A Good Result Is Not Proof of Control

Boards judge fast-moving organisations mostly by what they produce. Revenue, capacity, delivery dates and service levels are outcome evidence, and outcome evidence is direct evidence of one thing, that the result occurred. By itself it cannot show whether management can still see and influence the conditions that produced the result, or whether the result could be produced again if the people, partners or technology changed. This paper calls those control evidence and capability evidence, and argues that a success claim should state which of the three kinds of evidence it supports before anyone relies on it. The idea that technology changes faster than the structures meant to manage it is old. Rasmussen set it out in 1997, and the three-way split has close relatives in Donabedian, Ouchi, audit practice and the replication literature. The contribution claimed here is narrower. It is a governance test that assembles those parts and adds a timing comparison. At each consequential interface, a board asks whether the organisation can refresh its working picture faster than the time left before the next commitment becomes hard to reverse. Nine documented cases show the test at work, seven failures and two examined well. OpenAI’s infrastructure, financing and governance changes between January 2025 and October 2026 provide a live public case. The paper closes with six dated observations, fixed now, that will show by mid-2027 whether OpenAI’s control evidence is keeping pace with its outcome evidence. Status: practitioner research, not peer reviewed. The governance test has not yet been validated.