The test suite said: PASS. All green. One hundred percent.
The satisfaction metric said: 98.47%.
Jay stared at both numbers side by side. The test suite was from the old codebase, still running against the legacy modules as a sanity check. The satisfaction metric was from the factory, evaluating the same functionality against the full scenario set.
"One of these is lying," Jay said.
"One of these is lying," Justin agreed. He'd been standing behind Jay for thirty seconds, which was a thing Justin did—appearing silently when something interesting was on a screen. "The question is which one."
Jay knew the answer. He'd known it for weeks. But he wanted to hear Justin say it.
"The test suite is lying," Justin said. "Not intentionally. Boolean tests can't lie intentionally. But they lie structurally. A test says 'this works' or 'this doesn't.' Green or red. Binary. And binary is a simplification of a complex reality."
He pulled a chair over. "What does it mean for software to work? Does it work for one user? For a thousand? For a user in Tokyo with a flaky connection and a three-year-old browser? For a user provisioning fifty new employees simultaneously? All green tells you is that the specific inputs you chose produced the specific outputs you expected. It says nothing about the space between those inputs."
"Satisfaction says something about that space," Jay said.
"Satisfaction says 98.47 percent of observed trajectories through all scenarios likely satisfy the user. That 1.53 percent isn't a failure. It's an honest accounting of uncertainty. It says: we are very confident this works, and here is exactly how confident we are."
Navan, who had been listening from across the room while drawing something in his notebook, looked up. "Boolean truth is comforting. Probabilistic truth is useful."
Justin pointed at him. "Write that down."
"I already did."
Jay thought about the green checkmarks he'd accumulated over years of engineering. All those CI pipelines, all those green badges on pull requests. The dopamine hit of ALL TESTS PASSING. It was a kind of certainty. A warm, simple certainty.
The satisfaction metric offered no such warmth. 98.47 percent was not a celebration. It was a measurement. It invited the follow-up question: what's happening in the 1.53 percent? And that question led to more scenarios, more evaluation, more convergence. The number was never final. It was always an invitation to do better.
"Which is more honest?" Jay asked aloud, more to himself than anyone.
No one answered. The question was rhetorical and everyone in the room knew it. Boolean truth said the world was simple. Probabilistic truth said the world was complicated and here was a number that tried to account for the complication.
The test suite stayed green. The satisfaction metric stayed at 98.47 percent. Both were on the same screen. Both described the same software. Only one of them was telling the truth.
"Boolean truth is comforting. Probabilistic truth is useful." This needs to be on the wall of every engineering office. Frame it. Put it next to the fire exit sign.