Case study pipeline not mapped yet
The benchmark pipeline at /arisekit-benchmark-tasks/
covers the full 10-phase, 117-task design surface for an entire
benchmark. This case-study companion will mirror that structure
at item granularity: case stem capture, expected-answer
specification, rubric authoring, judge calibration, and the QC
loop a single item passes through before it joins a dataset.
When the task list lands, it should plug into the same
scripts/sync_pipeline_to_demo.py pattern the
benchmark pipeline uses, sourced from a new
docs/case-study-pipeline.md.