Method

The analysis uses a reproducible pipeline:

1. Fetch public Google Sheet responses into ignored local raw data.
2. Redact optional contact information before writing processed response tables.
3. Generate provisional qualitative codes from an editable seed codebook.
4. Write coded excerpts, uncoded excerpts, candidate terms, and aggregate counts for human review.
5. Summarize each free-text column with LiteLLM, storing Markdown and JSON outputs.
6. Copy only aggregate and redacted artifacts into this static site.

Important Caveats

The survey is not a statistically representative sample of all AMSAT Status Page visitors. People who care enough to answer are likely more engaged than casual visitors. Theme counts are not mutually exclusive, and LLM summaries are interpretive aids rather than ground truth.

Source Artifacts

• data/processed/theme_counts.csv
• data/processed/response_theme_codes.csv
• data/processed/review_excerpts_uncoded.csv
• reports/llm_summaries/*.json

Raw survey exports and .env credentials are intentionally excluded from git.