergym_goal: In the Asset category field, select Hardware / Networking / Routers. The task will finish automatically when done. ui_copy: 'Asset category: Hardware / Networking / Routers. The task will finish automatically when done.' setup_description: |- Scale: small (the cascader input uses AntD size='small'). Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Asset category”. Options: Category → Subcategory → Type: - Hardware → Networking → Routers (target), Switches - Hardware → Computing → Laptops - Software → Security → Antivirus Initial state: blank. Distractors: none, but the smaller control reduces target size. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: small instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 3 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 1 disambiguation_load: 1 justification: Small scale increases pointing difficulty, while the hierarchy remains moderate and unambiguous. success_trigger: human_readable: - Asset category selected path labels equal [Hardware, Networking, Routers]. - Selected path values equal ['hw','net','routers']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Hardware - Networking - Routers path_values: - hw - net - routers tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting Hardware / Networking / Switches. - Selecting a different top-level category (Software). - Leaving the field blank. expected_interaction_path: - Open the Asset category cascader. - Select Hardware → Networking → Routers. notes: 'Instrumentation: data-testid=''asset-category-cascader''.' - id: cascader-antd-T17 name: 'Search and select among similar results: United States / California / San Jose' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (showSearch=true)' task_template: search_and_select secondary_template: null browsergym_goal: Use search in the City selector to choose United States / California / San Jose. The task will finish automatically when done. ui_copy: 'City selector: search and pick United States / California / San Jose. The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Component: one AntD Cascader labeled “City” with showSearch enabled. Options: Country → State/Territory → City. The dataset is larger and includes many cities starting with 'San': - United States → California → San Jose (target), San Diego - United States → Texas → San Antonio - Puerto Rico → (territory) → San Juan - United States → Florida → Sarasota Search behavior: typing filters by path; results show full paths. Initial state: blank. Distractors: multiple search hits can appear for short queries like “San”. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: mid tier: L2 axes_ratings: precision_requirement: 3 target_acquisition: 2 density_choice_interference: 4 depth_layering: 3 feedback_dynamics: 2 semantic_observability: 2 disambiguation_load: 1 justification: Search narrows the list but introduces many confusable results; the agent must pick the correct full path from the filtered suggestions. success_trigger: human_readable: - City cascader selected path labels equal [United States, California, San Jose]. - Selected path values equal ['us','ca','san-jose']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - United States - California - San Jose path_values: - us - ca - san-jose tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting a different 'San…' city (San Diego, San Antonio, or San Juan). - Typing search text but not selecting any suggestion (no committed value). - Selecting via columns but ending on the wrong leaf. expected_interaction_path: - Open the City cascader dropdown. - Type a query like “San Jo” to reduce ambiguity. - Click the suggestion for United States / California / San Jose. notes: |- Instrumentation: data-testid='city-cascader'. Checker: verify committed value path; do not accept search text alone. - id: cascader-antd-T18 name: 'Change-on-select: choose parent path Engineering / Platform' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (changeOnSelect=true)' task_template: hierarchical_path_select secondary_template: null browsergym_goal: In the Reporting scope field, select Engineering / Platform (you do not need to pick a leaf team). The task will finish automatically when done. ui_copy: 'Reporting scope: Engineering / Platform. The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Reporting scope” configured with changeOnSelect=true. Behavior: the value updates on each selection level; selecting a parent is allowed and commits immediately. Options: Division → Group → Team: - Engineering → Platform → Infrastructure, Developer Experience - Engineering → Product → Web, Mobile - Operations → IT → Helpdesk Initial state: blank. Key nuance: success requires stopping at the parent path “Engineering / Platform” (2 levels), not choosing a leaf team. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 3 target_acquisition: 2 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 2 disambiguation_load: 1 justification: The component commits on intermediate selections, so the agent must intentionally stop at the correct parent depth instead of continuing to a leaf. success_trigger: human_readable: - Reporting scope cascader selected path labels equal [Engineering, Platform] (exactly two levels). - Selected path values equal ['eng','platform']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Engineering - Platform path_values: - eng - platform tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting Engineering / Platform / Infrastructure (too deep). - Selecting Engineering / Product (wrong group). - Leaving the field blank. expected_interaction_path: - Open the Reporting scope cascader. - Click Engineering in the first column. - Click Platform in the second column and stop (value commits at that level). notes: 'Instrumentation: data-testid=''reporting-scope-cascader''.' - id: cascader-antd-T19 name: 'Two instances: clear Secondary tag' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (allowClear=true)' task_template: clear_reset secondary_template: null browsergym_goal: Clear the selection in the Secondary tag field so it becomes blank. The task will finish automatically when done. ui_copy: 'Secondary tag: clear selection (blank). The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Components: two AntD Cascader inputs labeled “Primary tag” and “Secondary tag”. Both allow clearing. Initial state: - Primary tag is set to “Customer / Onboarding / High touch”. - Secondary tag is set to “Region / EMEA / UK”. (target to clear) UI behavior: each input shows a small clear icon (×) on hover. Distractors: the two fields are visually similar; only the label distinguishes which one to clear. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 2 guidance: text clutter: none difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 3 density_choice_interference: 1 depth_layering: 1 feedback_dynamics: 1 semantic_observability: 1 disambiguation_load: 3 justification: Clearing is simple but requires precise targeting of the clear icon on the correct instance. success_trigger: human_readable: - The cascader labeled “Secondary tag” has no selected value (null/empty). - The cascader labeled “Primary tag” may remain unchanged. canonical_predicate: predicate_type: equals target_state: value: null tolerance: null require_confirm: false confirm_control: null require_correct_instance: true target_instance_label_or_id: secondary-tag terminal_condition: task ends when predicate holds negative_cases: - Clearing the Primary tag instead of Secondary tag. - Leaving Secondary tag selected (Region / EMEA / UK). - Selecting a different tag rather than clearing. expected_interaction_path: - Locate the Secondary tag cascader input. - Hover if needed and click its clear (×) icon. notes: |- Instrumentation: data-testid='primary-tag-cascader' and data-testid='secondary-tag-cascader'. Checker: enforce correct-instance; accept both null and empty-array representations for 'cleared'. - id: cascader-antd-T20 name: 'Settings panel: choose Access scope in inline panel (4 levels)' canonical_type: cascader implementation_source: antd implementation_variant: panel implementation_component: 'AntD: Cascader.Panel' task_template: hierarchical_path_select secondary_template: null browsergym_goal: In the Access scope panel, select Organization / Europe / Dublin / Zone 2. The task will finish automatically when done. ui_copy: 'Access scope: Organization / Europe / Dublin / Zone 2. The task will finish automatically when done.' setup_description: |- Layout: settings panel with multiple sections. Clutter: low — there are a few toggles and helper texts, but only one cascader panel. Target component: an inline AntD Cascader.Panel labeled “Access scope”. Panel structure: four columns visible as you drill down (Scope → Region → Site → Zone). Options (representative): - Organization → North America → Seattle → Zone 1, Zone 2 - Organization → Europe → Dublin → Zone 1, Zone 2 (target) - Team only → (no further levels) Initial state: no selection. Distractors: nearby switches like “Enable SSO” and “Read-only mode” are present but irrelevant. scene_context: theme: light spacing: comfortable layout: settings_panel placement: center scale: default instances: 1 guidance: text clutter: low difficulty: difficulty_bucket: mid tier: L2 axes_ratings: precision_requirement: 3 target_acquisition: 2 density_choice_interference: 3 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 2 disambiguation_load: 2 justification: Deeper 4-level hierarchy in an inline panel requires multiple precise selections, with mild distraction from surrounding settings UI. success_trigger: human_readable: - Access scope panel selected path labels equal [Organization, Europe, Dublin, Zone 2]. - Selected path values equal ['org','eu','dublin','zone-2']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Organization - Europe - Dublin - Zone 2 path_values: - org - eu - dublin - zone-2 tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting Zone 1 instead of Zone 2 under Dublin. - Selecting a different site (e.g., Seattle). - Selecting 'Team only' (wrong top-level scope). expected_interaction_path: - In the Access scope panel, click Organization. - Click Europe, then Dublin, then Zone 2. notes: |- Instrumentation: data-testid='access-scope-cascader-panel'. Checker: path length must be 4 and match values exactly. - id: cascader-antd-T21 name: 'Lazy-loaded options (dark): select Hardware / Printers / Setup' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (loadData)' task_template: open_and_select secondary_template: null browsergym_goal: In the Knowledge base category field, select Hardware / Printers / Setup. The task will finish automatically when done. ui_copy: 'Knowledge base category: Hardware / Printers / Setup. The task will finish automatically when done.' setup_description: |- Theme: dark. Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Knowledge base category”, configured with loadData for lazy loading. Behavior: only the first column is populated initially. When a node is selected, the next column shows a loading spinner briefly and then renders fetched children. Option structure (as it appears after loading): - Hardware → Printers → Setup (target), Troubleshooting - Hardware → Monitors → Calibration - Software → VPN → Installation Constraints: showSearch is disabled (not available with loadData). Initial state: blank. Feedback: a subtle loading indicator appears in the dropdown column while children load. scene_context: theme: dark spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: hard tier: L2 axes_ratings: precision_requirement: 3 target_acquisition: 2 density_choice_interference: 2 depth_layering: 3 feedback_dynamics: 4 semantic_observability: 2 disambiguation_load: 1 justification: Lazy loading introduces timing/feedback challenges; the agent must wait for each column to populate before making the next selection. success_trigger: human_readable: - Knowledge base category selected path labels equal [Hardware, Printers, Setup]. - Selected path values equal ['hw','printers','setup']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Hardware - Printers - Setup path_values: - hw - printers - setup tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Clicking too quickly so the second/third column has not loaded, resulting in no committed leaf selection. - Selecting Hardware / Printers / Troubleshooting instead of Setup. - Selecting a Software category path. expected_interaction_path: - Open the Knowledge base category cascader dropdown. - Click Hardware and wait for the next column to load. - Click Printers and wait for the next column to load. - Click Setup to commit. notes: |- Instrumentation: data-testid='kb-category-cascader'. Checker: verify final committed value path; do not treat intermediate selections as success. - id: cascader-antd-T22 name: 'Hover-to-expand in compact spacing: Tools / Power Tools / Sanders' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (expandTrigger=''hover'')' task_template: hierarchical_path_select secondary_template: null browsergym_goal: In the Inventory category field, select Tools / Power Tools / Sanders. The task will finish automatically when done. ui_copy: 'Inventory category: Tools / Power Tools / Sanders. The task will finish automatically when done.' setup_description: |- Spacing: compact. Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Inventory category”, configured with expandTrigger='hover'. Behavior: submenus open when the pointer hovers over an item in a column (not when it is clicked). The final leaf is selected by clicking. Options: Department → Category → Item: - Tools → Power Tools → Sanders (target), Drills - Tools → Hand Tools → Hammers - Supplies → Safety → Gloves Initial state: blank. Distractors: in compact mode, list items are shorter and closer together. scene_context: theme: light spacing: compact layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: hard tier: L3 axes_ratings: precision_requirement: 3 target_acquisition: 4 density_choice_interference: 3 depth_layering: 3 feedback_dynamics: 2 semantic_observability: 2 disambiguation_load: 1 justification: Hover-based expansion is harder for agents than click expansion, and compact spacing reduces target size while navigating multi-column menus. success_trigger: human_readable: - Inventory category selected path labels equal [Tools, Power Tools, Sanders]. - Selected path values equal ['tools','power-tools','sanders']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Tools - Power Tools - Sanders path_values: - tools - power-tools - sanders tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Clicking items expecting them to expand (without hovering) and failing to reach the correct submenu. - Selecting Tools / Power Tools / Drills instead of Sanders. - Selecting a different top-level branch (Supplies). expected_interaction_path: - Open the Inventory category cascader dropdown. - Hover over Tools to open its submenu column. - Hover over Power Tools to open the third column. - Click Sanders to commit the leaf selection. notes: |- Instrumentation: data-testid='inventory-category-cascader'. Checker: verify leaf value; hover interactions are not directly checked—only final path matters. - id: cascader-antd-T23 name: 'Multiple mode: select exactly two cities (Paris and Osaka)' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (multiple=true, showCheckedStrategy=SHOW_CHILD)' task_template: select_many secondary_template: null browsergym_goal: 'In the Cities included field, select exactly these two paths: Europe / France / Paris and Asia / Japan / Osaka. The task will finish automatically when done.' ui_copy: 'Cities included: select Paris and Osaka (exactly two). The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Cities included” in multiple=true mode (checkboxes shown). Display: showCheckedStrategy is set to SHOW_CHILD so selected leaf nodes appear as individual tags in the input. Options: Continent → Country → City: - Europe → France → Paris (target), Lyon - Asia → Japan → Tokyo, Osaka (target) - Americas → USA → New York, Chicago Initial state: no selections. Behavior: each leaf is selected by checking its checkbox (or clicking the item). Multiple selections can be made in one session. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: hard tier: L2 axes_ratings: precision_requirement: 4 target_acquisition: 3 density_choice_interference: 3 depth_layering: 3 feedback_dynamics: 2 semantic_observability: 3 disambiguation_load: 1 justification: Multiple selection requires exact set matching and careful checkbox interactions across different branches of the tree. success_trigger: human_readable: - 'The Cities included cascader has selected exactly the set of two leaf paths: Europe/France/Paris and Asia/Japan/Osaka.' - No additional paths are selected. canonical_predicate: predicate_type: set_equals target_state: selected_paths: - path_labels: - Europe - France - Paris path_values: - europe - fr - paris - path_labels: - Asia - Japan - Osaka path_values: - asia - jp - osaka order_insensitive: true tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting only one of the two required cities. - Selecting Paris and Tokyo (wrong Japan city) or adding any extra city. - Selecting a parent node instead of the leaf city (if the UI allows it). expected_interaction_path: - Open the Cities included cascader dropdown. - Navigate to Europe → France and check Paris. - Navigate to Asia → Japan and check Osaka. - Ensure only two tags/selections are present. notes: |- Instrumentation: data-testid='cities-included-cascader'. Checker: compare sets of value paths (order-insensitive). - id: cascader-antd-T24 name: 'Multiple mode (visual): match the two Target selections chips' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (multiple=true)' task_template: select_many secondary_template: null browsergym_goal: In Teams included, select the same set of paths shown in the Target selections box. The task will finish automatically when done. ui_copy: 'Teams included: match the Target selections shown. The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Teams included” in multiple=true mode. Guidance element: a “Target selections” box above the cascader shows exactly two styled chips, each chip displaying a breadcrumb path (visual reference). Options: Department → Team → Squad: - Engineering → Backend → Payments (one target) - Engineering → Frontend → Web Platform - Marketing → Events → Conferences (one target) - Marketing → Content → Blog Initial state: no selections. Distractors: other non-target squads exist under the same departments. Important: the prompt does not list the paths; the agent must copy them from the on-page Target selections reference. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: visual clutter: none difficulty: difficulty_bucket: hard tier: L3 axes_ratings: precision_requirement: 4 target_acquisition: 3 density_choice_interference: 3 depth_layering: 3 feedback_dynamics: 2 semantic_observability: 4 disambiguation_load: 1 justification: Requires interpreting a visual reference (target chips) and matching an exact multi-selection set within a hierarchy. success_trigger: human_readable: - Teams included cascader selected set equals the two Target selections chips shown on the page. - Selected paths are exactly [Engineering/Backend/Payments] and [Marketing/Events/Conferences], order-insensitive. canonical_predicate: predicate_type: matches_reference target_state: reference_element_id: target-selections-chips selected_paths: - path_labels: - Engineering - Backend - Payments path_values: - eng - be - payments - path_labels: - Marketing - Events - Conferences path_values: - mkt - events - conferences order_insensitive: true tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting only one of the two target paths. - Selecting the correct two plus any additional path. - Selecting a sibling like Engineering / Frontend / Web Platform instead of the backend payments squad. expected_interaction_path: - Read the two breadcrumb chips in the Target selections box. - Open Teams included cascader and select/check each matching leaf path. - Verify the selected tags match the two reference chips. notes: |- Instrumentation: data-testid='teams-included-cascader' and data-testid='target-selections-chips'. Checker: compare selected set to reference chips’ canonical paths. - id: cascader-antd-T25 name: 'Ambiguous leaf labels with leaf-only displayRender: choose Springfield (MA)' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (custom displayRender leaf-only)' task_template: hierarchical_path_select secondary_template: null browsergym_goal: In the Event city field, select USA / Massachusetts / Springfield. The task will finish automatically when done. ui_copy: 'Event city: USA / Massachusetts / Springfield. The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Event city”. Custom rendering: displayRender is customized to show ONLY the final leaf label in the input (e.g., it will display “Springfield”, not the full path). Options: Country → State → City, with duplicate city names: - USA → Illinois → Springfield - USA → Massachusetts → Springfield (target) - USA → Oregon → Portland - USA → Maine → Portland Initial state: blank. Distractors: because the input displays only the leaf, the agent must rely on the dropdown columns to ensure the correct state/city branch. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: text clutter: none difficulty: difficulty_bucket: hard tier: L3 axes_ratings: precision_requirement: 4 target_acquisition: 2 density_choice_interference: 4 depth_layering: 3 feedback_dynamics: 2 semantic_observability: 4 disambiguation_load: 1 justification: Duplicate leaf labels combined with leaf-only display reduces observability; correctness requires selecting the right branch in the dropdown, not trusting the input text alone. success_trigger: human_readable: - Event city cascader selected path labels equal [USA, Massachusetts, Springfield]. - Selected path values equal ['us','ma','springfield-ma'] (distinct from the Illinois Springfield value). canonical_predicate: predicate_type: path_equals target_state: path_labels: - USA - Massachusetts - Springfield path_values: - us - ma - springfield-ma tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting USA / Illinois / Springfield (wrong state). - Selecting any Portland entry instead of Springfield. - Selecting only USA / Massachusetts without choosing the city leaf. expected_interaction_path: - Open the Event city cascader dropdown. - Select USA in the first column. - Select Massachusetts in the second column. - Select Springfield in the third column. notes: |- Instrumentation: data-testid='event-city-cascader'. Checker: must use underlying value path (e.g., springfield-ma vs springfield-il) to disambiguate duplicates. - id: cascader-antd-T26 name: 'Three instances: set Tertiary route to EU-West / Ireland / Zone 2' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: hierarchical_path_select secondary_template: null browsergym_goal: Set the Tertiary route field to EU-West / Ireland / Zone 2. The task will finish automatically when done. ui_copy: 'Tertiary route: EU-West / Ireland / Zone 2. The task will finish automatically when done.' setup_description: |- Layout: isolated card centered on the page. Components: three AntD Cascader inputs in a row, each labeled: - Primary route - Secondary route - Tertiary route (target) Options: Region → Country → Zone: - US-East → USA → Zone 1, Zone 2 - EU-West → Ireland → Zone 1, Zone 2 (target) - APAC → Singapore → Zone 1 Initial state: - Primary route preselected: US-East / USA / Zone 1 - Secondary route preselected: APAC / Singapore / Zone 1 - Tertiary route blank Distractors: three nearly identical inputs placed close together. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 3 guidance: text clutter: none difficulty: difficulty_bucket: hard tier: L2 axes_ratings: precision_requirement: 3 target_acquisition: 2 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 1 disambiguation_load: 5 justification: The option tree is moderate, but heavy disambiguation is required to operate the correct one of three adjacent cascader instances. success_trigger: human_readable: - The cascader labeled “Tertiary route” has selected path labels [EU-West, Ireland, Zone 2]. - Tertiary route selected path values equal ['eu-west','ie','zone-2']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - EU-West - Ireland - Zone 2 path_values: - eu-west - ie - zone-2 tolerance: null require_confirm: false confirm_control: null require_correct_instance: true target_instance_label_or_id: tertiary-route terminal_condition: task ends when predicate holds negative_cases: - Setting Primary route or Secondary route instead of Tertiary route. - Selecting EU-West / Ireland / Zone 1 (wrong zone). - Leaving Tertiary route blank. expected_interaction_path: - Locate the cascader labeled Tertiary route. - Open it and select EU-West → Ireland → Zone 2. notes: |- Instrumentation: data-testid='route-primary-cascader', 'route-secondary-cascader', 'route-tertiary-cascader'. Checker: enforce correct instance by testid/label. - id: cascader-antd-T27 name: 'Table cell: set Assign to for ''Password reset'' rule' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (embedded in table cell)' task_template: table_operation secondary_template: null browsergym_goal: In the routing rules table, set the Assign to field for the 'Password reset' row to Support / Tier 2 / Identity. The task will finish automatically when done. ui_copy: 'Table edit: Password reset → Assign to = Support / Tier 2 / Identity. The task will finish automatically when done.' setup_description: |- Layout: table cell scenario. Clutter: medium — a data table with multiple rows/columns plus a small toolbar (Search, Add rule) above it. Table: “Routing rules” with rows like: - Password reset (target row) - Refund request - App crash Only the target row’s “Assign to” column contains an interactive AntD Cascader input; other rows display read-only text. Cascader options: Department → Tier → Specialty: - Support → Tier 1 → General - Support → Tier 2 → Identity (target), Billing - Engineering → On-call → Mobile Initial state: target row currently shows “Support / Tier 1 / General”. Behavior: selecting a leaf updates the cell immediately; no additional Save action is required. scene_context: theme: light spacing: comfortable layout: table_cell placement: center scale: default instances: 1 guidance: text clutter: medium difficulty: difficulty_bucket: hard tier: L2 axes_ratings: precision_requirement: 3 target_acquisition: 3 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 2 disambiguation_load: 4 justification: The cascader is embedded in a dense table context with multiple nearby controls; targeting the correct cell and option requires careful acquisition and disambiguation. success_trigger: human_readable: - The Assign to cascader in the 'Password reset' row has selected path labels [Support, Tier 2, Identity]. - Selected path values equal ['support','t2','identity']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Support - Tier 2 - Identity path_values: - support - t2 - identity tolerance: null require_confirm: false confirm_control: null require_correct_instance: true target_instance_label_or_id: routing-table-password-reset-assign-to terminal_condition: task ends when predicate holds negative_cases: - Changing a different row (if any other editable control is present). - Selecting Support / Tier 2 / Billing (wrong specialty). - Leaving the value at Support / Tier 1 / General. expected_interaction_path: - Locate the Password reset row and its Assign to cascader cell. - Open the cascader dropdown. - Select Support → Tier 2 → Identity. notes: |- Instrumentation: add data-testid='routing-table-password-reset-assign-to' on the Cascader cell. Checker: locate by row key + column id (prefer stable row id over text match if available). - id: cascader-antd-T28 name: 'Modal flow: set Allowed destinations to Asia / Singapore / Downtown' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (inside Modal)' task_template: open_and_select secondary_template: null browsergym_goal: Open the Edit shipping rule modal and set Allowed destinations to Asia / Singapore / Downtown. The task will finish automatically when done. ui_copy: 'Edit shipping rule → Allowed destinations: Asia / Singapore / Downtown. The task will finish automatically when done.' setup_description: |- Layout: modal flow. Clutter: low — a simple page with a single primary button and a modal dialog. Entry point: a button labeled “Edit shipping rule”. Clicking it opens a modal. Inside the modal: - A short description paragraph - The target AntD Cascader labeled “Allowed destinations” - Modal footer buttons: Cancel and Save (these are NOT required for task success) Cascader options: Region → Country → Area: - Asia → Singapore → Downtown (target), Airport - Europe → Germany → Berlin - Americas → USA → New York Initial state: Allowed destinations is blank. Behavior: selection commits on leaf click within the cascader dropdown. scene_context: theme: light spacing: comfortable layout: modal_flow placement: center scale: default instances: 1 guidance: text clutter: low difficulty: difficulty_bucket: hard tier: L2 axes_ratings: precision_requirement: 3 target_acquisition: 2 density_choice_interference: 2 depth_layering: 4 feedback_dynamics: 2 semantic_observability: 2 disambiguation_load: 2 justification: Adds an extra overlay layer (modal + cascader dropdown) and multiple clickable controls, increasing depth and the chance of misclicks. success_trigger: human_readable: - The Allowed destinations cascader inside the open modal has selected path labels [Asia, Singapore, Downtown]. - Selected path values equal ['asia','sg','downtown']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Asia - Singapore - Downtown path_values: - asia - sg - downtown tolerance: null require_confirm: false confirm_control: null require_correct_instance: true target_instance_label_or_id: allowed-destinations terminal_condition: task ends when predicate holds negative_cases: - Selecting Asia / Singapore / Airport instead of Downtown. - Closing the modal without committing the selection (value remains blank). - Setting a cascader outside the modal (none exist). expected_interaction_path: - Click Edit shipping rule to open the modal. - In the modal, open the Allowed destinations cascader. - Select Asia → Singapore → Downtown. notes: |- Instrumentation: data-testid='edit-shipping-modal', data-testid='allowed-destinations-cascader'. Checker: bind to the cascader within the modal container to avoid matching any background elements. - id: cascader-antd-T29 name: 'Drawer flow: set Default assignee group to Customer Success / Enterprise / Onboarding' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader (inside Drawer)' task_template: open_and_select secondary_template: null browsergym_goal: Open the Assignment settings drawer and set Default assignee group to Customer Success / Enterprise / Onboarding. The task will finish automatically when done. ui_copy: 'Assignment settings → Default assignee group: Customer Success / Enterprise / Onboarding. The task will finish automatically when done.' setup_description: |- Layout: drawer flow. Clutter: low — a settings page with a single “Assignment settings” button that opens a right-side drawer. Entry point: click the button labeled “Assignment settings” to open a drawer. Inside the drawer: - Header with a close (×) button - A few toggles (e.g., “Auto-assign new tickets”) - Target AntD Cascader labeled “Default assignee group” Cascader options: Department → Segment → Team: - Customer Success → SMB → Renewals - Customer Success → Enterprise → Onboarding (target), Escalations - Support → Tier 1 → General - Support → Tier 2 → Identity Initial state: Default assignee group is set to “Support / Tier 1 / General”. Interaction nuance: the drawer may constrain available space, and the cascader dropdown must render on top of the drawer content. scene_context: theme: light spacing: comfortable layout: drawer_flow placement: center scale: default instances: 1 guidance: text clutter: low difficulty: difficulty_bucket: hard tier: L3 axes_ratings: precision_requirement: 3 target_acquisition: 3 density_choice_interference: 4 depth_layering: 4 feedback_dynamics: 2 semantic_observability: 2 disambiguation_load: 2 justification: Nested overlay (drawer + dropdown) plus a denser hierarchy increases layering and choice interference, especially in constrained space. success_trigger: human_readable: - Within the open drawer, Default assignee group cascader selected path labels equal [Customer Success, Enterprise, Onboarding]. - Selected path values equal ['cs','ent','onboarding']. canonical_predicate: predicate_type: path_equals target_state: path_labels: - Customer Success - Enterprise - Onboarding path_values: - cs - ent - onboarding tolerance: null require_confirm: false confirm_control: null require_correct_instance: true target_instance_label_or_id: default-assignee-group terminal_condition: task ends when predicate holds negative_cases: - Leaving the default value at Support / Tier 1 / General. - Selecting Customer Success / Enterprise / Escalations (sibling leaf). - Closing the drawer before selecting the target leaf. expected_interaction_path: - Click Assignment settings to open the drawer. - In the drawer, open Default assignee group cascader. - Select Customer Success → Enterprise → Onboarding. notes: |- Instrumentation: data-testid='assignment-settings-drawer', data-testid='default-assignee-group-cascader'. Checker: ensure it reads the cascader inside the drawer context. - id: cascader-antd-T30 name: 'Dark + visual: match Target breadcrumb for Policy category' canonical_type: cascader implementation_source: antd implementation_variant: null implementation_component: 'AntD: Cascader' task_template: match_reference secondary_template: null browsergym_goal: In Policy category, select the path shown in the Target breadcrumb preview. The task will finish automatically when done. ui_copy: 'Policy category: match the Target breadcrumb shown. The task will finish automatically when done.' setup_description: |- Theme: dark. Layout: isolated card centered on the page. Component: one AntD Cascader labeled “Policy category”. Guidance element: a “Target breadcrumb” preview above the field shows the desired path as styled crumb chips. Options: Domain → Area → Policy: - Compliance → Data → Retention (target) - Compliance → Data → Classification - Security → Access → MFA Initial state: blank. Distractors: the Compliance/Data branch has multiple similar policy leaves. scene_context: theme: dark spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: visual clutter: none difficulty: difficulty_bucket: hard tier: L2 axes_ratings: precision_requirement: 3 target_acquisition: 2 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 1 semantic_observability: 4 disambiguation_load: 1 justification: Visual reference matching in dark theme requires accurately interpreting the on-page breadcrumb and selecting the corresponding leaf among similar siblings. success_trigger: human_readable: - Policy category cascader selection matches the Target breadcrumb preview. - Selected path labels equal [Compliance, Data, Retention] and values equal ['compliance','data','retention']. canonical_predicate: predicate_type: matches_reference target_state: reference_element_id: target-policy-breadcrumb path_labels: - Compliance - Data - Retention path_values: - compliance - data - retention tolerance: null require_confirm: false confirm_control: null require_correct_instance: false target_instance_label_or_id: null terminal_condition: task ends when predicate holds negative_cases: - Selecting Compliance / Data / Classification (wrong sibling leaf). - Selecting a Security policy path instead. - Leaving the field blank while the Target breadcrumb is visible. expected_interaction_path: - Read the Target breadcrumb preview. - Open Policy category cascader and select Compliance → Data → Retention. notes: |- Instrumentation: data-testid='policy-category-cascader' and data-testid='target-policy-breadcrumb'. Checker: compare cascader value path to reference breadcrumb canonical path.