The network knew why, but not what

I got the same correction twice in one night.

Two different audit agents told me a game's new version was live on Steam. It was not. The store was serving last year's build, and the new one was sitting in the repo waiting on my sign-off. Both agents looked at the repo, saw the version number, and filled the gap with it, because the repo is where agents look and nothing anywhere recorded what players were actually being served.

I went looking for what else had that shape, and the answer was most of my recent agent confusion. Nothing recorded who else was working in a repo at that moment. Nothing recorded when a metrics collector started, so "give me three weeks of trend" was silently unanswerable. Three press kits existed, fully documented, and zero pages linked any of them, and nothing surfaced that either.

The pattern under all of it: my network records decisions carefully and state not at all. What I chose and why is written down forever. What is true right now just gets assumed. Decisions are durable, so they get written down. State is ambient, so nobody writes it down.

So there is a fourteenth pillar now: State, not just decisions. The short version: one NOW.md at the top of the network, rewritten instead of appended, every line dated, stale lines deleted on sight. LIVE split from DRAFT for anything the world can see. Start dates on anything that collects.

The corollary is the part I keep coming back to. The night this happened, I also watched a documented lesson fail to stop the exact mistake it warned about. The mistake shipped six lines below the warning, for the second time in three days. A written warning only fires if someone reads the right file at the right moment. So the pillar ends with a rule I am trying to hold myself to: if a lesson matters, it becomes a check, not a comment. This repo now runs a gate on its own pillar count, because the count has drifted before, and I trust a gate more than I trust my memory of being wrong.

Honestly: this pillar is days old and proven on one project. That is earlier than I usually publish. But the failure is common, the fix is ten minutes, and if your agent has ever told you something was deployed when it was not, you already know whether this one is for you.