twin-experiment export-plan crashes (AttributeError) when context_questions is a JSON list

[johnjosephhorton]

Summary

zwill twin-experiment export-plan throws an unhandled AttributeError when the plan's context_questions is a JSON list instead of a comma-separated string. The sibling model field is a list and works fine, so the type contract is inconsistent between adjacent fields, and there is no plan validation to produce a clear error — just a raw traceback.

Repro

Plan defaults with context_questions as a list:

{
  "defaults": {
    "context_questions": ["AGEFaixas", "GENDER", "SEL", "REGION", "Q1"],
    "model": ["openai:gpt-5.5", "google:gemini-2.5-pro"]
  }
}

zwill twin-experiment export-plan --path plan.json

Traceback:

File ".../zwill/twin_jobs.py", line 492, in build_edsl_digital_twin_job_dict
    context_question_names = deps.selected_question_names(context_args, questions)
File ".../zwill/edsl_integration.py", line 145, in selected_question_names
    selected.extend(name.strip() for name in args.questions.split(",") if name.strip())
AttributeError: 'list' object has no attribute 'split'

Fix: use a comma-separated string ("AGEFaixas,GENDER,SEL,REGION,Q1") — but nothing in the scaffold or docs signals this, and model accepting a list sets the opposite expectation.

Suggested fixes

• Accept both a list and a comma-separated string for context_questions (and normalize other question-list fields consistently — heldout_questions, exclude_question, etc.).
• Validate the plan up front and emit a structured error (code: invalid_plan_field) instead of letting a .split traceback escape.
• Related: there is no way to lint/validate a hand-edited plan before approve/export-plan (plan-status requires --survey/--plan-id and only finds records post-export). A twin-experiment validate --path plan.json would catch this class of error early. (Filed separately.)