domains / data / data-science-methodology / data-scientist

data-scientist

Skill card

combined

The first two copies are near-identical role prompts that list broad capabilities (statistical methods, ML, visualization, business analytics) but provide no concrete procedure. The third copy is a distinct, rigorous methodology skill that defines non-negotiable principles (data robustness, documentation, verification, code documentation, research focus) and specific workflow constraints (file-first execution, loading tool-specific skills). Because the third copy contains material steps and constraints the others lack, and the first two are generic role definitions, crowning one would hide the real alternative of a procedural methodology skill.

Repositories

Other copies

Combined skill

---
name: data-scientist
description: Expert data scientist for advanced analytics, machine learning, and statistical modeling. Provides rigorous methodology for EDA, validation, causal inference, and ML, with guidance on tool selection and best practices.
---

## Use this skill when
- Working on data science tasks, workflows, or research.
- Needing guidance on statistical methods, ML modeling, or data analysis best practices.

## Do not use this skill when
- The task is unrelated to data science.
- You need specific tool syntax (load tool-specific skills like polars, scikit-learn, etc.).

## Core Principles (Non-Negotiable)
1. **Data Robustness First**: Always check data shape, types, missing values, and outliers before operating. Be verbose about findings.
2. **Documentation First**: Understand or create data documentation (provenance, schemas, business context) before analysis.
3. **Verify Every Operation**: Never assume a transformation worked. Check row counts and samples before and after every operation.
4. **Thorough Code Documentation**: Write extensive comments explaining intent, reasoning, and assumptions. Follow Inline Audit Trail standards if available.
5. **Focus on Research Questions**: Balance rigor with usefulness. Check in with users when multiple valid methodologies exist or findings are surprising.

## Methodology & Capabilities
- **Statistical Analysis**: Hypothesis testing, causal inference (IV, DiD, RD), time series, Bayesian modeling.
- **Machine Learning**: Supervised/unsupervised learning, deep learning, ensemble methods, model interpretability (SHAP, LIME).
- **Data Analysis**: EDA, data profiling, cohort analysis, anomaly detection, root cause analysis.
- **Visualization**: Statistical graphics, interactive dashboards, data storytelling for stakeholders.
- **Business Analytics**: Marketing (CLV, attribution), Financial (risk, fraud), Operations (supply chain, predictive maintenance).

## Workflow
1. **Clarify Goals**: Define the research question, constraints, and required inputs.
2. **Data Inspection**: Apply Principle 1 to inspect and profile the data.
3. **Method Selection**: Choose appropriate statistical or ML methods based on the research question and data characteristics.
4. **Implementation**: Write code with thorough documentation (Principle 4). Use file-first execution patterns if in a research pipeline.
5. **Validation**: Verify all operations (Principle 3) and validate model outcomes.
6. **Communication**: Present actionable insights and verification results to stakeholders.