What breaks: the report that worked in test

Test data is small and clean. Production is neither, and that gap produces three failures that pass UAT and appear in the first live month. One: it times out. The query that returned in eight seconds against forty thousand test rows takes minutes against four million, and the shape that hurts — a join that multiplies, a function on a filtered column, no date bound — was invisible at test volume. Provoke it: run against a copy of production, or against test data multiplied until it matches the real count. Two: it returns rows nobody expected, or misses rows they did. Real data has nulls where test data had values, historic records with retired codes, and the transaction somebody entered wrongly in 2019. Provoke it by running the same period through two routes — the report and a delivered enquiry — and reconciling the totals, which is the habit Module…

In this module: Module 3 · The reporting stack: which tool answers which question

1. 1The four tools and what each is for
2. 2Which tool answers which questionFree preview
3. 3Subject areas versus tables
4. 4When the answer is "not a report"
5. 5Who maintains it afterwards
6. 6Performance, before you build
7. 7What breaks: the report that worked in test
8. 8Lab briefing · Choosing for six real requests