Data Extract

A point-in-time copy of data pulled for one question, and how it drifts from the source.

Also known as: data extract, data pull, data snapshot, exported data

A data extract is a copy of data pulled out of a system at one moment, usually into a file, to answer one question. You export a table to a spreadsheet or a CSV, or pull a month of events onto your machine, and work on the copy.

Extracts exist because they are easy. No access request, no waiting on a slow query, no risk of hammering the live system, and the file is yours to slice offline with tools you already have. They also have one useful property: an extract is frozen, so a result computed from it can be reproduced exactly — as long as you still have the file and the date you pulled it.

The trap is drift. The source keeps moving. A week-old extract is an accurate description of last week and of nothing else. Rows have appeared, rows have been corrected, statuses have changed, and a “discrepancy” between it and a fresh query is often just time (data freshness).

What to do about it:

Record with the file: when it was pulled, the query or filters that produced it, and the row count. Without those, an extract becomes a number nobody can trace (metric definitions).
Date-stamp the finding: “as of the 5th”, not “currently”.
Reconcile before comparing: check the row count and a total against the source (data reconciliation).
Anything that has to be produced again belongs in a query against the source (spreadsheets vs SQL).

An extract is a good answer to “what was true when I looked?”. It is a poor answer to “what is true now?”.