Code tells you what was built. It can't tell you why, why Postgres over MySQL, why a single VPS, why this library and not that one. Without the why, the next maintainer relitigates every decision.
The fix is a decision log: a short file in the repo, one entry per significant choice, the decision, the alternatives considered, and the reason. Two paragraphs each, written when the decision is fresh.
It's the cheapest documentation there is, and the most valuable. The code shows the what; the log preserves the why.
