Unit Conversions in Analytical Models
How to store original units, normalized values, and conversion provenance without losing numeric meaning.
The ColdStart Data blog
Notes on data systems.
A blog about data engineering, analytics, and the details that make data reliable. Explore practical guides, field notes, and lessons on everything from a first warehouse to AI-ready data.
From the blog
Recent articles
Currency Conversion Dates in Reporting
How to tell whether a number should use the transaction date rate or the reporting date rate before it reaches a dashboard.
Exchange-rate direction and inversion
A practical guide to preventing multiplication-versus-division mistakes in currency normalization.
Percentage Points Versus Percentage Changes
A worked comparison for avoiding absolute-versus-relative change errors in metrics, models, and dashboards.
Approximate Distinct Counts and Error Bounds
How to explain estimated unique counts honestly, set acceptable error, and avoid presenting probabilistic metrics as exact facts.
Variance Aggregation Across Partitions
How to combine group statistics correctly without averaging incompatible variances.
Topics
Explore the archive
Modern Data Stack
Practical guidance for choosing and arranging ingestion, warehouse, modeling, orchestration, and BI tools.
Data Modeling
How to shape raw data into documented tables, shared definitions, and durable business concepts.
Dashboard Trust
Methods for making dashboards credible, explainable, and consistent with the way the business operates.
Migration
Move from spreadsheets, legacy databases, or tangled scripts into a cleaner stack without breaking reporting.
Automation
Use scheduling, retries, alerts, and runbooks to make routine data work quiet and dependable.
AI-Ready Data
Prepare clean, governed, contextual data so AI systems have something trustworthy to use.
Reading tracks
Read a little deeper
Collections of articles for exploring a subject from the foundations onward.
Data Foundations
Build the warehouse, naming conventions, models, and source-of-truth habits that make every later data project easier.
Pipeline Reliability
Turn fragile scripts and silent failures into monitored, observable, recoverable data movement.
Analytics & Metrics
Design metrics, dashboards, and business definitions that match reality and survive operational pressure.
AI-Ready Data
Prepare governed, documented, well-modeled data before adding AI workflows on top of the business.