Someone told me that a classmate of his died partway through the year, in middle school. What he remembers, decades later, is that no one sat in the boy’s lunch chair. That’s the whole memory. Not the boy — the chair.
Nobody decided this. No teacher announced a mourning period, no group of twelve-year-olds voted to leave a gap. The seating arrangement simply kept a slot open that had nothing to put in it, and went on keeping it, the way a hand goes on itching after it’s gone.
That similarity is not decorative. It’s the same mechanism running in two substrates, and taking it seriously changes what grief is a problem of.
Most people who lose a limb continue to feel it. The standard account is that the brain maintains a body schema — a running model of where the parts are and what they’re doing — and the model doesn’t update just because the input stops. Sensation gets generated from the map rather than from the territory. The limb is gone; the representation of the limb is intact, still firing, still expecting return signals that never arrive.
A close relationship builds the same kind of thing. You run a predictive model of another mind: what they’d say about the news, which of your jokes will land, how they’d take the diagnosis, what the silence on the other end of the phone means. The model is expensive to build and it runs continuously, mostly below awareness. When the person dies, nothing deletes it. It keeps producing predictions into a void. Reaching for the phone is the same motion as reaching for a cup with a hand that isn’t there — an intact program executing against a missing world.
This gets the first thing right that ordinary grief talk gets wrong. We rank losses by closeness, and we tend to operationalize closeness as affection or contact. But cortical territory isn’t distributed by sentiment; hands and lips get vast representation and phantom sensation tracks that allocation. The relational analogue isn’t how much you loved someone or how often you saw them. It’s how much of your predictive machinery they occupy — how much of your day was structured around them, how many of your reflexes route through them. Which is why the chair works. Almost nothing personal was invested in that boy by most of the kids at the table. But he held a fixed position in a shared arrangement, and the arrangement bore weight. The absence outlived every actual memory of the person.
Note also where that phantom lived. Not in one skull. The deflationary reading is that twelve children each independently declined the seat and their choices happened to coincide — parallel individual models, no distributed anything. Two features resist that. Each child was tracking not the boy, whom most of them barely modeled, but the others: the slot was held in place by everyone’s model of everyone else’s expectation, a structure standing one level up from any of its carriers. And institutions do the same thing across complete personnel turnover, until the people maintaining the shape of a role have never met the person who shaped it. Still, I can’t exclude the possibility that this is ordinary norm-following with no phantom in it — that the chair is etiquette, not architecture. If it is, this paragraph is decoration and the rest of the essay survives without it.
Now the part I got wrong the first time, where the correction improves the argument.
The convenient version says phantom limbs resolve. The map corrects, the sensation fades, and no amputee has any interest in preserving it — whereas grief persists, so grief must be a different kind of thing. That contrast is false. Phantom sensation is close to universal after amputation, with lifetime prevalence around eighty percent and phantom pain in the same range, and for a large fraction it becomes chronic rather than transient. The mirror box, Ramachandran’s device for feeding the brain corrective visual evidence, produces effects that are real but modest and unevenly replicated; one home-delivered trial reported a median pain reduction around fifteen percent.
So the asymmetry isn’t between amputees and the bereaved. It’s one rule with two variables, and separating them is the argument.
The first variable is error scarcity. Because it is the simpler explanation, it goes first. A sensorimotor system is punished densely and immediately — reach for the cup, watch the hand not arrive, forty times a day, every day. A social model gets almost none of that. Nothing arrives to contradict your model of the dead. You can run them in imagination indefinitely at no cost, and silence from a person is ambiguous in a way that a hand closing on nothing is not. This alone explains why relational models decay so much more slowly than body maps, with no appeal to meaning, loyalty, or love. Where it works, prefer it.
It does not explain the valence. Error scarcity predicts slow updating; it does not predict that an offer to help you update is experienced as an insult. Nobody resents having their slow watch corrected. The moral charge on the vocabulary of closure — the specific way it lands as intrusion rather than as bad advice — needs the second variable, which is that the model is not experienced as a model of someone else. Social psychology has a literature on including the other in the self: close others get incorporated into the self-concept rather than merely attached to it. On that reading, correcting the map is not information about the world. It’s an instruction to become smaller. To stop reaching is to concede the reaching was unwarranted.
I wanted to corroborate that from the other substrate, where measurement is better, and I should report that I couldn’t. The tidy version wanted amputees who decline correction because the limb has become part of them — refusal on grounds of identity rather than discomfort. What the record contains is thinner and points the other way. In the one series I found that documents refusals, four of thirty-three patients abandoned mirror therapy, and the complaints logged across that group were confusion, dizziness, and unspecified irritation; the authors attributed the trouble to poor patient selection and to interference with prosthetic rehabilitation. Nobody in that record declined in order to keep the limb.
That is evidence against me, and it is weak evidence, because nobody was asking. A study cataloguing side effects will find side effects; it has no instrument for a patient who wants the phantom. So the claim stands as a prediction rather than a finding, and it’s the cleanest external test the argument has: ask amputees who abandon correction whether the limb is experienced as part of who they are. If refusal tracks only pain and inefficacy, the identity variable is doing no work in the substrate where it can be measured, and I’d take that as a serious blow to its use in the other. What the grief case supplies on its own is the valence — the fact that closure-talk offends rather than merely fails — and error scarcity can’t produce that. The prediction is unconfirmed. The asymmetry it explains is not in doubt.
This also revises what funerals are doing. The obvious reading makes them mirror boxes: sit with the body, see it, update. If the binding constraint were ignorance that would be right, and for the part of the block that’s error scarcity, it partly is — a body in a room is the densest disconfirming signal the social case ever gets. But it can’t be only that, because it doesn’t explain why a ritual is required, or why it’s collective. The second thing a funeral does is license: here is the sanctioned occasion on which you may keep reaching, and we will reach with you. Not correction, and not instead of correction. Permission to hold what the correction would take.
Now the second axis, which is routinely collapsed into the first.
Some grief has the texture of an unpaid bill. The waiter remembers the tab that’s still open and forgets tonight’s order the instant it’s settled — attention stays clamped on unfinished business and releases on completion. (The laboratory version of this has had a rougher time in replication than its ubiquity suggests, so lean on the phenomenology rather than the effect size.) It predicts the case that otherwise embarrasses the whole picture: grief after a conflicted relationship, where you’d expect less investment to mean less pain. The evidence points the other way — ambivalence and unresolved dependency predict worse outcomes. Not more machinery. More outstanding state.
But look at what makes the waiter’s metaphor work. The tab clears because payment arrives, an external event supplied by the counterparty. That is precisely what death forecloses. The loop can no longer be closed in the normal way, by the other person doing something. It can only be declared closed, unilaterally, by the one still holding it.
The strongest objection to splitting these is that they can’t be split, because the relationship is itself a standing open loop. What died wasn’t a set of items but a negotiated joint future, and no deathbed conversation completes that; so the resolved death still carries an unfinished thing, and there’s only one axis after all.
I think this equivocates, and the equivocation does all the work. Tension is about goal states with completion conditions — things that could in principle be finished, which is why they clamp attention until they are. “The shared future is gone” has no completion condition. Nothing would count as finishing it, and no act performed by anyone, including the dead, would have discharged it. That isn’t unfinished business. That’s a machine producing outputs with nowhere to send them, which is the first axis wearing the second one’s clothes. Widen unfinished until it covers that, and the model collapses to a single axis by definition rather than by evidence, which is a way of losing an argument rather than winning one.
The two terms differ in kind, not just in magnitude, and this is what makes them worth naming. Tension is propositional. There is a thing that didn’t get said, and it can be said — to a chair, into a journal, at a grave. Amplitude contains no proposition at all. It isn’t mispredicting an event; it’s mispredicting that the world will go on being shaped a certain way. Which is why its ambushes arrive through doorways, ringtones, the passenger seat, the second toothbrush. Those aren’t reminders. They’re cues for a motor program with no target left. You cannot say the unsaid thing to a doorway, and not because you lack the words. There is no unsaid thing.
The falsifier is clean. Take the death that is maximally close and fully resolved — everything said, sat with to the end, nothing outstanding. A single-axis model predicts this should be less grief. The two-axis model predicts grief of undiminished magnitude in a different key: missing rather than gnawing, an ache without torment. If people who got the deathbed conversation reliably report reduced intensity rather than altered texture, the axes are one axis and I’m wrong. My impression is that they report the second thing. That is an impression, not a finding, and it is the hinge the argument swings on.
Two further features of phantom limbs transfer, and they are not equally well earned.
Phantoms telescope — less often than the vivid descriptions suggest. Roughly a quarter of amputees report it, with estimates across large samples running from about fourteen percent to a third. In those who do, the felt limb shortens over months or years: the hand drifts inward, the arm collapses, until the fingers seem attached at the shoulder. The most articulate part survives; the connective structure dissolves.
This happens in grief and we have an uglier name for it: forgetting. Years on, what remains of the dead is a compact set of gestures — a laugh, a phrase, one afternoon — attached directly to you, with the person’s interior gone. You keep the hand. You lose the arm. What people are frightened of when they say they’re afraid of forgetting is not erasure of the memory, which mostly doesn’t happen; it’s this, the person contracting from a separate mind capable of surprising you into a small fixed ornament of your own.
And here is the part I’d rather not have found. There is at least one documented case in which phantom pain intensity fell as telescoping set in, and recent work relates telescoping to pain and mood measures in ways that are not simply adverse. The association is thin and the direction isn’t settled. But if it transfers, contraction is not the enemy of relief; contraction is what relief looks like from inside. The dead shrinking into a laugh and a phrase and one afternoon would not be the failure mode. It would be the mechanism by which the ache becomes survivable, which makes the fear of forgetting a fear of the treatment. That is a real prediction — grief intensity and model compression should be coupled rather than independent — and it is the one claim here I’d most like to be wrong about.
The other one earns its place more cleanly. Some people born without a limb report phantom sensation in it: at least one in five on the standard figures, alongside roughly half of those who lose a limb before the age of six. The deflationary reading is that these phantoms are assembled from watching other people’s bodies rather than arriving with the wiring, and that reading is live. But if any part of the schema comes without experience, a category of grief stops being anomalous: mourning a relationship that never existed. The absent father, the child not born, the sibling nobody had. Here the framework does work that attachment accounts can’t. If mourning is the undoing of a bond, grief with no prior bond to undo is a category error, and people report it anyway, and report it as grief rather than as disappointment. A schema-first account needs no occupant. The slot generates predictions on its own.
The strongest objection to the whole thing is that it’s too accommodating. A metaphor that fits the close death and the distant one, the resolved and the conflicted, the remembered and the never-had forecloses nothing — and grief research already has better-specified constructs that pay their way empirically without borrowing neurological glamour. That objection lands, and the answer is to narrow. The metaphor earns its keep in three places. It makes structural embedding rather than affection the driver of magnitude, which is not what the folk model says and is checkable against cases like the chair. It separates the rate of persistence from the valence of correction, which is what explains the felt violence of closure-talk, and which now rests on the grief evidence alone since the cross-substrate corroboration didn’t survive checking. And it predicts grief without a prior bond. Everywhere else it is a useful way of talking, and I’d concede the whole of it under pressure without losing the three.
One warning about what this is for, because the shape of the thing invites the error. A two-term diagnostic is nearly irresistible to convert into a treatment matrix: identify the axis, apply the matching tool. I want to disown that in advance. I have named one axis with a plausible tool and one axis with no tool at all. Anything that fills the empty cell is an intervention invented because the table looked unbalanced.
The empty cell is the real finding.
Unilateral closure is available for tension. That’s what Gestalt therapy’s empty chair is for — you address the chair, you say the thing that never got said, and the point is to discharge an open loop when no one is coming to settle it. Notice that this is the same object as the boy’s lunch chair, put to the opposite use: one preserves the slot, the other empties it, and both are treatments of the same wound. But nothing in the empty chair touches amplitude, and now we know why it can’t. Amplitude has no propositional content for a speech act to operate on. Saying the unsaid thing doesn’t reduce by one line the amount of you that was built to run predictions on a person who is gone. Most grief counsel, most of what friends offer, most of what the bereaved try on themselves, aims at the tension term — resolve it, forgive them, let it go, get closure — and a great deal of grief is amplitude, where none of that is even the right kind of operation.
What the distinction can offer isn’t a tool. It’s the thing the funeral was doing: permission. This one is not yours to finish. There is nothing here to resolve, and you have not failed to resolve it. That may be worth something even in the first month, when the person is drowning in advice built for the other axis and adding their failure to take it to everything else they’re carrying.
The chair was never a message. Nobody left it empty to honor anyone. It stayed empty because a structure that had a boy in it went on having a boy-shaped place in it, and that is not sentiment, it’s architecture — which is also why you cannot decide to stop.
Sources for the figures. Phantom sensation and pain prevalence, and the ~25% telescoping rate: Stankevicius et al., European Journal of Pain 25(1), 2021 — a systematic rapid review. Telescoping range of 14–33% across large samples, and its relation to pain and mood: Aternali et al., Frontiers in Pain Research, 2026. Pain falling as telescoping onset: case report, Frontiers in Pain Research, 2024. Mirror therapy refusals and side effects: Casale et al., on contraindications in lower-limb amputation (33 patients, 4 refusals). Home-delivered mirror therapy effect size: uncontrolled pilot, n=40, median 15.4% reduction at two months. Congenital phantoms at ~20%, and ~50% for amputation before age six: Melzack et al., 1997, as reported in the reviews above rather than read directly. Claims here without figures — the Zeigarnik replication record, ambivalence predicting worse bereavement outcomes, inclusion of the other in the self — are stated from general familiarity and were not checked for this draft.
cafebedouin.org 