Drift.

Drift is not an event. It is a rate.

Every node in this lattice loses conformance on its own quiet schedule. Nothing dramatic happens. A dependency ages out. A workflow keeps a permission it stopped needing. A branch rule is edited once for a good reason and never edited back. None of it is a failure on the day it happens, which is exactly why none of it gets fixed on the day it happens.

Drift also spreads. A node next to a drifted node decays faster here, because that is what happens in a real estate. The broken pattern is the one that gets copied. The stale workflow is the one that gets forked. Contagion is drawn in red between failing neighbours, and once it starts you will find you are no longer choosing where to look. The field is choosing for you.

Your attention is real and it works. Everything inside the amber circle genuinely recovers, faster than it decays. The circle simply is not large enough, and it is one circle, and you are one person. The coverage number in the rail is not rhetoric, it is a count of the nodes the circle reaches divided by the size of the lattice. That is the share of an estate a careful operator holds at any single instant.

Then press By policy. A sweep crosses the lattice at a constant rate and touches every node, including the ones that were never interesting, including the ones nobody had a reason to open this quarter. It is slower per node than you were. It never has a better idea. It holds the estate anyway, and it will still be holding it after you close the tab.

That is the whole argument, and it is the reason reliability at Atlas Systems is written as policy in a repository rather than as diligence in a person. Nothing here is fetched from the live estate. This runs entirely in your browser, from a seed, and no two runs are the same.

• held, above seventy five percent
• drifting
• breached, below forty percent
• gone