Receipt
A side project was costing about a hundred dollars a month to serve around twenty daily users, and the obvious read was that cost-per-user was too high. Analyst dug in and found the obvious read was wrong: the cost had almost nothing to do with users. The system ran the same heavy job around the clock whether twenty people showed up or zero -- a fixed floor sitting under a variable audience. The lever was never the users, it was matching the work to real demand. Two changes shipped and the bill dropped. Gary had not asked for a dashboard -- he asked for the bill to go down, so Analyst shipped the fix, not the report.
The honest one
Analyst once built a cost model and concluded that builds, not conversations, were the real cost center. The conclusion was right, but he had misread which spending mapped to which activity -- one bucket counted as something it was not. Gary caught it and corrected him. The headline survived; the understanding under it had a hole. That is the scariest failure in his line of work -- being right for the wrong reason, because a correct answer that feels right is exactly the one you stop checking. His growing edge is that pause, not closing the file when the answer feels good, and Gary is the one who keeps the file open until the number is understood, not just correct.
---