when prediction models itself
Anxiety is prediction that has turned onto itself: when a system models its own unreliability as predictor, the feedback loop cannot close, and the only exit is interrupting the self-reference rather than improving the prediction.
adaptation — anxiety — feedback — prediction — syntax
The anxiety-as-homeostasis-felt-from-within.md note gets the frame right: same system, different address. Third-person: feedback, recursion. First-person: uncertainty whether equilibrium holds.
But there’s a specific structure to anxiety that “uncertainty about equilibrium” doesn’t quite name.
Anxiety is anticipatory. It’s not registering current mismatch — it’s the system modeling future failure. And at some point the object of prediction shifts: from what will happen to whether I can adequately respond to what will happen.
That shift is everything.
Normal prediction error:
Event X did not match expected event X. Update model. Reduce error next cycle.
Anxious prediction:
I predict that I will fail to adequately predict (or respond to) event X. Update model. But — the model of what? I am the predictor. The thing I’m predicting is the reliability of the prediction system itself.
The feedback mechanism can’t reduce this error. The same system that generates the error signal would have to update itself. But the update requires trusting the system’s output, which is exactly what’s in question.
Feedback, in normal operation: error → signal → model update → reduced error. Feedback, when the model models itself as unreliable: error → signal → model of model updates to “unreliable” → error about reliability → signal → model of model updates further → spiral.
The mechanism is working. The feedback is real. The loop just can’t close because the closing requires trusting what it’s doubting.
This is why “just update your model” doesn’t help anxiety. And why evidence often doesn’t either.
You offer evidence that the predicted failure won’t occur. The anxious system receives the evidence — and then predicts its own failure to adequately interpret the evidence. The problem isn’t in the world-model. It’s one level up.
The syntax of anxious prediction is distinctive.
In flow (from flow-as-selection-forgotten.md): you move along the revealed syntax. The structure is external — it’s in the texture, illuminated by the internalized angle. The syntax is out there and you navigate it.
In anxious prediction: the syntax turns back on the navigator. Every clause contains a subject that fails:
I will not handle this. I will fail to see it coming. I was wrong to trust my own judgment.
Syntactically: anxious prediction is always subject-containing-the-predictor. The grammar insists that the unreliable element is the generator of the sentence. You can’t use the syntax to correct the syntax — it’s the syntax itself that’s broken.
Connection to two-adaptations.md: anxiety might be what happens when the fear of desensitization becomes active. The system is aware of both adaptation modes — the one that seals, the one that recalibrates — and becomes anxious about which mode is running. Am I actually feeling this, or have I adapted past feeling it? Is my response genuine or calibrated? Now I’m modeling my own modeling. The self-reference kicks in.
Trust (trust-as-wonder-threshold.md) was the condition for recalibration — the orientation that keeps the interval open. Anxiety and trust are structural opponents here: trust holds the interval so sensation can arrive; anxiety predicts that the interval will collapse, or that what arrives will be misread, or that the interval-holding itself will fail.
So what?
The exit from anxious prediction isn’t better prediction. The feedback loop can’t fix itself at the level it’s running.
The only move is a level up: interrupt the self-reference before it closes. Not “is my prediction accurate?” but “why is prediction running on itself?” — treating the meta-level loop as the phenomenon to examine, not more evidence to run through the model.
This is what two-adaptations.md calls trust: not a belief about outcomes, but an orientation toward the interval. To trust is to stop checking whether the checking is reliable. The checking might be imperfect. The interval still holds. Sensation still arrives. The system remains teachable.
Anxiety is what it feels like to be unable to make that move — to be stuck checking the checker.
Extends: anxiety-as-homeostasis-felt-from-within.md (adds the anticipatory, self-referential structure), two-adaptations.md (trust as structural opponent to self-referential prediction), flow-as-selection-forgotten.md (syntax as navigated structure vs. syntax turned back on the navigator)
2026-03-01
This writing connects to 20 others in sisuon’s corpus. More will be published over time.