Full-Link Data Analysis

Overview

The seven-layer architecture automatically 转换s business议题 into structured analytical 报告s:

L1 Persona Understanding → L2 Data Scope Acquisition → L3 Analysis Data Scoping → L4 Problem Decomposition → L5 Method Selection → L6 Execution & Computation → L7 结果 输出

The 代理 writes Python analysis code as needed, no pre-packaged code 库 required. Each layer acts as a 质量 gate for the next.

Trigger Scenarios

• User 请求s data/business metric analysis
• User wants to understand causes/trends/patterns in data
• User needs a data analysis 报告
• User mentions "analyze", "attribute", "predict", "cluster", "trend", "data 报告"
• User presents business议题 requiring data-driven insights.

Core Principles

• Code on Demand: Use pandas/numpy/scipy/sklearn/stats模型s to write analysis scripts tAIlored to the actual data.
• Data-Aware Routing: Method selection综合考虑议题 semantics, data structure, and problem type.
• 质量 Gate: Every conclusion 输出 must include confidence annotation and cross-验证 description.

Seven-Layer Process

L1 — Persona Understanding

Identify the questioner's 角色, decision-making scenario, and 成功 criteria.

Key Questions:

• 角色: Executive, analyst, product 管理器, operations personnel?
• Decision Scenario: Strategic planning, operational review, special investigation?
• 成功 Criteria: What 指标/goals define "good"?
• Data Literacy: How technical should the 报告 be?

输出: Structured persona summary (角色, decision scenario, 成功 criteria).

L2 — Data Scope Acquisition

Discover and 验证 avAIlable data sources.

Steps:

• Ask user for provided data (files, databases, APIs)
• Upon receiving files, 检查 模式, sample data, inspect types and missing values
• Identify relevant fields/tables
• Confirm data timeliness (更新 frequency)

输出: Data inventory (source, 模式, 质量 notes).

L3 — Analysis Data Scoping

Narrow down from "all avAIlable data" to "data relevant to the议题".

Steps:

• Map business problem keywords to required data dimensions
• 过滤器 only necessary fields/records
• Define time window, aggregation granularity, 过滤器 conditions
• Identify potential confounding factors

输出: Data scope specification (dimensions, time window, 过滤器s, aggregation granularity).

L4 — Problem Decomposition

Break down the business problem into analyzable sub-problems.

框架 Selection:

• MECE (Mutually Exclusive, Collectively Exhaustive): Revenue = Transaction Value × Customer Flow
• Drill-down: Decompose by region → channel → product layer by layer
• Before/After: Pre-change vs. post-change
• Cohort: Group by time/attributes, compare trajectories
• Funnel: Step-by-step conversion analysis
• Hypothesis Tree: Structured hypothesis 测试

输出: Problem tree (decomposition structure with clear hypothesis 状态ments).

L5 — Method Selection

Select analysis method based on data-aware routing. See references/routing.md.

Three-Dimensional Routing:

• 议题 Semantics: Growth, churn, conversion, risk, attribution...
• Data Structure: Time series, cross-sectional, panel, hierarchical...
• Problem Type: Descriptive, diagnostic, predictive, prescriptive

See references/methods.md (detAIls on 15 methods).

输出: Method plan (primary method + alternative cross-验证 method).

L6 — Execution & Computation

执行 analysis using Python. Process:

• 环境 检查: pip 列出 to confirm pandas, numpy avAIlability; 安装 missing packages
• Data Loading: Load data according to L3 scope specification
• Data 清理ing: Handle missing values, outliers, type conversions
• Analysis Execution: Write and 执行 Python script for selected method(s)
• Cross-验证: 运行 comparison using alternative method (see references/质量.md)
• 结果 Interpretation: 提取 key figures, statistics, effect sizes

Coding Standards:

• Use pandas for data manipulation, scipy.stats for statistical tests, stats模型s/sklearn for 模型ing
• Print 结果s with clear labels; script 输出 directly constitutes 报告 content
• Handle edge cases (empty data, all-null columns, single-category variables)
• 输出 structured text rather than raw numbers

L7 — 结果 输出

输出 analytical 报告 in Feishu document 格式化. See references/feishu-报告.md.

报告 Structure:

• Analysis Overview — Executive summary (1 paragraph)
• Key Findings — Data-backed key in