Analyze Survey AI Preview

Add the staged AI People Science summary experience to the shared analyze-survey report pipeline. Do not duplicate the analysis runner, report builder, golden report, schemas, or source index.

Isolation

• Use a dedicated output directory whose name clearly includes ai-preview.
• Never add summaries to a standard analyze-survey output unless the user explicitly asks to convert that output into a preview.
• Keep respondent-level data local and outside the share ZIP.

Workflow

1. If no completed analysis exists, follow ../analyze-survey/SKILL.md and run scripts/analyze_survey_export.py with --summary-mode off.
2. Follow ../analyze-survey/references/interpretation.md to establish QA readiness and apply the interpretation process and guardrails.
3. Inspect people-science-summary-context.json and the AI-summary modes in ../analyze-survey/references/interactive-report-contract.md.
4. Use people-science-knowledge-vault to retrieve relevant externally published article bodies.
5. Write people-science-summaries.json using ../analyze-survey/references/people-science-summaries.schema.json. Include all six tabs.
6. Rebuild the report:

python scripts/build_interactive_report.py \
  --config <output-directory>/_input/analysis-config.json \
  --output-dir <output-directory> \
  --summary-mode required

7. Confirm all six narrative objects are embedded, the manifest records summary_mode as required, and the share ZIP contains the summary file. Tab-level AI cards remain hidden as specified in the report contract.
8. Open the HTML report before finishing.

Summary contract

Use the shared report contract and narrative schema rather than maintaining a second summary specification here.

Promotion

This skill is a staging entry point. When the preview is approved, promote the experience by changing the default summary mode in the shared pipeline and folding these orchestration instructions into analyze-survey; do not replace the pipeline with copied code.