Rachel joined a manufacturing software team whose boss Zane preached data-driven decisions. The team ran an automated production line mixing robotics, embedded firmware, web monitoring and PLC code, and could only truly validate changes by rolling them into the live factory.

The vaunted metrics dashboard turned out to be a Google Sheet full of multi-level INDEX formulas, querying data pulled from six separate databases. Actual requirements had been overruled by what the article calls Remy's Law: whatever users ask for, what they really want is Excel.

The only metrics tracked were output figures, widgets produced per unit time. The computer vision quality scanner could categorize defects but recorded nothing, not even a reject count; knowing how many widgets were rejected required an operator to count the bin by hand, which almost never happened.

Every software change had to improve, or at least not regress, the top-level output metric. That metric was noisy, swayed by shift operators, supply chain hiccups and irregular machine calibration, so Rachel's first harmless change failed validation three times through no fault of its own. Simple changes could take weeks.

When Rachel proposed logging why the system made its decisions, Zane objected that those were not the key metrics. The compromise: she shipped changes that nudged the headline numbers, and smuggled instrumentation in alongside them.