Ask a program why a decision was made eighteen months ago and watch what happens. Sometimes there is a register: the rationale, the owner, the date, in plain sight. More often there is a search through someone's email, a half-memory, and a shrug. The second program is one resignation away from amnesia, and it does not know it yet.
Memory leaks by default
Decisions, risks, and the reasons behind them are made in meetings and then evaporate. Nobody intends to lose them; the loss is ambient. The knowledge does not vanish all at once; it leaves with people, one departure at a time. The gap stays invisible until the day you need the answer and find the only person who had it walked out last quarter.
A register is not bureaucracy. It is the difference between a program that learns and one that relitigates the same decision every time a new face joins the room.
Capture is cheap; reconstruction is not
Writing down the decision, the owner, and the why takes minutes in the moment, while the context is still warm and free. Reconstructing it later, after the person has gone and the detail has faded, can take weeks, if it is possible at all. The economics are not subtle. You either pay a little, on time, or a great deal, late, for worse information. Programs choose late more often than they would admit.
The memory must outlive the people
The test is brutal and worth running anyway: if your most knowledgeable person left tomorrow, how much of the program would walk out with them? If the honest answer is "most of it", you have not built a program. You have built a dependency on a human who is, eventually, going to leave.
Treat the program's memory as a deliverable with an owner, not a by-product of meetings. The work you cannot remember, you are doomed to repeat, at full price.