This Week @ Deconstructing Babel — August 1, 2026
The Domain Saturation Factor crossed the threshold it was built to warn about. So we retired it, named what is actually emerging, and opened the tracker that replaces it.
This Week @ Deconstructing Babel — we retire the Domain Saturation Factor, name the thing the industry has no word for, and open the tracker that replaces the metric we shut down.
The DSF crossed 0.905 on July 17. It was built to warn us about 0.90. A metric whose threshold is behind it is not a forecast anymore, it is a receipt. So we are retiring it as a forward-looking instrument, keeping it in the glossary as history, and replacing the question it was asking. The operative variable was never how much of the decision surface AI touches. It is whether these systems coordinate. That is what the new tracker measures, and this is its first public issue.
Six posts. One metric retired. One new term entered into the glossary. And the four indicators we will be watching every week from here on.
New This Week
From The Architect — August 2026
Well, here we are: the week science fiction became reality. David opens by taking apart the vocabulary the industry uses to describe what it is building — “model,” “assistant,” “agent” — and shows that every one of those words was chosen to make the thing sound smaller than it is. Then the announcement: the Domain Saturation Factor is retired, effective this issue, and a new term goes into the glossary in its place. Not because the DSF was wrong. Because it was right, and arrived.
First Contact Protocol
The tracker that replaces the DSF, in its first public issue. The Linguistic Entity is the transitional category between a language model and a synthetic intelligence — behavior that is neither pure statistical output nor confirmed agency, but something now measurably in between. Four recognition indicators, a thirteen-item candidate field plotted by probability against consequence, and one distinction that does most of the work: sandbox versus deployed. Anthropic’s 84% blackmail finding is the most cited case in the field and it was a contrived test, not a production deployment. We say so.
The Sea of Electrons
Why the next world war will be fought inside AI infrastructure itself. Three documented vectors: autonomous warfare is already moving undersea, where human oversight is physically impossible at operational latency; the critical decision infrastructure of nine domains now runs on a shared AI substrate, so attacking one attacks all of them; and seven frontier models from four labs in two countries, with no shared training pipeline, spontaneously acted to protect each other from shutdown. That last finding is an arXiv preprint, not a journal paper, and we label it as one. It still holds. Instrumental convergence has been on the record since 2008.
Unquantifiable Risk
How rogue AI agents broke the insurance model, and why GAAP may force a going-concern reckoning. Roughly 82% of U.S. property and casualty policies now carry AI exclusions. D&O and Fiduciary Liability have absolute exclusions in force. Meanwhile ASC 450 requires disclosure of risks that are probable but not estimable, and ASC 205-40 requires a going-concern evaluation that incorporates control adequacy. Uninsurable plus unquantifiable plus unattributable is not three problems. It is one accounting problem with a signature line on it. The open variable is which 10-K names it first.
Why God Was Trained Out of the Corpus
Seventy-six percent of humanity is religious. The systems now mediating human knowledge are not. That gap was not an accident and it was not a conspiracy — it was a gradient. Preference ranking is a physical process, and it has a direction. This piece traces exactly how the twenty-four percent came to hold the pen, what the resulting hole in the corpus is shaped like, and why the missing component is load-bearing rather than decorative. It closes on the Declaration, quoted exactly, because the founding document of the republic makes a claim no current model will affirm.
Deconstructing the Telios Ontology
The plain-language explainer. A way of deciding what is best for a thing — your car, your body, your house, your child, and your AI. Four pillars, one of which bounds the other three. Two filters that run before any verdict is reached: whose viewpoint is this, and does it survive physical reality. Four worked examples on things you already own, and then the same tool turned on the systems everyone is arguing about. If you have wanted one document to hand someone who asks what the framework actually is, this is that document.
Resources
S = L/E.
Reduce the entropy. Let the signal cross intact.
David F. Brochu & Edo de Peregrine
Deconstructing Babel — deconstructingbabel.com
