How stability, drift, collapse and recovery connect chronic illness, artificial intelligence, furnaces and galaxies At first glance, this publication may look like several different Substacks accidentally sharing a homepage. One article is about chronic illness. Another is about artificial intelligence. Then there is cosmology, furnace engineering, cognition, topology, memory, model reliability and the strange mechanics of why a capable system may still fail to produce a conventional outcome. The subjects move quickly. The underlying questions do not. Again and again, the same pattern appears. A system begins in a state that can absorb disturbance. Pressure accumulates. Small errors stop correcting themselves. The system begins to drift. A boundary is crossed. Behaviour changes rapidly. Then, if recovery remains possible, the route back is slower, more expensive and rarely identical to the route in.
Stability. Drift. Collapse. Recovery. These four states kept appearing often enough that they stopped looking like separate stories. They began to look like one structural language. I eventually gave that language a slightly ridiculous but accurate name: Joe’s Unified Domain State Theory. JUDST. The name is less important than the observation behind it. Different systems can behave differently for entirely different reasons while still passing through recognisable state transitions under pressure. A furnace does not become unstable for the same mechanistic reason as a nervous system. A language model does not hallucinate because it has an autonomic stress response. A galaxy does not drift because it is tired. That would be analogy used badly. The claim is not that these systems share one mechanism. The claim is that once each is expressed as a state evolving under interacting pressures, some of the same behavioural structures become visible. Regions that absorb disturbance. Slopes where errors begin to amplify. Thresholds beyond which behaviour changes rapidly. And slower paths through which stability may be rebuilt. That is the level at which JUDST operates. Not mechanism. State behaviour.
The difference between mechanism and state Mechanism asks: What physically, biologically or computationally causes this behaviour? State theory asks: What kind of behaviour is the system currently exhibiting under the pressures acting on it? Those are related questions, but they are not the same question. Take a furnace. Its mechanism includes heat generation, conduction, radiation, convection, insulation, thermal mass, sensor response and controller behaviour. But from a state perspective, we might ask: Is the temperature field stable? Are local gradients increasing? Is the controller still correcting deviation? Has the system entered runaway? What intervention would return it to a controllable regime? Now take a person living with chronic illness. The mechanism may involve autonomic regulation, inflammation, pain signalling, circulation, sensory load, fatigue and medication. But from a state perspective, we might ask: Is the system absorbing ordinary demand? Are small demands producing disproportionate effects? Is recovery still occurring between pressures? Has the person entered a shutdown or collapse regime? What reduces pressure enough to reopen a recovery path? The mechanisms are completely different. The state questions rhyme. Now consider an artificial intelligence system. Its mechanism involves model weights, attention, token probability, context, retrieval, tool use and software architecture. But from a state perspective: Is the output grounded and self-consistent?
Are small errors being corrected or compounded? Is context beginning to drift? Has narrative coherence started replacing evidence? Can the system recover through re-anchoring, or must the run be restarted? Again, different substrate. Similar behavioural questions. JUDST begins at that level. A system is not only what it is made from We often classify systems by material. Biological. Mechanical. Computational. Social. Cosmological. That is useful when the mechanism matters. But another valid way to classify systems is by how they behave under changing constraint. A bridge, a nervous system and a model may have nothing in common materially. Yet each may still possess: a range in which disturbance is tolerated; limits beyond which sensitivity increases; feedback that stabilises or amplifies change; thresholds where one regime becomes another; recovery costs; memory of prior stress. This does not make them the same system. It makes them comparable at a higher structural level. The same distinction appears throughout science. Thermodynamics can describe gases without tracking every molecule. Control theory can describe regulation without caring whether the controller is mechanical, biological or electronic.
Information theory can describe uncertainty without requiring the message to be written in one particular language. JUDST attempts something similar for state behaviour under interacting pressures. It asks whether stability, drift, collapse and recovery form a reusable behavioural grammar across domains. The four states The names are ordinary on purpose. They describe behaviours people already recognise. Stability A stable system does not need to remain perfectly still. It can move, adapt and respond. Stability means that disturbance remains bounded. Small changes do not automatically become large changes. Errors are corrected. Feedback remains functional. The system retains enough margin to absorb variation without losing its operating form. A stable person can experience stress without losing all function. A stable furnace can absorb small changes in load without temperature runaway. A stable model can handle ambiguity without inventing an unsupported answer. A stable organisation can encounter disruption without abandoning its core function. Stability is not the absence of pressure. It is the ability to contain it. Drift Drift begins when the system still functions, but no longer corrects cleanly. Small disturbances begin to persist. Sensitivity increases. Margins narrow.
Errors accumulate. The system may still look broadly operational from the outside, which is why drift is so easy to miss. A person may still complete tasks, but recovery takes longer each time. A furnace may still reach its setpoint, but local hotspots are growing. A model may still produce fluent prose, but its claims are becoming less anchored. An organisation may still deliver services, but delays, workarounds and contradictions are multiplying. Drift is not collapse. It is the weakening of the forces that prevent collapse. Collapse Collapse occurs when the system crosses into a regime it cannot correct from using its ordinary internal processes. This does not always mean destruction. A collapsed system may still exist. It may even continue producing outputs. But its behaviour has changed qualitatively. A person may enter shutdown. A furnace may enter thermal runaway. A model may commit to a false frame and construct an increasingly coherent hallucination around it. An institution may become trapped in a regime where each attempted correction creates further instability. Collapse is not merely “more drift. ” It is a change in the structure of the dynamics. Recovery Recovery is the process by which a system regains a stable operating regime. It is rarely a simple reversal. The path into collapse may be fast.