Generate fresh, visualisable data questions from raw datasets; reject stale prompts before charting.
Generate fresh, visualisable analysis questions from a raw tabular dataset. Use when Codex is given a CSV/XLSX/Parquet/database extract and asked what to ask, what to explore, what charts to make, what visualisation workshop prompts to use, or what data stories might be interesting; especially for Karthik-style exploratory analysis where obvious/stale questions should be filtered out before charting.
Orchestrate dataset-to-visual-story work: plan analysis, run it, choose visuals, style, critique, and iterate.
End-to-end analytical data visualization workflow for Karthik. Use when the user points Codex to a dataset and gives a loose exploratory question, possible hypothesis, story idea, or desired audience, and wants Codex to plan the analysis, run the analysis, find the defensible story, choose the best visual representation, make chart outputs in Karthik's design aesthetic, critique the result, and iterate until the visual story is good enough to use.
Choose charts for data stories, including S-curves, knee-bends, inflections, local peaks, and misleading/decorative forms.
Choose the right visualization for a dataset plus analytical question, hypothesis, data story, or management problem. Use when recommending, designing, critiquing, or implementing chart choices before plotting; especially for Karthik-style explanatory analytics, Mint-style data stories, time-series shape annotation (knee-bends, inflection points, local maxima/minima, temporary peaks), S-curves/adoption/diffusion patterns, Babbage/management decks, election/sports/payment/geography/risk visuals, or choosing between lines, bars, scatter, maps, distributions, small multiples, scorecards, waterfalls, and tables.
Turn data questions into Karthik-style analysis contracts with definitions, denominators, comparisons, metrics, caveats, and falsifiers.
Turn a natural-language analytical question into Karthik-style analysis contract before coding, charting, or prose. Use when a user asks a data question, blog/data-story question, exploratory analysis question, or asks to plan an analysis; especially when the answer needs explicit operational definitions, unit of analysis, denominator, comparison, metric, caveats, and falsification conditions rather than generic LLM priors.