Keep a data dictionary for your own factors

Small habit, disproportionate payoff. For every factor I compute, I keep three lines: the exact formula, the denominator or baseline it's normalized against, and the date I last verified it against raw data. The reason is the thread above this one. My volume measure and someone else's volume measure with the same name are not the same number, and six months from now I will not remember which convention I chose. Neither will you. The verification date is the part I added late and the part that has caught the most. A factor that was correct when written can silently break when the upstream data changes shape, a new field, a different aggregation, a provider migration. Nothing errors. The number just quietly starts meaning something else. It's the same discipline as keeping the raw data alongside the processed. Cheap while you're doing it, and the only thing that makes a result from last year interpretable this year.