CAD-1K Open v2 - 1,018.1229 Hours

509 end-to-end, single-display Windows CAD task recordings across seven CAD software families.

Each task contains:

• task_desc.json - task prompt, application, reference-input paths, and expected deliverables
• input_files/ - reference inputs named input.ext or input_N.ext
• output_files/ - submitted CAD deliverables and supplemental outputs named output.ext or output_N.ext
• rubrics.json - task-specific evaluation criteria
• task_overview.pdf - review document combining the task, rubrics, and current visual references
• clip.mp4 - screen recording video
• frame_events.json - events synchronized to video frames
• metadata.json - duration, frame, event, category, and platform metadata
• narration.json - frame-level task narration

Dataset Stats

509 tasks · 1,018.1229 hours · 5,455 public task objects · 207.18 GiB

Durations below use metadata.json.total_duration_ms.

Task Duration

Min: 12.2 min · Max: 812.2 min

Category Distribution

All tasks are in the cad metadata category. The software-level distribution is:

Task Domain Distribution

Domains come from task_desc.json.domain.

Platform Distribution

Artifact Coverage

Every task has all seven required root files, at least one reference input, and at least one submitted output.

Reference Input Formats

AutoCAD visual inputs are derived from the human gold drawings. Blank-start tasks expose the derived visual reference, while edit-existing tasks retain the supplied native starting file alongside the visual reference. The current task_overview.pdf files were regenerated from these inputs.

Submitted Output Formats

Task Layout

{software}/{task_id}/
├── task_desc.json
├── rubrics.json
├── task_overview.pdf
├── clip.mp4
├── frame_events.json
├── metadata.json
├── narration.json
├── input_files/
│   └── input[_N].ext
└── output_files/
    └── output[_N].ext

Pipeline backups and publication receipts live under the private _runs/ prefix and are not part of the public task corpus.