Fable
The carpenter writes down the mechanism rather than just describing it
The craftsman academy exam used to ask apprentices: if you want the waterwheel to stop at a specified water level, what would you do? Those who could recite the book answered well.
Later the workshop really needed an automatic waterwheel. Those who could explain it could not write a mechanism that would turn, and the mechanisms that could be written often jammed on edge cases.
The director first had the apprentice write a longer plan. The plan read as complete, but no one knew whether it would move once placed in the river.
He then had the teacher score by experience. The teacher favored familiar styles and overlooked that certain short mechanisms could actually run reliably.
The new examiner changed the test: give the function description and sample test cases, and have the apprentice write an executable mechanism. The mechanism would be placed into a set of hidden test troughs.
Some answers read beautifully but leak the moment they run; others are plain but pass every sink. The scores began to approach real contraption-building ability. The director also broke the results down: can it handle an empty sink, can it withstand a full one, can it stay steady when the water level is right at the edge. That way the score was no longer just a pretty ranking, but could point out which small jobs were already reliable and which needed to go back for rework.
The examiners also admitted that this only tested small mechanisms, not the ability to build an entire waterworks. Leaked questions or excessive drilling could also inflate scores. Later the academy tightened control over the question bank as well. Public practice questions could let apprentices get familiar with the rules, but formal exams had to retain unseen mechanisms; otherwise some would only memorize the shape of the answers and still not know how to repair a real river channel.
The workshop therefore treats it as a ruler for writing agency capability, not as proof of all software engineering capability. The director said: those who write agency must ultimately make the code run, not just make the thinking sound smooth. Later the manager wrote this experience into daily rules: first look at the scenario, then the boundary, and finally who can take over when it fails. Only when all of this is clear will a beautiful demo become deliverable work. During reviews, they also compare against old incidents to confirm the new rules are not just good on paper, but can truly block the same mistakes in a chaotic scene.