Data Model

核心功能：提供6阶段标准化数据分析建模流程，覆盖业务粒度定义、一致性维度、事实度量分类、SCD类型选择、键约束与性能优化，搭配最终检查清单。典型场景：数仓/湖仓/BI层设计、报表数据重复问题排查、维度模型重构。输入为业务需求与工具选型（如dbt、BigQuery），输出为规范的星型/雪花型数据模型，保障分析准确性与性能。

clawkk

热度 63.28

最近更新 2026-05-11 17:37:15

分类：数据依赖：纯提示词应用场景：数据处理 / 报告撰写 / 模板

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add data-model

Metadata

name: data-model
description: Deep data modeling workflow—grain, facts and dimensions, keys, slowly changing dimensions, normalization trade-offs, and analytics query patterns. Use when designing warehouse/analytics models or reviewing star/snowflake schemas.

Data Model

Analytics models succeed when grain is explicit, keys are stable, and slowly changing dimensions are chosen deliberately—not “star schema by default.”

When to Offer This Workflow

Trigger conditions:

Designing a warehouse, lakehouse, or BI layer
Confusion on one row per what; duplicate counts in reports
Refactoring dimensional models for performance or clarity

Initial offer:

Use six stages: (1) business questions & grain, (2) conformed dimensions, (3) facts & measures, (4) dimensions & SCD types, (5) keys & integrity, (6) performance & evolution). Confirm tooling (dbt, dimensional DW, BigQuery, etc.).

Stage 1: Business Questions & Grain

Goal: Grain = the atomic row: e.g., “one line item per order per day” not “sort of per order.”

Practices

List questions the model must answer; derive grain from smallest needed detail

Exit condition: One sentence grain per fact table.

Stage 2: Conformed Dimensions

Goal: Same customer/product definitions across facts—shared dimension tables or SCD policy aligned.

Stage 3: Facts & Measures

Goal: Additive vs semi-additive vs non-additive measures documented (balances, distinct counts).

Practices

Degenerate dimensions vs junk dimensions—avoid wide fact sprawl without reason

Stage 4: Dimensions & SCD Types

Goal: SCD1 overwrite vs SCD2 history with valid_from/valid_to vs SCD3 limited history—match compliance and reporting needs.

Stage 5: Keys & Integrity

Goal: Surrogate keys in facts; natural keys preserved as attributes; referential integrity strategy in the warehouse layer.

Stage 6: Performance & Evolution

Goal: Partition and cluster keys for large facts; late-arriving facts policy; version dims when schema evolves.

Final Review Checklist

Grain explicit per fact table
Conformed dimensions planned
Measure additivity documented
SCD strategy per critical dimension
Keys and late-arriving data handled

Tips for Effective Guidance

Fan traps and chasm traps in BI—flag when joining across facts incorrectly.
Snapshot fact tables for point-in-time balances vs transaction facts.

Handling Deviations

Event-only pipelines: still model curated dimensions for analysis, not only raw JSON.