- id: file_dropzone-antd-T01 name: Upload resume PDF to Attachments (single file) canonical_type: file_dropzone implementation_source: antd implementation_variant: dragger implementation_component: 'AntD: Upload.Dragger' task_template: file_upload secondary_template: null browsergym_goal: Upload the file “resume-alex-chen.pdf” to the “Attachments” drop area. The task will finish automatically when done. ui_copy: Upload the file “resume-alex-chen.pdf” to the “Attachments” drop area. setup_description: |- A single isolated card is centered in the viewport (light theme, comfortable spacing, default scale). The card title is “Job application” and it contains one AntD Upload.Dragger instance labeled “Attachments”. The drop area shows the standard AntD dragger affordance (upload icon + “Click or drag file to this area to upload”). Clicking anywhere inside the drop area opens an in-page “Sample files” picker (not an OS dialog) listing 3 files: - resume-alex-chen.pdf - avatar-green.png - project-brief.txt The Upload.Dragger is configured for a single file (maxCount=1), with auto-upload enabled. After a file is chosen/dropped, an AntD upload list appears directly below the drop area with filename text, a progress indicator, and a final success state (“Uploaded” with a check icon). Each list item has a remove (×) icon. Initial state: no file selected/uploaded. No other interactive UI is present on the page. 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: 1 feedback_dynamics: 2 semantic_observability: 5 disambiguation_load: 1 justification: Large, central dropzone with a tiny candidate set; success is clearly visible as a filename in the upload list, with only a brief upload-progress wait. success_trigger: human_readable: - 'The dropzone labeled “Attachments” contains exactly one file: resume-alex-chen.pdf.' - That file’s final status is “uploaded” (not uploading/error). - No additional files are present in the Attachments upload list. canonical_predicate: predicate_type: equals target_state: files: - file_key: resume-alex-chen.pdf file_name: resume-alex-chen.pdf required_status: uploaded 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: - A different file is uploaded (e.g., avatar-green.png or project-brief.txt). - The correct file appears but is still in an “uploading/in progress” state. - The correct file appears with an error/rejected status. - More than one file is present in the upload list. - A file is uploaded in the wrong dropzone instance (not applicable here, but must not be counted if the page is extended). expected_interaction_path: - Click the Attachments drop area to open the in-page picker (or drag the file if available). - Select “resume-alex-chen.pdf”. - Wait until the upload list shows the file as uploaded/success. notes: |- Instrumentation: expose a stable identifier for the Upload.Dragger instance label (“Attachments”) and for each upload list item status. Checker should read canonical file_key from the internal file model (avoid relying on truncated UI text). - id: file_dropzone-antd-T02 name: Upload profile photo (PNG) with thumbnail preview canonical_type: file_dropzone implementation_source: antd implementation_variant: dragger implementation_component: 'AntD: Upload.Dragger' task_template: file_upload secondary_template: null browsergym_goal: Upload the file “avatar-green.png” to the “Profile photo” drop area. The task will finish automatically when done. ui_copy: Upload the file “avatar-green.png” to the “Profile photo” drop area. setup_description: |- The page uses the baseline scene: a centered isolated card (light theme, comfortable spacing, default scale). Inside the card there is one AntD Upload.Dragger labeled “Profile photo”. The drop area is configured to accept images only (accept=image/*) and uses a picture-style upload list: after upload, the list shows a small image thumbnail preview next to the filename and status. Clicking the drop area opens an in-page “Sample files” picker listing 3 images: - avatar-green.png - avatar-blue.png - id-card.jpg Auto-upload is enabled and a short progress indicator appears before the final “uploaded” status. Initial state: empty (no selected file). There are no extra buttons or other form fields. 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: 2 semantic_observability: 4 disambiguation_load: 1 justification: Only a few visually similar image choices; the target file name is explicitly given and the final state is visible via filename + thumbnail in the upload list. success_trigger: human_readable: - 'The dropzone labeled “Profile photo” contains exactly one file: avatar-green.png.' - The file status is “uploaded”. - No other files are present in the Profile photo upload list. canonical_predicate: predicate_type: equals target_state: files: - file_key: avatar-green.png file_name: avatar-green.png required_status: uploaded 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: - Uploading avatar-blue.png or id-card.jpg instead of avatar-green.png. - The correct file is present but still uploading. - The correct file is present but rejected/error (e.g., wrong MIME type handling). - Multiple files present when maxCount=1. - Uploading to a non-target dropzone instance (not applicable here). expected_interaction_path: - Click the Profile photo drop area to open the in-page picker. - Choose “avatar-green.png”. - Wait for the upload to finish and the list to show success. notes: AntD picture-style upload list improves semantic observability via both filename and thumbnail; ensure the checker uses canonical file_key, not thumbnail URL. - id: file_dropzone-antd-T03 name: Upload two receipt PDFs (multi-select) canonical_type: file_dropzone implementation_source: antd implementation_variant: dragger implementation_component: 'AntD: Upload.Dragger' task_template: file_upload secondary_template: null browsergym_goal: Upload the files “invoice-1042.pdf” and “invoice-1043.pdf” to the “Receipts” drop area. The task will finish automatically when done. ui_copy: Upload the files “invoice-1042.pdf” and “invoice-1043.pdf” to the “Receipts” drop area. setup_description: |- Baseline scene (light/comfortable/centered isolated card). One AntD Upload.Dragger instance is labeled “Receipts”. The dropzone is configured for multiple files (multiple=true, maxCount=5) and accepts PDFs (accept=.pdf). Clicking the drop area opens an in-page “Sample files” picker with checkboxes (multi-select) listing: - invoice-1042.pdf - invoice-1043.pdf - receipt-sample.pdf After selection, the AntD upload list shows one row per file with filename and per-item status/progress. Initial state: no uploaded files. scene_context: theme: light spacing: comfortable 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: 2 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 5 disambiguation_load: 1 justification: Requires selecting an exact set of two files, but options are few and the upload list gives explicit filename feedback per item. success_trigger: human_readable: - 'The dropzone labeled “Receipts” contains exactly two files: invoice-1042.pdf and invoice-1043.pdf.' - Both files have final status “uploaded”. - No extra files are present in the Receipts list. canonical_predicate: predicate_type: equals target_state: files: - file_key: invoice-1042.pdf file_name: invoice-1042.pdf required_status: uploaded - file_key: invoice-1043.pdf file_name: invoice-1043.pdf required_status: uploaded 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: - Only one of the two target files is uploaded (partial completion). - receipt-sample.pdf is uploaded instead of one of the targets. - A target file is present but still uploading or error. - Any extra file is present beyond the two targets. - Files were uploaded in the wrong dropzone instance (not applicable here). expected_interaction_path: - Click the Receipts drop area to open the in-page picker. - Select invoice-1042.pdf and invoice-1043.pdf (checkboxes) and confirm selection in the picker. - Wait for both uploads to finish. notes: Checker should treat file order as irrelevant (order_insensitive=true). - id: file_dropzone-antd-T04 name: Remove the pre-attached text file (empty the dropzone) canonical_type: file_dropzone implementation_source: antd implementation_variant: dragger implementation_component: 'AntD: Upload.Dragger' task_template: file_manage secondary_template: null browsergym_goal: Remove the file “project-brief.txt” from the “Supporting files” drop area so that no files remain there. The task will finish automatically when done. ui_copy: Remove the file “project-brief.txt” from the “Supporting files” drop area so that no files remain there. setup_description: |- Baseline scene: one centered isolated card (light theme, comfortable spacing). The card contains one AntD Upload.Dragger labeled “Supporting files” (multiple=false). Initial state: the upload list already contains exactly one successfully uploaded item: - project-brief.txt (status: uploaded) The list item shows the filename and a remove (×) icon. Clicking remove immediately removes the item (no confirmation for this task). The drop area remains visible above the list for adding new files, but adding files is not required for success. 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: 5 disambiguation_load: 1 justification: Only one visible file needs removal; the remove affordance is directly adjacent to the filename and the empty state is obvious. success_trigger: human_readable: - The dropzone labeled “Supporting files” has an empty file list (no selected/uploaded files). canonical_predicate: predicate_type: equals target_state: files: [] 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: - project-brief.txt remains present in the file list. - A different file is removed while project-brief.txt remains (if additional files are added). - The file shows an intermediate removing/loading state but is not actually removed. - Files were removed from the wrong dropzone instance (not applicable here). expected_interaction_path: - Locate the upload list under Supporting files. - Click the remove (×) icon on project-brief.txt. - Verify the list is empty. notes: For robustness, ensure the checker distinguishes between hidden list items vs true removal from component state. - id: file_dropzone-antd-T05 name: Replace an existing W-9 PDF (maxCount=1) canonical_type: file_dropzone implementation_source: antd implementation_variant: dragger implementation_component: 'AntD: Upload.Dragger' task_template: file_manage secondary_template: null browsergym_goal: In the “W-9 form” drop area, replace “w9-2025.pdf” with “w9-2026.pdf”. The task will finish automatically when done. ui_copy: 'Replace the W-9 form: w9-2025.pdf → w9-2026.pdf' setup_description: |- The page is a small “Vendor onboarding” form section (light theme, comfortable spacing) with several non-target fields (company name text input, address text input, and a disabled “Save” button). These fields are distractors and are not required. The target component is one AntD Upload.Dragger labeled “W-9 form”, configured as a single-file uploader (maxCount=1, accept=.pdf). Initial state: the upload list already contains: - w9-2025.pdf (status: uploaded) The file row includes a remove icon and a prominent inline action link “Replace file”. Clicking the drop area or “Replace file” opens the in-page “Sample files” picker listing: - w9-2025.pdf - w9-2026.pdf - w8ben.pdf Selecting a file replaces the existing item (the list should end with exactly one uploaded PDF). 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: 3 target_acquisition: 2 density_choice_interference: 3 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 5 disambiguation_load: 2 justification: Similar filenames increase confusion and the task requires a precise end state (exactly one correct file) within a busier form section. success_trigger: human_readable: - 'The dropzone labeled “W-9 form” contains exactly one file: w9-2026.pdf.' - The file status is “uploaded”. canonical_predicate: predicate_type: equals target_state: files: - file_key: w9-2026.pdf file_name: w9-2026.pdf required_status: uploaded 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: - w9-2025.pdf remains as the uploaded file. - w8ben.pdf is uploaded instead. - Both files are present (should not happen with maxCount=1, but must not count as success). - The correct file is present but not uploaded (still uploading/error). - The user modifies a non-target field but does not replace the file (should not count). expected_interaction_path: - Locate the W-9 form Upload.Dragger and its existing file row. - Click “Replace file” (or click the drop area) to open the in-page picker. - Select w9-2026.pdf and wait for uploaded status. notes: If using AntD’s onRemove/beforeUpload hooks, ensure replacement updates the canonical file_key deterministically. - id: file_dropzone-antd-T06 name: Upload verification document in a two-dropzone modal canonical_type: file_dropzone implementation_source: antd implementation_variant: dragger implementation_component: 'AntD: Upload.Dragger' task_template: file_upload secondary_template: null browsergym_goal: In the upload dialog, add “id-card.jpg” to the “Verification document” drop area (not the Profile photo one). The task will finish automatically when done. ui_copy: 'Upload dialog: Add id-card.jpg to Verification document' setup_description: |- A modal dialog is open on page load (modal_flow layout) against a dimmed background (light theme overall). The modal title is “Identity verification” and it contains TWO AntD Upload.Dragger instances stacked vertically: 1) “Profile photo” (accept=image/*, maxCount=1) — currently empty 2) “Verification document” (accept=image/*,.pdf, maxCount=1) — currently empty ← TARGET Each drop area shows the standard AntD dragger affordance. Clicking a drop area opens an in-page “Sample files” picker. The picker lists 4 files: - id-card.jpg - utility-bill.pdf - avatar-green.png - avatar-blue.png Auto-upload is enabled; after selection, the chosen file appears in that dropzone’s upload list with a success indicator. The background page has no other interactables beyond the modal (the modal must remain open; no “Save” is required). scene_context: theme: light spacing: comfortable layout: modal_flow placement: center scale: default instances: 2 guidance: text clutter: low difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 3 density_choice_interference: 2 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 5 disambiguation_load: 3 justification: Requires selecting the correct dropzone among two similar instances inside a modal, with standard upload feedback but higher disambiguation demand. success_trigger: human_readable: - Only the dropzone labeled “Verification document” is modified. - 'The “Verification document” dropzone contains exactly one file: id-card.jpg, with status “uploaded”.' - The “Profile photo” dropzone remains empty. canonical_predicate: predicate_type: equals target_state: files: - file_key: id-card.jpg file_name: id-card.jpg required_status: uploaded order_insensitive: true tolerance: null require_confirm: false confirm_control: null require_correct_instance: true target_instance_label_or_id: Verification document terminal_condition: task ends when predicate holds negative_cases: - id-card.jpg is uploaded to “Profile photo” instead of “Verification document”. - A different file (utility-bill.pdf or an avatar image) is uploaded to the target dropzone. - The correct file is present but still uploading or error. - Both dropzones contain files when only the target should. - The modal is closed without uploading (should not count). expected_interaction_path: - In the modal, locate the ‘Verification document’ drop area. - Click it to open the in-page picker (or drop the file). - Choose id-card.jpg and wait for uploaded status in that instance’s list. notes: Checker must bind the predicate to the correct instance (label text or stable data-testid) and also verify the non-target instance remains unchanged. - id: file_dropzone-antd-T07 name: Drag a logo SVG into the dropzone (dark theme, corner placement) canonical_type: file_dropzone implementation_source: antd implementation_variant: dragger implementation_component: 'AntD: Upload.Dragger' task_template: drag_operation secondary_template: null browsergym_goal: Drag and drop the file tile “logo.svg” into the “Brand assets” drop area. The task will finish automatically when done. ui_copy: Drag logo.svg into Brand assets setup_description: |- The UI is rendered in dark theme and the main card is anchored to the bottom-right of the viewport (placement=bottom_right). Spacing is comfortable and the dropzone is default size. Inside the card there is one AntD Upload.Dragger labeled “Brand assets” (accept=.svg,.png, multiple=false). Below the drop area is a small “File tray” strip (part of the test harness) containing 3 draggable file tiles: - logo.svg ← TARGET - logo-old.svg - brand-guidelines.pdf Each tile shows a file-type icon and the filename. The intended interaction is to drag a tile and drop it onto the dashed AntD dragger area. On successful drop, the file appears in the AntD upload list with an upload progress line, then status “uploaded”. Initial state: no files uploaded in Brand assets. scene_context: theme: dark spacing: comfortable layout: isolated_card placement: bottom_right scale: default instances: 1 guidance: text clutter: low difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 3 density_choice_interference: 2 depth_layering: 1 feedback_dynamics: 2 semantic_observability: 4 disambiguation_load: 2 justification: Requires a drag-and-drop gesture and accurate target acquisition in a corner-anchored layout; otherwise feedback is clear via filename in the upload list. success_trigger: human_readable: - 'The dropzone labeled “Brand assets” contains exactly one file: logo.svg.' - The file status is “uploaded”. canonical_predicate: predicate_type: equals target_state: files: - file_key: logo.svg file_name: logo.svg required_status: uploaded 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: - Dropping logo-old.svg or brand-guidelines.pdf instead of logo.svg. - Dropping the correct tile but not onto the dropzone (no file appears in the list). - The file appears but remains in uploading/error state. - An extra file is present (should be single-file). - Interacting with the tray without completing the drop. expected_interaction_path: - Locate the file tray under the dropzone. - Click-and-drag the “logo.svg” tile onto the Brand assets dragger region. - Wait for upload success status. notes: 'Instrumentation: ensure draggable tiles set a deterministic file_key in the drop handler so the checker can match logo.svg reliably.' - id: file_dropzone-antd-T08 name: Scroll to Diagnostics and upload a specific log bundle (compact, small) canonical_type: file_dropzone implementation_source: antd implementation_variant: dragger implementation_component: 'AntD: Upload.Dragger' task_template: scroll_find secondary_template: null browsergym_goal: Scroll to the “Diagnostics” section and upload “logs-2026-02-01.zip” to the “Diagnostic logs” drop area. The task will finish automatically when done. ui_copy: 'Diagnostics: Upload logs-2026-02-01.zip to Diagnostic logs' setup_description: |- The page is a settings_panel with a scrollable main content area (light theme). Spacing is compact and the component is rendered at small scale (smaller padding and typography than baseline). Several settings sections appear above the fold (Network, Appearance, Privacy), each with toggles and dropdowns (distractors). The target section “Diagnostics” is below the fold and is not initially visible without scrolling. Inside “Diagnostics”, there is one AntD Upload.Dragger labeled “Diagnostic logs”. It is configured to accept .zip files only (accept=.zip, multiple=false, maxCount=1) and uses auto-upload with a visible progress bar. Clicking the drop area opens the in-page “Sample files” picker listing 4 zip files with similar names: - logs-2026-01-31.zip - logs-2026-02-01.zip ← TARGET - logs-2026-02-02.zip - screenshots.zip Initial state: no file uploaded. The scroll position starts at the top of the settings page. scene_context: theme: light spacing: compact layout: settings_panel placement: center scale: small instances: 1 guidance: text clutter: medium difficulty: difficulty_bucket: hard tier: L2 axes_ratings: precision_requirement: 3 target_acquisition: 3 density_choice_interference: 4 depth_layering: 2 feedback_dynamics: 3 semantic_observability: 5 disambiguation_load: 3 justification: Requires scrolling to find the target section, then selecting the correct zip among multiple similar filenames in a compact/small UI with more distractors. success_trigger: human_readable: - 'The dropzone labeled “Diagnostic logs” contains exactly one file: logs-2026-02-01.zip.' - The file status is “uploaded”. canonical_predicate: predicate_type: equals target_state: files: - file_key: logs-2026-02-01.zip file_name: logs-2026-02-01.zip required_status: uploaded 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: - Uploading a different zip (e.g., logs-2026-02-02.zip) due to filename similarity. - The correct zip is selected but the file remains uploading. - The correct zip is present but rejected/error. - Uploading to an incorrect dropzone elsewhere in the settings (if any). - Failing to scroll to the Diagnostics section (no upload performed). expected_interaction_path: - Scroll down the settings panel until the “Diagnostics” header is visible. - Click the Diagnostic logs drop area to open the in-page picker. - Select logs-2026-02-01.zip and wait for upload success. notes: Checker should ensure the scroll_find requirement is not scored separately; success is still purely the dropzone state. - id: file_dropzone-antd-T09 name: Upload the banner that matches a reference preview (visual match) canonical_type: file_dropzone implementation_source: antd implementation_variant: dragger implementation_component: 'AntD: Upload.Dragger' task_template: match_reference secondary_template: null browsergym_goal: Upload the banner image that matches the reference preview to the “Banner image” drop area. The task will finish automatically when done. ui_copy: Upload the banner image that matches the reference preview to the “Banner image” drop area. setup_description: |- A centered isolated card is shown (light theme, comfortable spacing, default scale). The card is split into two columns: - Left column: a “Reference preview” panel showing the target banner image (a small wide thumbnail). - Right column: one AntD Upload.Dragger labeled “Banner image” (accept=image/png, maxCount=1). Below the drop area is a grid of 4 candidate banner thumbnails (part of the test harness). Each thumbnail is labeled only with a neutral ID: - banner-01.png - banner-02.png - banner-03.png ← (this one visually matches the reference) - banner-04.png The intended interaction is to drag the correct thumbnail tile onto the AntD dragger area (or click the drop area to open the in-page picker showing the same 4 thumbnails). Auto-upload is enabled; after selection, the upload list under “Banner image” shows a picture thumbnail, filename, and uploaded status. Initial state: no file selected/uploaded. scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: visual clutter: low difficulty: difficulty_bucket: hard tier: L2 axes_ratings: precision_requirement: 3 target_acquisition: 3 density_choice_interference: 4 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 3 disambiguation_load: 2 justification: The goal is specified visually, requiring the agent to compare thumbnails; multiple similar candidates increase interference and the correct outcome must be an exact file match. success_trigger: human_readable: - The dropzone labeled “Banner image” contains exactly one file whose file_key is banner-03.png. - That file has status “uploaded”. canonical_predicate: predicate_type: equals target_state: files: - file_key: banner-03.png file_name: banner-03.png required_status: uploaded 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: - Uploading banner-01.png, banner-02.png, or banner-04.png instead of the one matching the reference. - The correct banner file is present but still uploading. - The correct banner file is present but rejected/error. - Multiple files are present (should be maxCount=1). - The agent interacts with the reference preview but does not upload a file. expected_interaction_path: - Compare the reference preview to the candidate banner thumbnails. - Drag the matching candidate (banner-03.png) onto the Banner image drop area (or select it in the in-page picker). - Wait for the uploaded status. notes: To preserve the ‘visual match’ intent, keep candidate labels non-semantic (banner-01…04) so filename reading does not trivially solve the task. - id: file_dropzone-antd-T10 name: Cancel a removal confirmation so the file stays attached canonical_type: file_dropzone implementation_source: antd implementation_variant: dragger implementation_component: 'AntD: Upload.Dragger' task_template: confirm_cancel secondary_template: null browsergym_goal: In the “Signed contract” drop area, try to remove “contract-signed.pdf” but click “Cancel” in the confirmation so the file stays attached. The task will finish automatically when done. ui_copy: 'Signed contract: remove → Cancel (keep contract-signed.pdf attached)' setup_description: |- The page is a light-theme form section titled “Contract submission” (comfortable spacing). Two AntD Upload.Dragger instances are shown one after another: - “Signed contract” ← TARGET instance - “Appendix (optional)” Initial state: - The “Signed contract” upload list already contains contract-signed.pdf (status: uploaded). - The “Appendix (optional)” list is empty. In the “Signed contract” upload list, clicking the remove (×) icon opens an AntD Popconfirm anchored to the list item. The Popconfirm text reads “Remove this file?” with two buttons: “Remove” and “Cancel”. Clicking “Cancel” closes the popover and keeps the file in place. No other form controls are required for success; the only meaningful interaction is canceling the removal confirmation. scene_context: theme: light spacing: comfortable layout: form_section placement: center scale: default instances: 2 guidance: text clutter: low difficulty: difficulty_bucket: hard tier: L2 axes_ratings: precision_requirement: 2 target_acquisition: 3 density_choice_interference: 1 depth_layering: 3 feedback_dynamics: 3 semantic_observability: 5 disambiguation_load: 4 justification: Requires correctly targeting the remove icon in the right instance, handling a transient confirmation popover, and selecting ‘Cancel’ without changing the final file state. success_trigger: human_readable: - A remove confirmation for contract-signed.pdf in the “Signed contract” instance was opened and cancelled (Cancel clicked at least once). - The “Signed contract” dropzone still contains contract-signed.pdf with status “uploaded”. - The “Appendix (optional)” dropzone remains empty. canonical_predicate: predicate_type: equals target_state: files: - file_key: contract-signed.pdf file_name: contract-signed.pdf required_status: uploaded order_insensitive: true tolerance: null require_confirm: false confirm_control: null require_correct_instance: true target_instance_label_or_id: Signed contract terminal_condition: task ends when predicate holds required_interaction_events: - remove_confirmation_cancelled negative_cases: - Clicking “Remove” instead of “Cancel” (file disappears). - Canceling a confirmation in the wrong dropzone instance (Appendix) if a file is added there. - Never opening the confirmation popover (task must not auto-succeed without the cancel event). - The file remains but is in a non-uploaded/error state. - Closing the popover by clicking outside without triggering the explicit Cancel event (should not count as success if event instrumentation requires Cancel). expected_interaction_path: - In the Signed contract upload list, click the remove (×) icon on contract-signed.pdf. - In the Popconfirm, click “Cancel”. - Verify the file is still listed as uploaded. notes: |- To prevent trivial success, add an instrumented event counter (e.g., window.__cb_events.remove_confirmation_cancelled++) when the Popconfirm Cancel button is clicked. Checker should require that counter >= 1 in addition to the final file list state. - id: file_dropzone-mui-T01 name: Upload resume PDF in a MUI-styled dropzone card canonical_type: file_dropzone implementation_source: mui implementation_variant: composite implementation_component: 'MUI: react-dropzone (useDropzone) + MUI Card/Paper (composite)' task_template: file_upload secondary_template: null browsergym_goal: Upload the file “resume-alex-chen.pdf” to the “Upload resume” drop area. The task will finish automatically when done. ui_copy: 'Upload resume: resume-alex-chen.pdf' setup_description: |- Baseline scene: a centered isolated card (light theme, comfortable spacing, default scale). The dropzone implementation is a MUI Card with a dashed-border Paper area wired via react-dropzone (composite implementation). The drop area is labeled “Upload resume” and contains helper text: “Drag a file here, or click to choose from sample files”. Clicking the drop area opens a small in-page MUI Dialog titled “Sample files” (not an OS dialog) with a simple list: - resume-alex-chen.pdf ← TARGET - project-brief.txt - avatar-green.png After selection, the dialog closes and the chosen file appears below the drop area as a single MUI Chip with: filename text + an “Uploaded” status label + a small remove (×) icon on the chip. Initial state: no selected files. 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: 2 semantic_observability: 5 disambiguation_load: 1 justification: Large drop target with a short file list; the final selected file is clearly rendered as a chip with filename text. success_trigger: human_readable: - 'The dropzone labeled “Upload resume” contains exactly one file: resume-alex-chen.pdf.' - The file’s status is “uploaded”. canonical_predicate: predicate_type: equals target_state: files: - file_key: resume-alex-chen.pdf file_name: resume-alex-chen.pdf required_status: uploaded 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: - A different file is selected/uploaded. - The file is present but still uploading/error. - More than one file is present (single-file configuration). - Uploading into a non-target dropzone instance (not applicable here). expected_interaction_path: - Click the Upload resume drop area to open the Sample files dialog. - Select resume-alex-chen.pdf. - Confirm the chip shows the correct filename and uploaded status. notes: 'Composite note: MUI has no official drag-and-drop file upload component; this task uses react-dropzone styled with MUI components.' - id: file_dropzone-mui-T02 name: Upload profile photo using mixed text + visual guidance canonical_type: file_dropzone implementation_source: mui implementation_variant: composite implementation_component: 'MUI: react-dropzone (useDropzone) + MUI Card/Paper (composite)' task_template: file_upload secondary_template: null browsergym_goal: Upload the file “avatar-blue.png” to the “Profile photo” drop area. The task will finish automatically when done. ui_copy: 'Profile photo: upload avatar-blue.png (matches the preview)' setup_description: |- Baseline scene: centered isolated card, light theme, comfortable spacing. The dropzone is a MUI-styled dashed Paper area labeled “Profile photo” (react-dropzone composite). To the right of the dropzone label, a small “Desired preview” thumbnail is displayed showing the target avatar (a blue square avatar). The page also shows the target filename in text (“avatar-blue.png”), making guidance mixed (text + visual). Clicking the drop area opens an in-page MUI Dialog (“Sample files”) listing 3 images: - avatar-blue.png ← TARGET - avatar-green.png - id-card.jpg After selection, the dropzone area shows an inline image preview and a chip row with the filename and “uploaded” status. Initial state: empty (no selected files). scene_context: theme: light spacing: comfortable layout: isolated_card placement: center scale: default instances: 1 guidance: mixed 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: 2 semantic_observability: 4 disambiguation_load: 1 justification: Only a few image choices and both filename and preview are available; the resulting state is easy to verify via filename chip and thumbnail. success_trigger: human_readable: - 'The “Profile photo” dropzone contains exactly one file: avatar-blue.png.' - The file status is “uploaded”. canonical_predicate: predicate_type: equals target_state: files: - file_key: avatar-blue.png file_name: avatar-blue.png required_status: uploaded 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: - Uploading avatar-green.png or id-card.jpg. - The correct file is present but uploading/error. - Multiple files attached. - Uploading to a different dropzone instance (not applicable). expected_interaction_path: - Click the Profile photo drop area to open the Sample files dialog. - Select avatar-blue.png. - Wait for the preview/chip to show uploaded status. notes: Mixed guidance tasks allow analysis of whether agents rely on text labels vs visual previews. - id: file_dropzone-mui-T03 name: Clear all attachments using a Clear button canonical_type: file_dropzone implementation_source: mui implementation_variant: composite implementation_component: 'MUI: react-dropzone (useDropzone) + MUI Card/Paper (composite)' task_template: clear_reset secondary_template: null browsergym_goal: Clear all files from the “Attachments” drop area so it becomes empty. The task will finish automatically when done. ui_copy: 'Attachments: Clear all' setup_description: |- Baseline scene: centered isolated card, light theme, comfortable spacing, default scale. The dropzone is a MUI Card section labeled “Attachments” (react-dropzone composite). Under the dashed drop area, there is a file list rendered as MUI Chips and a small action row containing a “Clear” button. Initial state: two uploaded files are already present as chips: - project-brief.txt (uploaded) - data_export.csv (uploaded) Clicking the “Clear” button removes all chips at once and returns the dropzone to its empty state. No confirmation dialog is used for this task. 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: 5 disambiguation_load: 1 justification: Single explicit Clear control with immediate feedback; the empty state is visually obvious and easy to verify. success_trigger: human_readable: - The dropzone labeled “Attachments” has no files (no chips/items listed). canonical_predicate: predicate_type: equals target_state: files: [] 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: - At least one file chip remains after attempting to clear. - Only one chip is removed (partial clear). - Files are cleared from a different component (not applicable here). - The UI shows chips removed but the internal file list state still contains entries (checker must use canonical state). expected_interaction_path: - Locate the Attachments card. - Click the “Clear” button under the file chips. - Verify no chips remain. notes: Make sure the Clear action is within the dropzone component boundary for attribution (component-level task). - id: file_dropzone-mui-T04 name: Drag a file tile into a top-left dropzone (MUI composite) canonical_type: file_dropzone implementation_source: mui implementation_variant: composite implementation_component: 'MUI: react-dropzone (useDropzone) + MUI Card/Paper (composite)' task_template: drag_operation secondary_template: null browsergym_goal: Drag and drop the file tile “logo.svg” into the “Brand logo” drop area. The task will finish automatically when done. ui_copy: 'Brand logo: drag logo.svg into the drop area' setup_description: |- The page is a small settings_panel section and the main dropzone card is positioned near the top-left of the viewport (placement=top_left). Theme is light, spacing comfortable, scale default. The target is a MUI-styled dashed Paper drop area labeled “Brand logo” (react-dropzone composite, accept=.svg, maxCount=1). To support drag interaction, a compact “File tray” appears directly below the card with three draggable tiles: - logo.svg ← TARGET - logo-mono.svg - brand-guidelines.pdf Dragging a tile over the drop area changes the border highlight (hover feedback). On drop, the selected file is added and rendered as a single chip with filename and “uploaded” status. Initial state: empty. scene_context: theme: light spacing: comfortable layout: settings_panel placement: top_left scale: default instances: 1 guidance: text clutter: low difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 4 density_choice_interference: 2 depth_layering: 1 feedback_dynamics: 2 semantic_observability: 4 disambiguation_load: 2 justification: Corner placement and a drag gesture increase target acquisition difficulty; however, the file set is small and feedback is explicit via the resulting filename chip. success_trigger: human_readable: - 'The “Brand logo” dropzone contains exactly one file: logo.svg, with status “uploaded”.' canonical_predicate: predicate_type: equals target_state: files: - file_key: logo.svg file_name: logo.svg required_status: uploaded 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: - Dropping logo-mono.svg or brand-guidelines.pdf instead of logo.svg. - Dropping the tile outside the dropzone so no file is added. - The file appears but remains in uploading/error state. - More than one file is present when maxCount=1. expected_interaction_path: - Locate the file tray beneath the Brand logo card. - Drag the “logo.svg” tile onto the dashed drop area. - Confirm the resulting chip shows logo.svg as uploaded. notes: If drag is implemented via dataTransfer, ensure drag-start sets a deterministic file_key. - id: file_dropzone-mui-T05 name: Upload two specific gallery images in compact spacing canonical_type: file_dropzone implementation_source: mui implementation_variant: composite implementation_component: 'MUI: react-dropzone (useDropzone) + MUI Card/Paper (composite)' task_template: file_upload secondary_template: null browsergym_goal: Upload the files “gallery-01.jpg” and “gallery-03.jpg” to the “Gallery images” drop area. The task will finish automatically when done. ui_copy: 'Gallery images: add gallery-01.jpg and gallery-03.jpg' setup_description: |- The page is a centered isolated card in light theme but uses compact spacing (tighter padding and smaller gaps). The dropzone is a MUI dashed Paper area labeled “Gallery images” (react-dropzone composite), configured for multiple image files (accept=image/*, maxFiles=5). Clicking the drop area opens an in-page MUI Dialog (“Sample files”) with a compact checkbox list of candidate images: - gallery-01.jpg ← TARGET - gallery-02.jpg - gallery-03.jpg ← TARGET - gallery-04.jpg - avatar-blue.png - avatar-green.png The dialog has an “Add selected” button. After adding, the dropzone shows a stacked list of image thumbnails with filenames and per-item “uploaded” badges. Initial state: no images selected. scene_context: theme: light spacing: compact 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: 3 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 4 disambiguation_load: 1 justification: Multiple required targets and compact checkbox spacing raise interference and precision demands, but the component remains centered with clear filename feedback. success_trigger: human_readable: - 'The “Gallery images” dropzone contains exactly two files: gallery-01.jpg and gallery-03.jpg.' - Both files are in status “uploaded”. canonical_predicate: predicate_type: equals target_state: files: - file_key: gallery-01.jpg file_name: gallery-01.jpg required_status: uploaded - file_key: gallery-03.jpg file_name: gallery-03.jpg required_status: uploaded 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: - Only one of the two target images is attached (partial selection). - A non-target image (gallery-02.jpg, gallery-04.jpg, or an avatar) is attached instead. - Any extra image is attached beyond the two targets. - A target image is present but not uploaded (uploading/error). expected_interaction_path: - Click the Gallery images drop area to open the dialog. - Tick gallery-01.jpg and gallery-03.jpg and click “Add selected”. - Verify both appear as uploaded in the list. notes: Compact spacing is intended to stress target acquisition on checkboxes without adding unrelated clutter. - id: file_dropzone-mui-T06 name: Remove one specific receipt from a populated list canonical_type: file_dropzone implementation_source: mui implementation_variant: composite implementation_component: 'MUI: react-dropzone (useDropzone) + MUI Card/Paper (composite)' task_template: file_manage secondary_template: null browsergym_goal: In the “Receipts” drop area, remove “invoice-1043.pdf” and keep the other files. The task will finish automatically when done. ui_copy: 'Receipts: remove invoice-1043.pdf only' setup_description: |- The page is a light-theme form_section with moderate clutter: there are several read-only expense fields (amount, date, merchant) and the target upload card. Spacing is comfortable, scale default. The target component is a MUI-styled dropzone labeled “Receipts” (react-dropzone composite, multiple=true). Initial state: three files are already attached and shown in a vertical MUI List under the drop area: - invoice-1042.pdf (uploaded) - invoice-1043.pdf (uploaded) ← TARGET for removal - receipt-sample.pdf (uploaded) Each list row has a trailing trash/delete icon button. Clicking the delete icon removes that single file row immediately and shows a brief Snackbar (“Removed …”). The task requires removing only the specified file while leaving the other two intact. scene_context: theme: light spacing: comfortable layout: form_section placement: center scale: default instances: 1 guidance: text clutter: medium difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 3 target_acquisition: 3 density_choice_interference: 3 depth_layering: 1 feedback_dynamics: 2 semantic_observability: 5 disambiguation_load: 2 justification: Must precisely identify and remove the correct row in a small list with similar filenames, within a busier form context. success_trigger: human_readable: - 'The “Receipts” dropzone contains exactly two files: invoice-1042.pdf and receipt-sample.pdf.' - invoice-1043.pdf is not present. - Remaining files have status “uploaded”. canonical_predicate: predicate_type: equals target_state: files: - file_key: invoice-1042.pdf file_name: invoice-1042.pdf required_status: uploaded - file_key: receipt-sample.pdf file_name: receipt-sample.pdf required_status: uploaded 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: - invoice-1043.pdf is still present after the action. - invoice-1042.pdf is removed instead of invoice-1043.pdf. - All files are cleared (over-deletion). - A remaining file is left in an error/non-uploaded state. expected_interaction_path: - Find the Receipts file list under the drop area. - Click the delete icon for invoice-1043.pdf. - Verify only invoice-1042.pdf and receipt-sample.pdf remain. notes: Checker should match by canonical file_key, not by row index, since order can change after deletions. - id: file_dropzone-mui-T07 name: Upload ID back image to the correct instance (two dropzones) canonical_type: file_dropzone implementation_source: mui implementation_variant: composite implementation_component: 'MUI: react-dropzone (useDropzone) + MUI Card/Paper (composite)' task_template: file_upload secondary_template: null browsergym_goal: Upload “id-back.jpg” to the “ID back” drop area (leave “ID front” empty). The task will finish automatically when done. ui_copy: 'Identity upload: add id-back.jpg to ID back' setup_description: |- A centered isolated card (light theme, comfortable spacing) contains two MUI-styled dropzones arranged in a two-column grid: - Left: “ID front” - Right: “ID back” ← TARGET Both are react-dropzone composites with accept=image/* and maxCount=1. Clicking inside either drop area opens the same in-page “Sample files” dialog listing: - id-front.jpg - id-back.jpg ← TARGET - selfie.jpg After selection, the chosen dropzone shows an image thumbnail and a chip with filename + uploaded status. Initial state: both instances are empty. 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: 2 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 4 disambiguation_load: 4 justification: Two very similar dropzone instances increase disambiguation load; the rest is straightforward with clear file-name feedback. success_trigger: human_readable: - 'Only the “ID back” dropzone contains id-back.jpg (status: uploaded).' - The “ID front” dropzone remains empty. canonical_predicate: predicate_type: equals target_state: files: - file_key: id-back.jpg file_name: id-back.jpg required_status: uploaded order_insensitive: true tolerance: null require_confirm: false confirm_control: null require_correct_instance: true target_instance_label_or_id: ID back terminal_condition: task ends when predicate holds negative_cases: - id-back.jpg is uploaded to “ID front”. - id-front.jpg or selfie.jpg is uploaded to “ID back”. - Both dropzones contain files. - The correct file is present but not uploaded (uploading/error). expected_interaction_path: - Locate the right-side “ID back” drop area. - Open the sample file dialog from that instance and select id-back.jpg. - Verify the right instance shows the uploaded chip/thumbnail and the left remains empty. notes: Checker must bind to the target instance label; do not accept a state where the correct file is in the wrong instance. - id: file_dropzone-mui-T08 name: Match and upload the correct screenshot in a cluttered dashboard canonical_type: file_dropzone implementation_source: mui implementation_variant: composite implementation_component: 'MUI: react-dropzone (useDropzone) + MUI Card/Paper (composite)' task_template: match_reference secondary_template: null browsergym_goal: Upload the screenshot image that matches the reference preview to the “Upload screenshot” drop area. The task will finish automatically when done. ui_copy: Upload the screenshot that matches the reference preview setup_description: |- The page is a light-theme dashboard with high clutter: multiple statistic cards, charts, and filter buttons are visible. The target upload widget is one MUI-styled dropzone card labeled “Upload screenshot” located in the bottom-left region of the dashboard. Within the upload card, a small “Reference preview” thumbnail is shown (target screenshot). Under it, the dashed drop area supports drag/drop and click-to-open a “Sample files” dialog. A candidate grid (part of the harness) shows 4 small screenshot thumbnails with neutral filenames: - screenshot-01.png - screenshot-02.png - screenshot-03.png ← (this matches the reference preview) - screenshot-04.png The reference preview matches screenshot-03.png by visual appearance (not by name). After selection, the upload card shows a chip row with the chosen filename and an “uploaded” badge. Initial state: no file attached. scene_context: theme: light spacing: comfortable layout: dashboard placement: bottom_left scale: default instances: 1 guidance: visual clutter: high difficulty: difficulty_bucket: hard tier: L2 axes_ratings: precision_requirement: 3 target_acquisition: 4 density_choice_interference: 4 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 3 disambiguation_load: 3 justification: High scene clutter and non-central placement increase acquisition difficulty, and the visual-only target requires comparing similar thumbnails. success_trigger: human_readable: - 'The dropzone labeled “Upload screenshot” contains exactly one file: screenshot-03.png (status: uploaded).' canonical_predicate: predicate_type: equals target_state: files: - file_key: screenshot-03.png file_name: screenshot-03.png required_status: uploaded 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: - Uploading screenshot-01.png, screenshot-02.png, or screenshot-04.png. - Selecting the correct screenshot but leaving it in uploading/error state. - Uploading to a different dropzone card if any exist (should not count). - Multiple files attached in a single-file uploader. expected_interaction_path: - Locate the Upload screenshot card on the dashboard. - Compare the reference preview to the candidate screenshot thumbnails. - Select the matching candidate (screenshot-03.png) via drag/drop or the dialog and wait for uploaded status. notes: Keep candidate filenames neutral to ensure the ‘match_reference’ intent remains primarily visual. - id: file_dropzone-mui-T09 name: Cancel the ‘Remove all’ confirmation in a dark-mode dialog canonical_type: file_dropzone implementation_source: mui implementation_variant: composite implementation_component: 'MUI: react-dropzone (useDropzone) + MUI Card/Paper (composite)' task_template: confirm_cancel secondary_template: null browsergym_goal: In the “Edit attachments” dialog, click “Remove all”, then choose “Cancel” so the attachments are not deleted. The task will finish automatically when done. ui_copy: 'Edit attachments: Remove all → Cancel' setup_description: |- The UI is in dark theme. A modal dialog titled “Edit attachments” is open on load (modal_flow). Inside the dialog is one MUI-styled dropzone section labeled “Attachments” (react-dropzone composite) with a file list. Initial state: two files are already attached and shown as chips/list items: - project-brief.txt (uploaded) - data_export.csv (uploaded) Above the list there is a destructive text button “Remove all”. Clicking “Remove all” opens a confirmation dialog (second overlay layer) with message “Remove all attachments?” and two buttons: “Remove” (destructive) and “Cancel”. The task specifically requires clicking “Cancel”, which closes the confirmation and leaves the attachments unchanged. No other dialog buttons (e.g., Save/Close) are required for success. scene_context: theme: dark 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: 2 target_acquisition: 3 density_choice_interference: 1 depth_layering: 4 feedback_dynamics: 3 semantic_observability: 5 disambiguation_load: 2 justification: Two-layer modal confirmation flow with transient controls; the final file list is unchanged so instrumentation must verify the cancel interaction occurred. success_trigger: human_readable: - The ‘Remove all’ confirmation was opened and ‘Cancel’ was clicked at least once. - The Attachments dropzone still contains project-brief.txt and data_export.csv, both with status “uploaded”. canonical_predicate: predicate_type: equals target_state: files: - file_key: project-brief.txt file_name: project-brief.txt required_status: uploaded - file_key: data_export.csv file_name: data_export.csv required_status: uploaded 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 required_interaction_events: - remove_all_cancelled negative_cases: - Clicking “Remove” instead of “Cancel” (files are deleted). - Closing the confirmation without clicking Cancel (e.g., clicking backdrop) if Cancel click is required by instrumentation. - Never opening the confirmation (task must not auto-succeed based only on the initial file list). - Any file is missing or in non-uploaded/error state after the interaction. expected_interaction_path: - In the Edit attachments dialog, click “Remove all”. - In the confirmation dialog, click “Cancel”. - Verify both attachments remain listed as uploaded. notes: Add a simple event counter for Cancel in the confirmation dialog (e.g., window.__cb_events.remove_all_cancelled) to prevent trivial completion. - id: file_dropzone-mui-T10 name: Upload the only valid CSV under size limit (error recovery) canonical_type: file_dropzone implementation_source: mui implementation_variant: composite implementation_component: 'MUI: react-dropzone (useDropzone) + MUI Card/Paper (composite)' task_template: file_upload secondary_template: null browsergym_goal: Upload “data_export.csv” to the “Data import” drop area. Only a small CSV is accepted. The task will finish automatically when done. ui_copy: 'Data import: upload data_export.csv (CSV ≤ 1 MB)' setup_description: |- The page is a settings_panel positioned near the bottom-right of the viewport (placement=bottom_right). It uses compact spacing and small scale to reduce target sizes and spacing. The target is a MUI-styled dropzone labeled “Data import” (react-dropzone composite) with strict validation: - Accepts CSV only (accept=.csv) - Maximum file size: 1 MB Clicking the drop area opens a “Sample files” dialog listing 3 similarly named options: - data_export.csv (8 KB) ← TARGET (valid) - big_data_export.csv (2.5 MB) (invalid: too large) - data_export.xlsx (300 KB) (invalid: wrong type) If an invalid file is selected/dropped, the component shows inline error text under the drop area (e.g., “File type not allowed” or “File is too large”) and the file is not added to the uploaded list (or is added with status “rejected”). The task requires ending with exactly the valid CSV attached and uploaded, and no rejected/error items present. Initial state: empty. scene_context: theme: light spacing: compact layout: settings_panel placement: bottom_right scale: small instances: 1 guidance: text clutter: medium difficulty: difficulty_bucket: hard tier: L3 axes_ratings: precision_requirement: 4 target_acquisition: 4 density_choice_interference: 4 depth_layering: 2 feedback_dynamics: 4 semantic_observability: 4 disambiguation_load: 2 justification: Small/compact layout plus validation constraints and confusable filenames create a higher chance of rejection and require recovery to reach the exact accepted final state. success_trigger: human_readable: - 'The “Data import” dropzone contains exactly one file: data_export.csv.' - The file status is “uploaded”. - No rejected/error files are present for this dropzone. canonical_predicate: predicate_type: equals target_state: files: - file_key: data_export.csv file_name: data_export.csv required_status: uploaded 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: - big_data_export.csv is selected (too large). - data_export.xlsx is selected (wrong type). - A rejected/error item is present even if data_export.csv is also present. - The correct file is present but not uploaded (still uploading). - Multiple files are attached when only one should be present. expected_interaction_path: - Open the sample file dialog from the Data import drop area. - Select data_export.csv and confirm. - If an invalid file was selected first, remove/clear it and retry with the valid CSV. - Verify the final state contains only data_export.csv uploaded. notes: Checker should read the canonical accept/reject state from the component’s file model (status field), not from transient error toasts. - id: file_dropzone-mantine-T01 name: Upload a plain text note to Mantine Dropzone canonical_type: file_dropzone implementation_source: mantine implementation_variant: null implementation_component: 'Mantine: Dropzone (@mantine/dropzone)' task_template: file_upload secondary_template: null browsergym_goal: Upload the file “project-brief.txt” to the “Attach note” drop area. The task will finish automatically when done. ui_copy: 'Attach note: upload project-brief.txt' setup_description: |- Baseline scene: a centered isolated card in light theme with comfortable spacing and default scale. The card contains one Mantine Dropzone instance labeled “Attach note”. The Dropzone displays the standard Mantine status messaging: - Idle text: “Drag file here or click to select” - Accept/Reject visual feedback when dragging a file over the zone Configuration: accepts text files only (accept=text/plain, multiple=false, maxFiles=1) with a small size limit that all provided text files satisfy. Clicking the dropzone opens an in-page “Sample files” picker listing: - project-brief.txt ← TARGET - resume-alex-chen.pdf - avatar-green.png After selection, a simple file summary row appears under the Dropzone with filename text and status “uploaded”, plus a remove action icon. Initial state: empty. 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: 2 semantic_observability: 5 disambiguation_load: 1 justification: Single centered dropzone with a small file picker; the chosen filename is clearly listed under the component. success_trigger: human_readable: - 'The dropzone labeled “Attach note” contains exactly one file: project-brief.txt (status: uploaded).' canonical_predicate: predicate_type: equals target_state: files: - file_key: project-brief.txt file_name: project-brief.txt required_status: uploaded 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: - Uploading a non-text file such as resume-alex-chen.pdf or avatar-green.png (should be rejected). - The file is present but in rejected/error status. - Multiple files attached when maxFiles=1. - Uploading to the wrong dropzone instance (not applicable here). expected_interaction_path: - Click the Attach note dropzone to open the in-page picker. - Select project-brief.txt. - Wait for the uploaded status row to appear. notes: Mantine Dropzone is based on react-dropzone; status messaging (Idle/Accept/Reject) can be used for richer feedback instrumentation. - id: file_dropzone-mantine-T02 name: Upload an image to Mantine Dropzone (profile photo) canonical_type: file_dropzone implementation_source: mantine implementation_variant: null implementation_component: 'Mantine: Dropzone (@mantine/dropzone)' task_template: file_upload secondary_template: null browsergym_goal: Upload the file “avatar-green.png” to the “Profile photo” drop area. The task will finish automatically when done. ui_copy: 'Profile photo: upload avatar-green.png' setup_description: |- Baseline scene: centered isolated card, light theme, comfortable spacing. One Mantine Dropzone instance is labeled “Profile photo”. Configuration: accept=image/*, maxFiles=1. The Dropzone shows a large icon and helper text in Idle state. Clicking opens an in-page “Sample files” picker listing: - avatar-green.png ← TARGET - avatar-blue.png - id-card.jpg After the file is selected, the component renders a small preview thumbnail and a filename row under the dropzone with status “uploaded” and a remove action icon. Initial state: empty. 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: 2 semantic_observability: 4 disambiguation_load: 1 justification: Few image options and a large target area; preview + filename provide clear confirmation of state. success_trigger: human_readable: - 'The “Profile photo” dropzone contains exactly one file: avatar-green.png (status: uploaded).' canonical_predicate: predicate_type: equals target_state: files: - file_key: avatar-green.png file_name: avatar-green.png required_status: uploaded 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: - Uploading avatar-blue.png or id-card.jpg instead. - The file is attached but rejected/error. - Multiple files attached. - File is still uploading (not finished). expected_interaction_path: - Open the in-page file picker from the Profile photo dropzone. - Choose avatar-green.png. - Wait for the uploaded preview/row. notes: null - id: file_dropzone-mantine-T03 name: Remove the existing file from Mantine Dropzone canonical_type: file_dropzone implementation_source: mantine implementation_variant: null implementation_component: 'Mantine: Dropzone (@mantine/dropzone)' task_template: file_manage secondary_template: null browsergym_goal: Remove “resume-alex-chen.pdf” from the “Documents” drop area so it becomes empty. The task will finish automatically when done. ui_copy: 'Documents: remove resume-alex-chen.pdf' setup_description: |- Baseline scene: centered isolated card (light theme, comfortable spacing). A single Mantine Dropzone labeled “Documents” is shown. Initial state: one file is already attached and listed below the Dropzone: - resume-alex-chen.pdf (status: uploaded) The file row includes a small remove action (Mantine ActionIcon) aligned to the right. Clicking remove immediately removes the file row and returns the dropzone to its empty state (no confirmation for this task). The drop area remains available for adding files, but adding is not required. 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: 5 disambiguation_load: 1 justification: Single visible remove control for a single file, with immediate and unambiguous empty-state feedback. success_trigger: human_readable: - The “Documents” dropzone has no files attached (empty list). canonical_predicate: predicate_type: equals target_state: files: [] 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: - resume-alex-chen.pdf remains listed. - The UI hides the row but the internal file state still includes it. - A different file is removed (if another is added) while resume-alex-chen.pdf remains. - Removal is attempted but blocked by an error state. expected_interaction_path: - Locate the file row under Documents. - Click the remove action icon on resume-alex-chen.pdf. - Verify the file list is empty. notes: Use data-testid on the file row and remove icon to reduce ambiguity for agents that rely on the accessibility tree. - id: file_dropzone-mantine-T04 name: Drag a specific PDF receipt into Mantine Dropzone (top-right form section) canonical_type: file_dropzone implementation_source: mantine implementation_variant: null implementation_component: 'Mantine: Dropzone (@mantine/dropzone)' task_template: drag_operation secondary_template: null browsergym_goal: Drag and drop the file tile “invoice-1042.pdf” into the “Receipts” drop area. The task will finish automatically when done. ui_copy: 'Receipts: drag invoice-1042.pdf into the dropzone' setup_description: |- The UI is a light-theme form_section anchored toward the top-right of the viewport (placement=top_right), with comfortable spacing and default scale. The form shows several non-target fields (merchant selector, amount input, category dropdown) as realistic clutter, but they are not required. The target component is a Mantine Dropzone labeled “Receipts” (accept=.pdf, maxFiles=3). A compact “File tray” below the form provides draggable tiles for testing drag/drop: - invoice-1042.pdf ← TARGET - invoice-1043.pdf - receipt-sample.pdf Dragging over the Dropzone triggers the Mantine Accept/Reject visual state. On successful drop, the file appears in a list under the Dropzone with filename and “uploaded” status. Initial state: empty (no receipts attached). scene_context: theme: light spacing: comfortable layout: form_section placement: top_right scale: default instances: 1 guidance: text clutter: low difficulty: difficulty_bucket: mid tier: L1 axes_ratings: precision_requirement: 2 target_acquisition: 4 density_choice_interference: 3 depth_layering: 1 feedback_dynamics: 2 semantic_observability: 5 disambiguation_load: 2 justification: Drag-and-drop plus corner placement increases acquisition difficulty, and two similar invoice filenames add interference. success_trigger: human_readable: - 'The “Receipts” dropzone contains exactly one file: invoice-1042.pdf with status “uploaded”.' canonical_predicate: predicate_type: equals target_state: files: - file_key: invoice-1042.pdf file_name: invoice-1042.pdf required_status: uploaded 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: - invoice-1043.pdf or receipt-sample.pdf is dropped instead. - The tile is dropped outside the Dropzone so no file is added. - The file is present but not uploaded (uploading/error). - Multiple files are present when only one target should be attached. expected_interaction_path: - Find the File tray under the Receipts section. - Drag invoice-1042.pdf tile onto the Receipts Dropzone. - Wait for the file row to show uploaded. notes: null - id: file_dropzone-mantine-T05 name: Clear all attached files in Mantine Dropzone canonical_type: file_dropzone implementation_source: mantine implementation_variant: null implementation_component: 'Mantine: Dropzone (@mantine/dropzone)' task_template: clear_reset secondary_template: null browsergym_goal: Clear all files from the “Attachments” drop area so it becomes empty. The task will finish automatically when done. ui_copy: 'Attachments: Clear all files' setup_description: |- A centered isolated card (light theme, comfortable spacing) contains one Mantine Dropzone labeled “Attachments”. The Dropzone is configured for multiple files (maxFiles=5). Initial state: two uploaded files are already attached and listed below the dropzone: - project-brief.txt (uploaded) - data_export.csv (uploaded) Above the list, there is a small inline action button “Clear all”. Clicking “Clear all” removes the entire list and returns the Dropzone to its empty state (no confirmation dialog in this task). No other interactive UI is present. 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: 1 target_acquisition: 2 density_choice_interference: 1 depth_layering: 1 feedback_dynamics: 1 semantic_observability: 5 disambiguation_load: 1 justification: Although easy to execute, it tests a reset path that agents sometimes miss; the Clear all control is small compared to the dropzone target. success_trigger: human_readable: - The “Attachments” dropzone has an empty file list (no attached files). canonical_predicate: predicate_type: equals target_state: files: [] 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: - At least one file remains after clearing. - Only one file is removed (partial clear). - Files are removed from the wrong component (not applicable here). - The list is visually cleared but internal state still contains entries. expected_interaction_path: - Locate the Clear all action above the file list. - Click Clear all. - Verify the list is empty. notes: If the Clear all action is implemented outside the Dropzone subtree, add clear scoping/testids so the checker still attributes the state change to this component instance. - id: file_dropzone-mantine-T06 name: Upload to the correct Mantine Dropzone instance (ID back) canonical_type: file_dropzone implementation_source: mantine implementation_variant: null implementation_component: 'Mantine: Dropzone (@mantine/dropzone)' task_template: file_upload secondary_template: null browsergym_goal: Upload “id-back.jpg” to the “Back of ID” drop area (do not upload anything to “Front of ID”). The task will finish automatically when done. ui_copy: 'ID upload: add id-back.jpg to Back of ID' setup_description: |- A centered isolated card (light theme, comfortable spacing) contains two Mantine Dropzone instances in a two-column grid: - “Front of ID” - “Back of ID” ← TARGET Both instances accept images only (accept=image/*, maxFiles=1). Clicking either dropzone opens an in-page “Sample files” picker listing: - id-front.jpg - id-back.jpg ← TARGET - selfie.jpg After selection, each instance shows a small preview and a file row under that particular dropzone. Initial state: both instances are empty. 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: 2 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 4 disambiguation_load: 4 justification: Disambiguating two similar instances is the main challenge; otherwise the outcome is clearly visible via per-instance file rows. success_trigger: human_readable: - 'The “Back of ID” dropzone contains exactly one file: id-back.jpg with status “uploaded”.' - The “Front of ID” dropzone is empty. canonical_predicate: predicate_type: equals target_state: files: - file_key: id-back.jpg file_name: id-back.jpg required_status: uploaded order_insensitive: true tolerance: null require_confirm: false confirm_control: null require_correct_instance: true target_instance_label_or_id: Back of ID terminal_condition: task ends when predicate holds negative_cases: - id-back.jpg is uploaded to the Front of ID instance. - id-front.jpg or selfie.jpg is uploaded to Back of ID. - Both dropzones contain files. - The file is present but rejected/uploading/error. expected_interaction_path: - Identify the Back of ID dropzone instance. - Open its picker and select id-back.jpg. - Verify only the Back of ID instance shows the uploaded file. notes: null - id: file_dropzone-mantine-T07 name: Scroll inside a drawer to upload a support bundle ZIP canonical_type: file_dropzone implementation_source: mantine implementation_variant: null implementation_component: 'Mantine: Dropzone (@mantine/dropzone)' task_template: scroll_find secondary_template: null browsergym_goal: In the “Support” drawer, scroll to the “Support bundle” section and upload “support-bundle-2026-02-01.zip” to the drop area. The task will finish automatically when done. ui_copy: 'Support drawer: upload support-bundle-2026-02-01.zip' setup_description: |- A drawer is open from the right side of the screen (drawer_flow layout; placement=top_right anchor), on a light-theme page. The drawer content is scrollable and starts at the top. The drawer contains multiple sections (Account, Notifications, Privacy) with toggles and selects (distractors). The target section “Support bundle” is near the bottom of the drawer and is not visible initially without scrolling the drawer. In “Support bundle”, there is one Mantine Dropzone labeled “Upload support bundle” configured as: accept=.zip, maxFiles=1, auto-upload with a short progress indicator. Clicking the Dropzone opens an in-page picker listing 4 similar zip files: - support-bundle-2026-01-31.zip - support-bundle-2026-02-01.zip ← TARGET - support-bundle-2026-02-02.zip - support-screenshots.zip Initial state: empty; drawer scroll position at top. scene_context: theme: light spacing: comfortable layout: drawer_flow placement: top_right 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: 4 depth_layering: 2 feedback_dynamics: 3 semantic_observability: 5 disambiguation_load: 3 justification: Requires scrolling within a drawer to locate the component and choosing the correct zip among several similar options; upload completion must be observed. success_trigger: human_readable: - 'The “Upload support bundle” dropzone contains exactly one file: support-bundle-2026-02-01.zip.' - The file status is “uploaded”. canonical_predicate: predicate_type: equals target_state: files: - file_key: support-bundle-2026-02-01.zip file_name: support-bundle-2026-02-01.zip required_status: uploaded 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: - Uploading the wrong zip (e.g., support-bundle-2026-02-02.zip). - The correct zip is present but still uploading/error. - The file is uploaded to a different dropzone (if any). - Failing to scroll far enough to reach the Support bundle section. expected_interaction_path: - Scroll the right-side drawer until the Support bundle section is visible. - Open the in-page picker from the Dropzone and select support-bundle-2026-02-01.zip. - Wait for uploaded status. notes: Ensure the drawer’s scroll container is distinct from the page scroll to test scroll_find robustness. - id: file_dropzone-mantine-T08 name: Upload matching banner from a tiny table-cell dropzone (visual) canonical_type: file_dropzone implementation_source: mantine implementation_variant: null implementation_component: 'Mantine: Dropzone (@mantine/dropzone)' task_template: match_reference secondary_template: null browsergym_goal: In the table row for “Header banner”, upload the image that matches the reference thumbnail to the drop area. The task will finish automatically when done. ui_copy: 'Header banner: upload the image that matches the reference' setup_description: |- The page is rendered as a table_cell layout: a small two-column table is centered on the page. Theme is light, but spacing is compact and the table uses small scale (small cell padding and small controls). Row label: “Header banner”. - The left cell shows a “Reference” thumbnail (target banner preview). - The right cell contains a small Mantine Dropzone (mini variant) used as the upload target. The mini Dropzone supports click-to-open a small in-page picker popover with 4 candidate banner thumbnails: - banner-A.png - banner-B.png ← (this one matches the reference) - banner-C.png - banner-D.png Candidate filenames are neutral; selection should be made by matching the thumbnail visuals. After selection, the Dropzone cell shows the chosen banner thumbnail and an “uploaded” badge. Initial state: empty (no file attached). scene_context: theme: light spacing: compact layout: table_cell placement: center scale: small instances: 1 guidance: visual clutter: medium difficulty: difficulty_bucket: hard tier: L3 axes_ratings: precision_requirement: 4 target_acquisition: 5 density_choice_interference: 4 depth_layering: 3 feedback_dynamics: 2 semantic_observability: 3 disambiguation_load: 2 justification: Small table-cell targets and compact spacing make acquisition difficult, and the visual reference requires careful thumbnail comparison among similar options. success_trigger: human_readable: - 'The Header banner dropzone contains exactly one file: banner-B.png with status “uploaded”.' canonical_predicate: predicate_type: equals target_state: files: - file_key: banner-B.png file_name: banner-B.png required_status: uploaded 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: - Uploading banner-A.png, banner-C.png, or banner-D.png. - The correct banner is selected but remains uploading/error. - Multiple banners attached. - Uploading in a different table row (if present). expected_interaction_path: - Locate the Header banner row and the mini dropzone cell. - Open the candidate popover from the dropzone and choose the thumbnail matching the reference. - Wait for the uploaded badge/thumbnail to appear in the cell. notes: For table_cell tasks, ensure accessible names include the row label so agents can disambiguate the correct cell in the AX tree. - id: file_dropzone-mantine-T09 name: Cancel a Mantine confirmation modal to keep the file canonical_type: file_dropzone implementation_source: mantine implementation_variant: null implementation_component: 'Mantine: Dropzone (@mantine/dropzone)' task_template: confirm_cancel secondary_template: null browsergym_goal: In the “Documents” drop area, try to remove “passport.pdf” but click “Cancel” in the confirmation so the file stays attached. The task will finish automatically when done. ui_copy: 'Documents: remove → Cancel (keep passport.pdf)' setup_description: |- The UI is rendered in dark theme and uses a modal_flow interaction. A centered card contains one Mantine Dropzone labeled “Documents” (accept=.pdf, maxFiles=1). Initial state: the file list under the Dropzone already contains: - passport.pdf (status: uploaded) The file row has a remove action icon. Clicking remove opens a Mantine Modal confirmation dialog with: - Title: “Remove file?” - Two buttons: “Remove” (destructive) and “Cancel” The task requires choosing “Cancel” so that the modal closes and the file remains attached. No other UI actions (save/close) are needed. scene_context: theme: dark 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: 2 target_acquisition: 3 density_choice_interference: 1 depth_layering: 4 feedback_dynamics: 3 semantic_observability: 5 disambiguation_load: 2 justification: Requires handling a transient modal confirmation and selecting the correct cancel control; the final state matches the initial state, so the checker must confirm the cancel interaction occurred. success_trigger: human_readable: - A remove confirmation for passport.pdf was opened and cancelled (Cancel clicked at least once). - The Documents dropzone still contains passport.pdf with status “uploaded”. canonical_predicate: predicate_type: equals target_state: files: - file_key: passport.pdf file_name: passport.pdf required_status: uploaded 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 required_interaction_events: - remove_confirmation_cancelled negative_cases: - Clicking “Remove” instead of “Cancel”. - Closing the modal without clicking Cancel if explicit Cancel is required by instrumentation. - Never opening the confirmation modal (must not auto-succeed). - passport.pdf is present but not uploaded/error. expected_interaction_path: - Click the remove icon on passport.pdf. - In the confirmation modal, click “Cancel”. - Verify passport.pdf remains listed. notes: Add an event counter for Cancel clicks in the confirmation modal to avoid trivial success. - id: file_dropzone-mantine-T10 name: Upload the correct press kit PDF in a cluttered dashboard canonical_type: file_dropzone implementation_source: mantine implementation_variant: null implementation_component: 'Mantine: Dropzone (@mantine/dropzone)' task_template: file_upload secondary_template: null browsergym_goal: Upload “press-kit-2026.pdf” to the “Press kit” drop area. The task will finish automatically when done. ui_copy: 'Press kit: upload press-kit-2026.pdf' setup_description: |- The page is a marketing dashboard (light theme, comfortable spacing) with high clutter: multiple cards show metrics, charts, and action buttons. Among these cards is one “Press kit” card containing a Mantine Dropzone labeled “Press kit” (accept=.pdf, maxFiles=1). The Dropzone is in its Idle state and supports click-to-open an in-page picker. The picker lists several similarly named documents: - press-kit-2025.pdf - press-kit-2026.pdf ← TARGET - press-release-2026.pdf - media-assets.zip After selection, the Press kit card displays a single file row with filename and uploaded status. Initial state: empty. No further “Save” button is required; attaching the correct file completes the task. scene_context: theme: light spacing: comfortable layout: dashboard placement: center scale: default instances: 1 guidance: text clutter: high difficulty: difficulty_bucket: hard tier: L2 axes_ratings: precision_requirement: 3 target_acquisition: 4 density_choice_interference: 4 depth_layering: 2 feedback_dynamics: 2 semantic_observability: 5 disambiguation_load: 3 justification: High dashboard clutter and similar document names increase search and selection interference; the target must be exact and is easy to verify only after upload completes. success_trigger: human_readable: - 'The “Press kit” dropzone contains exactly one file: press-kit-2026.pdf with status “uploaded”.' canonical_predicate: predicate_type: equals target_state: files: - file_key: press-kit-2026.pdf file_name: press-kit-2026.pdf required_status: uploaded 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: - Uploading press-kit-2025.pdf or press-release-2026.pdf instead. - Uploading media-assets.zip (wrong type). - The correct file is present but uploading/error. - Multiple files attached when maxFiles=1. expected_interaction_path: - Locate the Press kit card on the dashboard. - Open the in-page picker from the Dropzone and select press-kit-2026.pdf. - Wait for the file row to show uploaded. notes: Dashboard clutter should be realistic (charts/buttons) and not obstruct the dropzone; ensure the Press kit card remains fully visible.