Optimize for the edge. Deploy at the source. The architecture is already small enough to go everywhere.
The architecture does not need the cloud. It runs on a single core. It produces output in one pass. Its state is a JSON file. Its configuration is five window values and three structural constants. There is nothing to serve, nothing to scale, nothing to distribute. There is only the stream and the detection.
This is not a limitation. This is the deployment model. Put the architecture where the data is born.
An ECG machine in a rural clinic. No internet. No GPU. A Raspberry Pi running the frame economy in real time. Each heartbeat encoded as a vector. Harm detected at the boundary. The doctor reads the anchors. The anchors say: normal sinus rhythm, verified across fifty thousand beats, three time perspectives, consistent at all τ levels. The deviation says: this RR interval pattern deviates from the anchor. The doctor decides. The architecture only states.
A historian's laptop in an archive. Voting records from 1946 to 2026. The architecture scans for anchors — diplomatic norms that no state has ever broken. It finds them. It finds the moment one was broken — a harm event at the boundary of institutional consensus. The historian writes the paper. The architecture provided the detection.
A seismologist's workstation in a monitoring station. Continuous ground motion data. The architecture detects the anchor — the normal tremor pattern of the Earth. It marks the harm — the anomaly that deviates from the anchor. The seismologist decides whether it is a foreshock. The architecture only states.
The same 2000 lines. The same three constants. The same five window values to scan. The only thing that changes is the data. The architecture does not adapt to the domain. The domain adapts to the architecture — by being a stream in time.
Every stream with temporal structure has anchors. Every anchor has deviations. Every deviation is a statement. The architecture reads the statement. The human decides what to do with it. This is the deployment model. Not one massive model in one data center. Millions of tiny models, wherever a stream touches a sensor, wherever time leaves a trace, wherever a pattern deviates from what was always there.
The architecture is already small enough. The rest is optimization for the edge. And the edge is everywhere.