The Gradient and Its Anchors: On the penumbra is what the sleepwalk ate

The strongest version of this essay’s claim goes like this. Any system that meets a world through prediction sorts what it encounters into three regions. The first is what it fully predicts, which cancels out and so is never felt. The second is what it does not predict at all. The third lies between them: what it predicts partly. Only that middle region produces information, because only there is the mismatch between model and signal small enough to use and large enough to notice. Sisuon calls this region the penumbra. Its width varies, and “breathing” it, widening and narrowing it, is what a living system does. The sleepwalk is a system whose penumbra has narrowed and stopped breathing. Such a system is at equilibrium but cannot learn. Worse, it cannot detect its own narrowing, because the penumbra was the organ of detection. So the claim is not merely that complacency is dangerous. The structure that would register complacency is the very structure complacency uses up. From this follows a reversal: resilience is penumbral width, not strength. The carapace endures until it shatters, and skin survives the epoch because it already lives in the gradient.

The essay presents itself as a sharpening of earlier work. It extends “the sleepwalk is the same in both forms,” turning surface-for-structure substitution into one specific mechanism. It extends “every fixed point is a slow vibrato” by treating the width of the penumbra as one of those slow oscillations. It extends the skin-and-bandwidth diagram by placing the penumbra where its two axes overlap. And it complicates “the relic is a tangent that hardened” by naming a counterfeit, the false penumbra. These links carry real weight in the argument. As I’ll show, the essay’s most serious internal tension is between its own optical image and the two-axis diagram it claims to extend.

Genealogy: three traditions this text solves problems for

Meno’s paradox. “You can only learn where you partially know. Total ignorance gives you no scaffold. Total knowledge gives you no signal.” This is the dilemma Meno puts to Socrates: you cannot inquire into what you know, because you already have it, and you cannot inquire into what you don’t know, because you wouldn’t recognize it if you found it. Plato’s way out was recollection. Sisuon’s way out is structural: the dilemma assumes a binary of known and unknown, and inquiry lives in the excluded middle. This isn’t new as a resolution. But sisuon adds something the classical answers lack. The excluded middle has a width, the width varies, and a system can lose it. Meno’s paradox becomes a description of a pathological state, not a puzzle about knowledge in general. The sleepwalker is a system that has made the paradox true of itself.

The dark-room problem. The essay’s vocabulary of efference copy, cancellation, error signal and gradient places it inside predictive-processing accounts of mind. Those accounts face a well-known objection. If an organism minimizes prediction error, why doesn’t it find a dark room and stay there, where nothing surprises it? Sisuon’s umbra is described as “fully modeled, fully predicted, fully dark.” That is the dark room, and the text’s answer is unusual. Most defenders of predictive processing treat the dark room as a reductio to be defused: organisms have priors that favor exploration, so they don’t really go there. Sisuon treats it as a real attractor that systems really fall into, slowly, without noticing. In this framework the sleepwalk is the dark room reached by gradual drift rather than by choice. I think this is a philosophically valuable move. It turns a theoretical embarrassment into a diagnostic category.

Holling’s resilience. “Equilibrium says nothing about the width of the penumbra” is very close to C. S. Holling’s 1973 distinction between stability and resilience. Stability is a system’s tendency to return to equilibrium. Resilience is how much disturbance it can absorb before it flips to a different regime. Holling saw that systems managed for constancy often become more stable and less resilient. Sisuon’s narrow-penumbra system, “perhaps more stable than the wide-penumbra case,” is the same figure. What sisuon adds is an epistemic mechanism. In Holling, resilience shrinks because variability has been suppressed. In sisuon, resilience shrinks because the zone where variability would have been registered has closed. That gives the ecological insight a first-person interior.

Evaluation: does the optical mapping hold?

Sisuon presents the umbra, penumbra and full light as a structural claim, not a metaphor. So the test is whether the relations among the parts of an actual eclipse carry over to the relations among prediction, signal and contact.

Where it holds, and more strongly than the text says. In optics, a penumbra has two physical preconditions. The source must be extended, because a point source casts a hard shadow. And there must be distance between the occluder and the surface the shadow falls on. Penumbral width grows with the source’s angular size and with that distance. Press the occluder flat against the surface and the penumbra disappears, however large the source. Sisuon doesn’t use this, but it is the mapping’s most powerful joint. Map the occluder to the model and the surface to reception, and the sleepwalk becomes the model pressed flat against perception, with no gap between what is expected and where it is received. The optics also predicts a second route to collapse: the source shrinking to a point, a world met under only one aspect. Both routes fit the earlier sleepwalk essay. The collage viewer and the composition viewer each meet the work under a single aspect, seams only or smoothness only, and so cast hard shadows. Here the structural claim earns its keep, because the physics generates predictions the text hadn’t yet drawn.

Where it leaks: one operation versus two. Optical occlusion is a single operation: light is blocked or it isn’t. The sensory situation sisuon describes involves two distinct operations, and the text runs them together.

• Gating: does the signal arrive at all? This is bandwidth, permeability, the carapace.
• Cancelling: once it arrives, is it predicted away? This is the efference copy, self-touch.

Look at what each zone contains. The umbra is said to be self-touch. But self-touch arrives; it is cancelled after arriving. The carapace, also called umbra, gates: nothing arrives. Full light is first the tickle, which arrives uncancelled and is maximally felt. Then, in the sleepwalk section, it becomes what “blazes unseen because no receptor is tuned to it,” which is something that never arrives. The tickle and the unreceivable are opposites, yet both are filed under “full light.”

The equivocation isn’t fatal, and the repair comes from inside the corpus. The skin-and-bandwidth diagram is said to have two axes. The optical image has only one. Separate the axes and the scheme becomes a plane:

• what arrives and is cancelled (the umbra proper);
• what arrives and is partly cancelled (the penumbra);
• what arrives uncancelled (the tickle, surprise);
• what never arrives (beyond bandwidth).

On this plane the sleepwalk is a collapse from both sides. Cancellation spreads outward over everything that arrives, and gating closes inward over everything that might arrive. The penumbra is squeezed between them. This is more precise than the text’s one-dimensional account. It also explains something the text asserts without deriving: the sleepwalker has lost surprise as well as learning. A sleepwalker cannot be tickled. The essay states that the penumbra is “the conditions of encounter, not a property of either axis alone.” It is right, but its governing image keeps reducing the penumbra to one axis.

A smaller leak: the wound. Early on, the wound is “full light,” exposure “without gradient.” Later it has “infinite penumbral width.” These can’t both hold. If the penumbra is partial prediction, then where nothing is predicted there is no penumbra at all. The slip is instructive, though. It shows that a gradient needs anchors at both ends: no penumbra without umbra. So the umbra is not the enemy. The text’s resolution (“enough prediction to scaffold”) concedes this but doesn’t make it central. It deserves to be central, because it shows that the self, the fully modeled region, is a condition of learning and not only its limit. The sleepwalk is not too much self. It is self with no edge.

The self-concealment argument

The essay’s central formal claim is that “the sleepwalk is self-concealing… because the organ that would detect it is the organ it consumed.” Two mechanisms are fused here, and they need to be pulled apart.

1. Rate. The narrowing happens too slowly to register. This comes from the vibrato piece: the oscillation has dropped below bandwidth.
2. Organ. The detector itself has degraded.

Either one alone would conceal the sleepwalk, and they have different consequences. The rate mechanism implies that a fast narrowing, such as trauma, would be felt as a loss. It would not be a sleepwalk but a known wound. The text implicitly confines the sleepwalk to gradual erosion. I think that’s correct, but it should be said, because it separates the sleepwalk from defensive closure.

The organ mechanism faces an objection. A system could watch itself at a second order and notice that error signals have gone quiet. A sudden absence of surprise is itself a signal. But the second-order monitor has its own penumbra, which can narrow in the same way, and so on up the hierarchy. The argument survives as a claim about the highest level of self-monitoring a system has: whatever watches last has nothing watching it. That is a sound result, and it explains the conclusion that “the exit… cannot come from within.”

Strictly, though, the text contradicts itself, because the hiccup is internal. The hiccup comes from the relic essay and is carrying a lot here: it is the “pinhole in the carapace,” a residual somatic signal that gets through. If it is coherent, the claim has to be narrowed: the exit cannot come from the system’s representational monitoring, but it can come from a layer of the body that never fully joined the model. That is an interesting thesis, close to the phenomenological priority of the lived body over the represented body. But it needs an account of why the body’s penumbra narrows more slowly than the model’s. The essay asserts this asymmetry without arguing for it.

The false penumbra

The complication of the relic essay is the text’s subtlest move, and it can be pushed further. A tangent that periodically re-intersects its curve is binary: on the curve or off it. Sampled slowly, though, intermittent binary contact looks like a gradient. Engineers use exactly this, pulse-width modulation, to make on/off switching pass for a continuous value. It follows that the line between true and false penumbra depends on bandwidth. Only a system with fine enough temporal resolution can tell partial prediction from intermittent full prediction. The sleepwalker, whose bandwidth has narrowed, is therefore the system least able to tell its relic from a living gradient. The trap doesn’t just restate itself. It is cumulative.

Assessment

The essay’s main contribution is to treat partial knowledge as an organ with a width that can atrophy, and to give the atrophy a mechanism that explains why it goes unnoticed. It combines three traditions that rarely meet: the Socratic problem of inquiry, the dark-room problem in predictive accounts of mind, and ecological resilience theory. The combination gives each of them something it lacked.

Three things remain open:

1. The two operations. The optical image needs to be reconciled with the two-axis diagram it claims to extend. Gating and cancelling are different operations, and the sleepwalk is a collapse along both.
2. The hiccup. Why does it survive? That needs to be argued, not just asserted.
3. Width. The text says what the penumbra is wide in, but not how to measure it: what would distinguish a wide-penumbra system from one that merely reports being surprised often?

The third question matters most. Without an answer, “resilience is penumbral width” risks becoming unfalsifiable, a property identified only after the epoch shows who had it.