They hired a knowledge management consultant six months before Renata retired. He interviewed her for forty hours. He filmed the processes, catalogued the decision trees, mapped the workflow dependencies. The manual ran to three hundred and twelve pages, cross-referenced and indexed.
“It’s all here,” he told the operations director. “Every procedure, every contingency, every variation she uses.”
The manual was accurate. They verified each section against Renata’s actual practice. Every step matched.
The first failure came eleven days after she left. A pressure reading looked normal to the technician following the manual. Renata would have noticed that the number was normal at the wrong time of day—that the morning spike hadn’t come, which meant the evening compensator had been working too hard, which meant the upstream seal was starting to go.
The consultant was called back. He added the case study to the troubleshooting section. A paragraph about temporal context for pressure readings.
The second failure was different. The manual said to check the filter every seventy-two hours. Renata checked it every seventy-two hours too—except when she didn’t. Sometimes she checked it at forty-eight. Sometimes at ninety. The manual couldn’t capture why, because Renata couldn’t explain it either. She said she listened. She said it sounded different. The consultant wrote “use professional judgment regarding inspection frequency” and knew it meant nothing.
By the sixth month, the manual had grown to four hundred and eighteen pages. Every failure had been incorporated. The troubleshooting section was meticulous. And the failure rate was three times what it had been under Renata, because every addition taught the technicians to follow the manual more carefully, and following the manual more carefully was the problem.
The knowledge was never in the steps. It was in the decision about which step to skip.