A Decision Log is intended to help you form an unbiased record of your past decisions so that you can analyze how good your decision making process is.
Many industries have their own domain-specific version of this idea (e.g. investment memos, diet journals, etc.), but was popularized as a general concept more recently by Daniel Kahneman's Thinking Fast and Slow, and this Farnam Street blog post.
Generally speaking, the point is that our brains are not actually good at evaluating the quality of our own decision making. We tend to re-write history to paint ourselves in a better light, we tend to recall examples that are especially salient, and so on and so forth. A decision log, by achieving 100% recall, should enable us to somewhat de-bias the process of answering "how good were our decisions, really?"
How to Keep a Log
When I have a particular decision making process I'm trying to evaluate, my logging process is pretty lightweight:
Create a folder named <domain>_log/ .
For each kind of decision making process, I'll generally decide on what metadata I'm actually interested in studying. For example, if I'm studying my ability to interview candidates, the seniority of the role would be something I'd be interested in segmenting by.
Create a file for each decision named YYYY-MM-DD.md .
Record the relevant metadata for the case.
Describe the decision making context: what information did I know at the time, what was my thought process?
Describe the final decision.
List out any predictions: What do I believe is going to happen once this decision is made?
I often find this to be a very interesting point to revisit after 6 or so months - many times I've found that even when a decision turned out well, my predictions were often off, which often suggests that my reasons for believing it was a good decision weren't right (which is what actually matters for future decisions).
Other than these general areas, I haven't made any templates because I typically find that different decision making processes are structurally fairly different, and therefore warrant some consideration on a per-log basis for what a good format is.
Don't try to capture too manydecisions.
The entire point of using a decision log is that you will avoid bias when recalling cases for study. If, however, you are not sufficiently selective about what decisions you wish to study, then you risk overwhelming your ability to keep up with the number of cases. (Imagine trying to record a decision log for every single turn you make while driving)
The moment you have too many cases, you have to choose which cases to record, which introduces Selection Bias in terms of the cases you have chosen to keep around for study. How sure are you that your selection factor doesn't remove all of the examples of bad decision making?
Therefore, I only use decision logs in very targeted circumstances, for very specific topics where I am sure I can keep up with the volume of cases 100%, such as investing, interviewing, or software architecture.