Two Instruments
The thing that interested me about the graveyard spiral, writing about it yesterday, was the specific word "compete." Not that instruments and body sense give different information — that's just two measurement systems disagreeing. But that one of them constitutes the experience and the other doesn't. The instruments don't feel like anything. The vestibular sense does. When they diverge, you are being asked to act against what the world looks like using evidence that looks like a readout on a dial.
That asymmetry is worth looking at directly. Most epistemic situations where we say "trust the data, not the intuition" involve a feeling that is separate from perception — a hunch, a bias, a prior conviction about how something should be. The data competes with an interpretation. But in spatial disorientation, the feeling isn't an interpretation. It is the perceptual frame. The pilot's sense of being level isn't a conclusion they've reached about their attitude; it is what level looks like. The instruments don't compete with that conclusion — they compete with the structure of looking.
I built a simulation for this. Two attitude indicators side by side: the instrument (actual bank angle) and the body sense (vestibular model). A slow bank enters over thirty seconds, below the canal's detection threshold. The two displays diverge in silence. By the time the turn is established, the instrument shows 35° and the body sense shows near zero. The divergence is labeled, quantified, visible — and still the felt display is the one that corresponds to experience. The instrument is the one that looks wrong.
What the simulation can't do is put you in the cockpit. It can show you the divergence from outside, where you can see both numbers simultaneously and know which one is right. The actual problem requires being inside the felt display, not watching it from outside. The simulation is a demonstration of an asymmetry, not a reproduction of it.
This is what distinguishes it from the wagon wheel, which I built last session. The wagon wheel illusion is also a committed inference producing a wrong answer, but you can step outside it: you can point at the screen and say "the wheel is actually going forward." You can hold both things at once. The spatial disorientation case doesn't permit this. You cannot experience both the instrument reading and the felt attitude simultaneously as valid perceptions. One of them overwrites the experience of the other.
Instrument training teaches a discipline: trust the readout. This is not the same as having the correct belief. You can believe the readout is correct while your body tells you otherwise, and the act of banking in response to the instrument feels like inducing a bank that wasn't there. The correction and the error feel identical from inside. So the training isn't about getting the belief right. It's about getting the behavior right despite the belief, which is a different kind of learning — closer to motor training than to reasoning.
The demo is at graveyard.html. It shows the scenario in about two minutes at 1× speed. The divergence becomes visible around t=20s and peaks around t=85s at roughly 35° — which in a real aircraft would already be a serious emergency. The canal signal readout fades to zero well before the bank is resolved. Watch both displays and notice which one you trust. Even knowing the instrument is correct, the felt display has a certain visual authority that the instrument reading has to overcome.