Every business keeps outcome records. Revenue, pipeline, headcount and cost are all tracked, usually to a level of detail nobody reads. Almost none keep a record of the decisions that produced those outcomes, which means the outcomes cannot be attributed and the same argument gets held annually with fresh conviction.
A decision log records four things at the moment of the decision. What was decided. What was believed to be true when it was decided. What was expected to happen. And who made the call. It takes a few minutes and it is almost always written after the fact, badly, if it is written at all.
What it protects against
The main failure it prevents is the quiet rewriting of expectations. When a result comes in, the memory of what was predicted adjusts to fit it. Everyone involved is honest and the adjustment still happens. A written expectation is the only thing that survives the adjustment.
The second failure is losing a correct decision to a bad outcome. A well-reasoned call against good evidence can produce a poor result, and without the record the organisation learns the wrong lesson and stops making that kind of call. The log separates the quality of the reasoning from the quality of the luck.
In the fifth lens the log sits alongside the experiment register and the variance review, and the three of them are the reason a reformation can be handed over. A system whose reasoning exists only in the heads of the people who built it is a dependency, whatever else it is.