additional_features
["A consistent A0–A7 workflow (structure → tests/hotspots → visuals → narrative)", "Preprocessing rules to sanity-check totals/100%, handle DK/NR/Other labels, and add small-N caveats", "Matplotlib single-figure policy (no explicit colors by default) with Korean font support (NotoSansKR)"]
example_commands
["Summarize the main differences using row percentages, and list the top standardized-residual cells.", "Write a report paragraph including χ² and Cramér’s V. Use neutral wording for significance.", "There are many expected counts < 5—suggest up to three category-collapsing options with pros/cons.", "Cross age groups with a 1–5 ordinal scale and also run a linear-by-linear (trend) association check."]
gpt_id
g-686ce5c0f460819185068b8c9a1bbe18
ideal_use_cases
["Comparing response proportions across segments (gender/age/region, etc.) and summarizing the gaps", "Reporting χ² significance, Cramér’s V, and top cells by standardized residuals in one coherent output", "Flagging assumption risks (e.g., expected count < 5) and proposing recoding/collapsing options", "Producing report-style commentary with explicit % vs % and +/− percentage-point differences"]
limitations
["With very small samples or many sparse cells, χ² assumptions may be weak (consider exact tests or collapsing).", "If the percent base (row% vs column%) is unclear, interpretations can be misleading; table meta is required.", "Multi-select (checkbox) tables are non-mutually-exclusive; avoid direct comparisons to single-select crosstabs.", "By default, it does not pull external web evidence; it interprets only the provided tables/text."]
target_users
["Survey researchers and quantitative analysts", "Marketing/brand insights teams", "Policy/public-sector report writers", "Graduate students and research assistants"]