Data Quality Analyst Pro

The Problem: The 'Looks Right' Trap in Data Analysis

In daily work, we often deal with multiple Excel sheets or database query outputs. Directly aggregating, comparing, and visualizing this data can lead to the 'Looks Right' trap: aligning summary columns incorrectly, mixing parent-child levels, using inconsistent metrics, or inheriting arithmetic errors. Insights built on such flawed foundations can mislead decisions with serious consequences.

Core Method: An Embedded 6-Step Quality Gate

The skill's core value isn't flashy analysis, but establishing a repeatable, error-preventing workflow. It mandates that any task involving 'multi-source extraction → computing derived metrics → generating reports/charts' must follow a 6-step quality gate (sourced from the data-quality-gate methodology):

1. Source Data Review: Understand the structure before touching numbers. Identify headers, total rows, and potential traps (e.g., misaligned summary columns, parent-child hierarchies).
2. Create a Lineage Map: For every number, document its data lineage: file + sheet + row number + extraction logic. This is the foundation of traceability.
3. Extract Only Once: Pull raw values into an intermediate structure (like JSON) in one go. All subsequent calculations derive from this intermediate data, eliminating errors from manual re-entry.
4. Self-Consistency Check: Verify logical consistency between intermediate figures (e.g., do percentages sum to 100%? Is the basis for YoY/QoQ calculations uniform?).
5. Three-Way Reconciliation: Before final output, perform a line-by-line comparison: Source File ↔ Intermediate Data ↔ Final Output. Any discrepancy >1% requires immediate investigation.
6. Delivery Lockdown: The final deliverable must include the data lineage and lineage map, making the conclusion fully traceable.

Key Capabilities & Workflow

Beyond the quality gate, the skill provides a structured analytical framework to handle diverse scenarios:

• 10-Scenario Routing: Based on your request (e.g., 'analyze sales data', 'review financial report', 'HR metrics'), it automatically selects the most appropriate analysis blueprint. Each scenario defines key metrics, red lines (e.g., 'no investment advice', 'must anonymize personal data'), and detailed procedures.
• Six-Stage Analysis Method: The standard analytical path follows Description → Comparison → Trend → Anomaly → Causality → Prediction, ensuring a logical and rigorous progression.
• Fast & Slow Lane Diagnosis: For data questions, it offers a 'fast lane' for quick format/calculation checks and a 'slow lane' with a 6-question root cause analysis, suitable for different urgency levels.
• Hard Boundaries & Safety Rails: The skill explicitly states it does not auto-publish, auto-overwrite source files, fabricate data, or provide financial advice. All critical actions (like overwriting files or sending externally) require your explicit confirmation.

Applicable Boundaries & Collaboration

It's crucial to understand this skill's strength is in 'how to rigorously analyze data once it's in hand,' not in data collection or automated preprocessing. It defines clear boundaries for collaboration:

• Data Collection & Cleaning: Handled by office-auto-pro (external skill) or similar.
• Analytical Insight & Visualization: The core domain of data-analyst-pro.
• Report Polishing: Can be passed to humanizer-pro for language refinement.
• Presentation Creation: Analytical conclusions can be formatted into slides by ppt-pro.

The Fundamental Principle: AI-generated analysis is a tool and a draft. The human author bears ultimate responsibility for the data's authenticity, the appropriateness of statistical methods, and the interpretation of results. In academic contexts, strict adherence to data transparency, complete reporting, and AI-content annotation is mandatory.