What are the four properties of a data warehouse in Inmon's definition?
Subject-oriented, integrated, time-variant, nonvolatile: four adjectives, one architecture.
W. H. Inmon · Building the Data Warehouse
What does data independence mean in Codd's model?
Every attribute depends on the key, the whole key, and nothing but the key.
E. F. Codd · A Relational Model of Data for Large Shared Data Banks
Why not run reports directly on the production database?
OLTP touches whole rows a few at a time; OLAP touches a few columns billions of times.
Michael Stonebraker, Uğur Çetintemel · “One Size Fits All”: An Idea Whose Time Has Come and Gone
What are the four steps of dimensional design, in order?
The grain is a sentence: "one row = one receipt line, in one store, on one day".
Ralph Kimball, Margy Ross · The Data Warehouse Toolkit
What do the hub, the link and the satellite each hold in Data Vault?
Hub, link, satellite: keys, relations, attributes with history — three kinds of tables, one auditable warehouse.
Dan Linstedt · Data Vault Series 1 — Data Vault Overview
What does a lakehouse add on top of a data lake?
Bronze is the evidence, silver is the cleaned truth, gold is the answer.
Michael Armbrust, Ali Ghodsi, Reynold Xin, Matei Zaharia · Lakehouse: A New Generation of Open Platforms that Unify Data Warehousing and Advanced Analytics
What are the four principles of the data mesh?
A data product has an owner, a schema, an SLA and an address — otherwise it is just a table.
Zhamak Dehghani · Data Mesh Principles and Logical Architecture
Why is column storage faster for analytics?
Columns compress well because neighbours look alike; rows do not.
Michael Stonebraker et al. · C-Store: A Column-oriented DBMS
What advantage does CDC have over periodic extraction with queries?
A periodic SELECT does not see deletions; the transaction log sees every one.
Jay Kreps · The Log: What every software engineer should know about real-time data's unifying abstraction
How do you correct an error in an append-only system?
A correction is a new row, not an eraser.
Pat Helland · Immutability Changes Everything
Why does SCD type 2 require a surrogate key?
Type 1 overwrites, type 2 adds a row with a period, type 3 keeps a column with the previous value.
Ralph Kimball, Margy Ross · The Data Warehouse Toolkit
What do backward and forward compatibility mean for a schema?
Add, backfill, switch, only then drop — never change in place.
Martin Kleppmann · Designing Data-Intensive Applications
What makes a load job idempotent?
A rerun for the same window must give the same result — that is idempotency.
Joe Reis, Matt Housley · Fundamentals of Data Engineering
Name four dimensions of data quality.
A test at intake costs one line; the same error found in the report costs a week.
W. Edwards Deming · Out of the Crisis
What is lineage (provenance) for in a warehouse?
Every figure in a report must be able to say which rows and which file it came from.
Charles Babbage · Passages from the Life of a Philosopher
What is leakage in a training set?
A feature is computed "as it was at the time of the event", not "as it is now".
Alon Halevy, Peter Norvig, Fernando Pereira · The Unreasonable Effectiveness of Data
Why can't you compare two embedding vectors with equality?
The embedding model's version is part of the schema: change the model, recompute the column.
John Rupert Firth · A synopsis of linguistic theory 1930–1955
What is the difference between pseudonymisation and anonymisation?
Pseudonymised still means personal data — only the key lives in another room.
Parlamentul European și Consiliul Uniunii Europene · Regulamentul general privind protecția datelor (GDPR)
What is crypto-shredding and what problem does it solve?
Deleting a person is destroying their key, not rewriting history.
Bruce Schneier · Data Is a Toxic Asset, So Why Not Throw It Out?
What does k-anonymity guarantee, and what does it not?
A quasi-identifier is harmless alone; three together are a name.
Latanya Sweeney · k-anonymity: a model for protecting privacy