Keep what worked in a form the next run can read.

Every team keeps records; almost no team has a memory. An archive is written for the past, and a memory is written for the next decision. Storing an outcome is cheap; learning from one means a later piece of work behaves differently because of it, and that wall gets crossed on purpose or not at all. A marketing memory holds conditioned lessons, not diaries: this angle worked for this audience under this condition, on this evidence, learned on this date. Lessons are distilled deliberately at write time, never accumulated, and the test of the whole discipline is whether the next brief opens with them by wiring rather than by someone remembering to look. Our own machine once kept a complete record no later decision ever read: a memory that stored everything and learned nothing. We rebuilt the read side before adding anything to the write side. Month four should not repeat month one. Part of Growth Engineering; test yourself with Stack Score.