Part 6 of Prediction, Control, and the Regulating Mind

The Loop That Won't Settle

Dr. Greg Moody · August 24, 2026

Somebody checks the lock, walks away, and comes back ten minutes later. That runs all evening. Nothing about the door has changed the whole time (the door has never once been unlocked, and that fact has never once helped).

Rumination, worry and compulsive checking get different labels and different treatment protocols. Put them into a control model and the correction keeps firing because the error never closes.

Part 1 set up the pattern. A loop like that doesn't just push too hard, it oscillates – overshoot, correct, overshoot the other way, hunt around the target and never land. Wiener wrote it up in 1948 watching cerebellar tremor, where the muscle is fine and the correction keeps arriving too hard and too late (Wiener, 1948).

When the Correction Won't Hold

Fradkin and colleagues built a computational model where the symptoms come from uncertainty about state transitions, meaning whether an action actually changed anything, rather than from a wrong belief about the world (Fradkin et al., 2020). Now you can ask whether they trust that the lock changed state, which gives you more to work with than "they're anxious about germs."

Run that on the lock. If you don't trust that turning the key changed the state of the lock, looking at it doesn't help, because looking is also an action and you don't trust that one either. The loop can't converge on anything. What's uncertain is the loop's own effect.

A family can use that. “Why can't they just stop?” has never once produced a useful answer. Asking what part of the correction won't hold gives you somewhere to start.

Rumination is messier and I'd rather say so. Watkins and Roberts review it as a transdiagnostic vulnerability with a stack of maintaining mechanisms – habit, executive control, abstract processing style, goal discrepancies, negative bias (Watkins & Roberts, 2020). The review is broader than the control model. Repetitive thought driven by a goal discrepancy may still behave like an unstable loop, because thinking about the discrepancy doesn't close it.

Not Acting IS the Correction

How does doing nothing fix a system that's already stuck? Because the doing was the thing keeping it going. Every check is a fresh correction, and the corrections are arriving faster than the situation can absorb them.

Tell them the next part before it happens, not after. The first few nights may get WORSE. An under-damped loop can overshoot as it settles, so a bad night early on can be consistent with a system on its way to settling (I have lost people at that point by not saying it in advance, which is a mistake you only need to make once).

Worry postponement moves the worrying to a set hour. Waiting a day before answering the email does it too. You're slowing the loop down.

Öst and colleagues meta-analyzed 37 randomized trials of CBT for OCD from 1993 to 2014 and found very large effects against waiting list (1.31) and placebo (1.33), and real superiority over antidepressant medication (0.55).

They also found no significant difference between exposure with response prevention and cognitive therapy, at 0.07 (Öst et al., 2015). If those two land in the same place, then explaining how one of them works hasn't explained the treatment. Whatever is doing the job, both treatments reach it, and I hold this section loosely because of it.

Telling Someone to Stop Raises the Gain

This is why "just don't think about it" fails, and who does it fail hardest on? The conscientious ones, because they actually try.

Somebody two hours into a rumination isn't short on discipline – they're running an accurate correction on an estimate that won't stabilize, and every extra demand you put on them goes in as more gain.

Part 4 gave you four questions for telling apart presentations that look identical from across the room. The first one does the work here: is the belief inaccurate, or is it accurate and being acted on too hard? Restructuring answers the first one and may do very little for the second (and on an oscillation case the second one is most of what's in front of you).

Depression Produces a Symptom and Its Opposite

Sleep and appetite each have to be controlled in two directions. One governor makes sure you sleep enough and another makes sure you wake up. One makes sure you eat enough and another keeps you from eating too much.

Powers put up the constraint that makes that necessary half a century ago – a neural comparator only signals error in one direction, so controlling something in both directions takes two loops (Powers, 1973). In this model, a fault in the shared machinery lands on one side or the other. Hit one and you get insomnia, hit the other and you get hypersomnia, and it may be the same KIND of fault either way.

Slime Mold Time Mold put the objection better than I can. Take your car in and the mechanic tells you it has a case of broken-downness. You would be back out in the parking lot before they finished the sentence! A real mechanic says the spark plugs are shot so they can't drive the pistons (Slime Mold Time Mold, 2025a).

Then watch what we do when somebody feels sad all the time. We tell them they have depression… which names no parts and no rules those parts follow, and it still gets a billing code.

Depression Is Three Different Faults

What would you actually see in a room?

A set point set too high. Nothing clears the bar, so nothing registers as a win. This person may be doing well by every external measure and getting no return on any of it.

A controller with the gain turned down. The error registers and the correction can't get mounted. From outside that's anhedonia and psychomotor slowing. They can tell you exactly what's wrong and can't move on it, which reads as not trying to anyone who isn't paying attention.

An estimator that stopped updating. Good news arrives and changes nothing, because the model making the prediction isn't taking new data. Stephan and colleagues built a version of this as allostatic self-efficacy, where depression follows from a belief, sitting under the mood, that your own regulatory actions no longer work (Stephan et al., 2016). Treat it as a proposal waiting on a test.

I haven't found that test. It's the clearest testable claim this model makes in a clinic, and it follows from the constraint Powers described.

What to Do With This in the Room

Not much, deliberately. There is no outcome data behind any of it, no trial has tested whether explaining it to a client helps, and it does not license a new technique. If it made me change a protocol I'd be doing exactly what I'm criticizing.

Use it as a question. When restructuring keeps sliding off, ask whether the set point, the gain or the estimator is causing the problem. Nobody has run the study that separates them, so don't treat the answer as established.

Try it on one case. If it changes nothing, you've lost four minutes. Part 7 is where I lay out what would show the whole framework is wrong – this section is the first place I'd look.

References