Seeing Things

Why a psilocybin session, a sleepless night and a failing retina all produce the same images, and what that tells us about how every brain builds the world.

By Reid Robison, MD. Adapted from my Grand Rounds talk. 7,823 words, with nine things to try.

Three people, one Tuesday

Three brains. Three entirely different causes: a drug, a sleep debt, a diseased retina. And yet all three land on the same visual vocabulary. If you ask a room of physicians how many have had the second experience, most hands go up. This isn't exotic.

So here is the question this whole post is built around: why does a healthy brain, a tired brain and a damaged brain converge on the same images?

Because the hallucination isn't the content. It's the machinery, briefly visible.

Perception is not a recording. It's a construction: a running best guess, continuously checked against evidence. Break the checking in any of several places and the guess keeps running, unchecked. What you see then isn't noise. It's the shape of the apparatus. This is a post about hallucination that happens to include psychedelics, not the other way round.

Part 1What a hallucination actually is

The clinical use of the word comes from the French psychiatrist Jean-Étienne Esquirol, and his definition still works almost two centuries later:

A person who has the intimate conviction of a presently perceived sensation, when no external object capable of exciting that sensation is within reach of the senses, is in a state of hallucination.Jean-Étienne Esquirol, Des maladies mentales, 1838¹

Notice what he puts into it: not strangeness, not madness, not disease. Just two things, a percept and no object. The modern working definition adds a little precision: a perception-like experience that (a) happens without an appropriate external stimulus, (b) has the full force of a real perception, and (c) isn't under voluntary control.²

What the definition deliberately leaves out matters as much:

• Belief. You can hallucinate and know it. Most people who hallucinate do.
• Pathology. Nothing in the definition requires an illness.
• A particular sense. Every sense can hallucinate. This post leans on vision because it's the best mapped, not the most common. In schizophrenia, for example, hearing voices dominates.

The clinically decisive question is almost never "is this a hallucination?" It's "which of four neighbouring things is this?", because each has different causes, a different workup and a different conversation.

The illusion cell is the one people under-use. A coat on a door that is briefly a person is an illusion, not a hallucination, and in someone who might be delirious that distinction changes what you look for. And the boundary between imagery and hallucination is a gradient, not a wall, which matters later when we get to visual changes that persist after psychedelics.

Not madness: a property of nervous systems

Much of this post rests on the frame Oliver Sacks set in his 2012 book Hallucinations.⁴ Across fifteen chapters of cases (blind people seeing crowds, deaf people hearing orchestras, the bereaved seeing the dead, migraineurs watching zigzag patterns assemble, the dying and the half-asleep), his argument is that hallucination belongs to ordinary neurology, not the asylum. The great majority of people who hallucinate are not psychotic, not demented, and not in danger. Many say nothing for years because they assume otherwise. Sacks was no detached observer, either: he wrote frankly about his own amphetamine and psychedelic years, including chasing a particular shade of indigo he never found again.

How common is it?

Imagery at the edge of sleep is close to universal when people are actually asked. Complex visual hallucinations occur in perhaps 10 to 15% of people with substantial vision loss. Grief, the ICU and extreme isolation all produce hallucinations often enough that their absence from medical teaching is the strange thing. The old rule that visual hallucinations mean "organic disease until proven otherwise" is wrong as stated. What survives is a weaker version: prominent visual hallucinations with a clear head and no voices should push you to look harder at neurology and drugs.

Part 2Perception is already a guess

This part is the spine of everything that follows. It needs no math, only two dials.

The psychic activities that lead us to infer that there in front of us at a certain place there is a certain object of a certain character, are generally not conscious activities, but unconscious ones.Hermann von Helmholtz, Treatise on Physiological Optics, 1867⁷

Your retina receives a flat, noisy, ambiguous, partly hidden and constantly moving sample of the world. What you experience is a stable three-dimensional scene. The gap between the two is filled by inference. The brain runs a model of what kinds of things are out there and how they behave, and constantly reconciles it with the evidence arriving from the senses. What you perceive is the reconciliation, not the evidence.

The key word is precision, meaning confidence. Each side carries a weight, and the percept is pulled toward whichever side the brain currently trusts more. That weighting is itself set by brain chemistry, which is why drugs, sleep loss and disease can all move it. That single idea is the whole model. Try it.

Four moves are worth making. Start balanced: the percept tracks the evidence and the world looks ordinary, which is the state you're in right now, and it is still inference. Drop sensory trust to the floor and leave the prediction alone: that's Charles Bonnet syndrome, phantom limbs, sensory deprivation. Nothing was added; something was subtracted. Put sensory trust back and push the prediction up: that's the strong-prior account of psychosis, where the percept detaches even though the evidence is fine. Then drop both: that's severe sleep loss and delirium, where nothing is trustworthy and the system wobbles.

Two completely different lesions produce the same symptom, because the symptom is a ratio, not a location.

Where each condition sits

The map uses the same two dials as the sandbox: trust in the senses runs across, trust in the prediction runs up. Ordinary perception sits near the middle of the dashed diagonal, where the two are balanced.

• Evidence starved (far left): an eye, ear or limb stops reporting, and the prediction runs unopposed onto a silent channel. Charles Bonnet, musical hallucinations with hearing loss, phantom limbs, sensory deprivation. Insight is usually kept.
• Prior dominant (top left): the evidence is still arriving, but the model overrides it. The strong-prior account of schizophrenia, conditioned hallucinations, stimulant psychosis. Insight is usually lost.
• Both degraded (bottom left): delirium, deliriant drugs, severe sleep loss, withdrawal. Fluctuating and disoriented. This is the dangerous corner.
• Senses up, beliefs loosened (bottom right): the psychedelic corner, and the contested one. Sensory gain rises while high-level beliefs loosen, which is why the experience is so rich and insight so often intact. That placement is a synthesis, not a settled finding, and I lay out the disagreement in Part 4.

The insight point deserves its own sentence. Whether a person believes their hallucination tracks where the fault is. A failed input channel leaves the rest of the model intact to say "that can't be right." An overweighted model has nothing left to check it against.

You can build a hallucination in an afternoon

In 2017, Albert Powers, Christoph Mathys and Philip Corlett at Yale paired a checkerboard with a faint tone, over and over, then started dropping the tone.⁹ They tested four groups of about fifteen people each: healthy controls, healthy voice-hearers (self-identified psychics with no illness), people with psychosis who don't hear voices, and people with psychosis who do. Everyone started "hearing" the absent tone. But the people who hear voices did so far more often, whether or not they had a diagnosis. The effect tracked hallucination, not illness. Computational modeling put it down to overweighted perceptual priors, with signals in auditory cortex, the insula and the cerebellum.

That experiment turned a metaphor into a measurement. And the idea is older than it looks: in 1895, Carl Seashore at Yale induced hallucinations of touch, sight and taste in healthy volunteers using learned cues and plain suggestion.¹⁰ The idea is 130 years old; the model is new. In fairness, the replication literature on the exact task is mixed on some details, and researchers still argue about whether people lean on priors as a standing trait or respond to manipulated expectation.¹¹ Neither reading rescues the simple equation "hallucination = psychosis."

Part 3Why the patterns are always the same patterns

This is the most beautiful part of the story, and the part most people have never seen.

Heinrich Klüver spent the 1920s taking mescaline and reading everyone else's accounts of it. He found that the simple imagery, across people and decades, kept collapsing into four recurring types.¹² He called them form constants: tunnels and funnels, spirals, lattices and honeycombs, and cobwebs. The same four families turn up after mescaline, LSD, psilocybin and cannabis; in migraine aura; under flicker; with pressure on the eyeball; at the edge of sleep; in occipital seizures; in near-death reports; and in Charles Bonnet syndrome.

To see form constants and other perceptual changes layered over an ordinary scene, try the perception lab.

The logic is the same as tapping very different objects and hearing them all ring at one frequency: it tells you about the room, not the hammers. If wildly different causes produce the same small set of shapes, the shapes aren't coming from the cause. They're coming from the thing being disturbed.

Your cortex, seen from the inside

Here is why. The wiring from the retina to primary visual cortex (V1) follows, roughly, a logarithmic map. Concentric circles in your visual field land as vertical stripes on the cortex; spokes radiating from the center land as horizontal stripes. Run that map backwards and any stripe pattern on the cortex becomes a form constant in what you see.^13,14

The person isn't seeing a tunnel. They're seeing their own primary visual cortex, in stripes, transformed by the map that put it there.

Where the stripes come from

1. An excitable sheet.V1 is a two-dimensional sheet of excitatory and inhibitory neurons, with local excitation and broader inhibition around it. At rest it sits in an even, low-activity state.
2. Turn up the gain.Raise excitability (with drugs, flicker, the wave of a migraine, or by removing input) and the even state becomes unstable. It breaks spontaneously into stripes, hexagons and other periodic patterns, a so-called Turing instability in neural tissue.¹³
3. Read it out through the map.Invert the retina-to-cortex map and the stripes become form constants. Paul Bressloff and colleagues added the cortex's orientation-tuned columns and long-range connections, which let the model produce lattices and cobwebs too, the two types the original model couldn't.^14,15

The pattern is set by the same lateral wiring that normally processes edges, contours and textures. Geometric hallucination is that machinery idling with the gain turned up. A local note: Bressloff did this work in the mathematics department at the University of Utah. The best formal account of what people see on psychedelics was written a few miles up the hill from me.

Eight doors into the same room

Part 4Nine mechanisms, one phenomenon

Drugs, sleep loss, lost input, neurological disease, grief and schizophrenia, all read through the same two dials. To keep this from becoming a list of diseases, every case gets the same six questions. The answers are what separate them.

Question three is worth dwelling on. A person with Charles Bonnet syndrome may have far more elaborate imagery than a person with schizophrenia and know perfectly well it's unreal, because their reality-testing is intact. It's the input channel that failed. Insight tracks where the fault is, not how big it is.

Classic psychedelics: three models that don't agree

These may not be rivals. They may describe different levels of the hierarchy. Loosen the abstract beliefs at the top (the self-model, the story, what you think is possible) while turning up the low-level pattern-making machinery at the bottom, and you get geometry and ego dissolution in the same hour, which is exactly what people report. I'm stating that as a hypothesis, because the field hasn't settled it.

Dissociatives: a different disinhibition

Ketamine, DXM, PCP and nitrous oxide preferentially block NMDA receptors on the fast-firing interneurons that act as the brake on cortical output. Take off the brake and the cortex disinhibits. More interestingly, the error signal that tells the brain when to update its model degrades, so predictions stop being corrected and the world stops feeling like it belongs to you.²⁰ Where psychedelics add, dissociatives subtract: disconnection, unreality, the body receding, featureless space, broken time. That's why ketamine has long been the more convincing drug model of schizophrenia's negative and disorganized symptoms.^21,22 Having sat with several thousand ketamine sessions, I can tell you the two states separate cleanly at the bedside: with psychedelics the visual field is alive and emotion is present; with dissociatives the field thins, the body is remote and feeling goes flat. Yet in clinical trials the two get coded into the same handful of terms. More on that in Part 6.

Deliriants: the class where insight goes

Datura, scopolamine and high doses of diphenhydramine (Benadryl) block acetylcholine, the signal that, among other things, tells cortex to attend to the input. Remove it and the internal generator wins completely. These are true hallucinations without insight: people hold conversations with people who aren't there and smoke cigarettes that don't exist, with a clouded mind, amnesia afterwards and a classic toxic syndrome (hot, dry, flushed, dilated pupils). This is delirium, not a psychedelic state, and it is dangerous. The same mechanism explains a very common, preventable cause of hallucination in older adults in hospital: the cumulative anticholinergic load of their medication list. Check it before calling psychiatry. And the mechanism runs in reverse, too: cholinesterase inhibitors reduce visual hallucinations in Lewy body dementia. Folk usage lumps datura and psilocybin together as "hallucinogens," but at the level of circuit, experience, insight, danger and aftermath they have almost nothing in common.

Stimulants and withdrawal

With methamphetamine and cocaine, repeated dopamine surges lead the brain to tag neutral events as urgently significant ("aberrant salience"), and the person builds an explanation to fit the feeling.²³ Paranoia comes first and outlasts the perceptions. The mind stays clear. Voices and touch dominate, including the classic sense of insects under the skin, which often sends people to dermatology first.

Alcohol and sedative withdrawal give the best teaching pair in this whole post. Alcoholic hallucinosis: clear-headed, oriented, mostly voices, vital signs near normal, often mistaken for a primary psychosis. Delirium tremens: clouded, disoriented, tremulous, mostly visual and tactile, with unstable vital signs. It's a medical emergency with real mortality. Same drug, same withdrawal, two syndromes that differ on exactly question four.

Sleep loss: the one everyone has done to themselves

A systematic review of 21 studies and 760 people kept awake from 24 hours to 11 nights found symptoms appearing in a nearly fixed sequence.²⁴ Vision was affected in 90% of studies, the body in 52% and hearing in 33%; by day three, all three. It's a dose-response curve for psychosis in healthy people, with no receptor touched.

What changes is the precision of everything: dream sleep pressure intruding on waking, patches of cortex briefly going offline while you're awake, and attention control failing. Both dials drop. Most people fully recover after sleeping roughly half as long as they were deprived, though recovery was slower in people with pre-existing mental health conditions. And an ethics footnote: keeping people awake past 48 hours is no longer considered acceptable research, so we know this trajectory from experiments we would not run today.

Lost input: the cleanest natural experiment we have

In release phenomena nothing is added to the brain. Something is subtracted, and the hallucination is what's left when the sensory channel stops arbitrating. In Charles Bonnet syndrome, vision loss from any cause leads to complex, often silent, often repetitive imagery with insight kept. Dominic ffytche's imaging work showed the content maps onto whichever specialized region lost its input: faces with the face area, colored patterns with the color area, text with the word area.²⁵ The hallucination tells you which cortical specialist lost its day job. The same principle gives musical hallucinations with hearing loss (usually familiar tunes, often hymns), phantom limbs, and the "prisoner's cinema" that appears within minutes in a featureless visual field.

If you remember one thing from this post, make it this script for the person in front of you:

People carry this secret for years. Sacks opens his book with exactly that patient: a woman who assumed her visions meant she was going mad, and told no one.⁴ Where eye clinics screen for it routinely, far more cases come to light.

Five neurological signatures

• Migraine aura. A slowly marching shimmering blind spot with zigzag edges over 20 to 30 minutes. The march is the giveaway. (Sacks's first book was on migraine.)
• Occipital seizures. Seconds to a couple of minutes, often colored circles or spots, the same picture every time. Temporal-lobe seizures give formed, dreamlike, multi-sense experiences instead.
• Peduncular hallucinosis. A brainstem lesion. Vivid, often tiny people or animals, usually in the evening, usually with insight kept. Often missed for months.
• Parkinson's and Lewy body disease. A progression: a sense that someone is there, then something passing at the edge of vision, then formed figures, animals or children. Evenings. Insight erodes over time. Families often report "he keeps thinking someone's in the room," which gets recorded as paranoia and treated with an antipsychotic, which in Lewy body disease can be catastrophic.
• Narcolepsy and sleep paralysis. Dream machinery running while the cortex is awake enough to notice: the intruder, the pressure on the chest, the "night hag" found across cultures.

The pattern: location predicts experience. Early visual cortex gives geometry, association cortex gives faces and scenes, the brainstem gives small formed figures with insight, and the temporal lobe gives meaning-saturated experience. The content is a map.

The haunted mind: grief

In W. Dewi Rees's classic 1971 study, nearly half of widowed people had sensed, seen or heard their late spouse, most often as a felt presence, sometimes for years, and most had told no one, including their doctor.⁶ The two-dial model makes this concrete. A lifetime of the most heavily weighted predictions you own (this face, this voice, this footstep, that car in the driveway) suddenly has no matching evidence. The model doesn't update on the day of the funeral. It runs, and it fills. That's why it's almost always that person, in their chair, at their hour, and why it's most common in the first year. It's lost input of a relationship.

Normalize it, ask about it, and keep concern for content that is distressing or commanding or comes with other features. A good way to ask: "A lot of people, after a loss like this, catch a glimpse or hear a voice. Has anything like that happened?"

Schizophrenia, and why "psychotomimetic" was a bad word

In two-dial terms, schizophrenia looks like overweighted high-level priors with reduced trust in sensory evidence. Add a failure to tag your own inner speech as your own: normally, when you speak, the motor system warns auditory cortex "this one's ours" and the response is damped. In schizophrenia that damping is reduced, most in people who hear voices.^26,27 Inner speech arrives unflagged, and the system reasonably concludes it came from somewhere else. Add dopamine-driven aberrant salience and the delusion becomes the explanation built over the feeling.²³

The point isn't that psychedelics are harmless or that psychosis is like a trip. It's that the surface similarity was over-read. The 1950s "model psychosis" framing set the field back decades, the better drug model of schizophrenia was always the NMDA one, and the residue of that confusion is still in our vocabulary.²⁸

Filter to "insight kept" and Charles Bonnet, musical hallucinations, peduncular hallucinosis, migraine, grief and psychedelics appear together: six conditions no textbook groups, grouped because insight is a mechanism-level fact. Filter to "clouded" and you get deliriants, delirium tremens and severe sleep loss: the emergency column. The table has more explanatory structure than the diagnostic categories, because its columns are mechanisms and the diagnoses are history.

Part 5When it doesn't stop: HPPD and visual snow

There is now a full post on this: When the trip doesn't end, with the neuroscience, the 2025–2026 numbers and the treatment evidence.

Hallucinogen persisting perception disorder (HPPD) is when someone, after stopping a hallucinogen, keeps re-experiencing perceptual changes that were present during the drug effect.^29–31 It was called post-hallucinogen perception disorder in DSM-III and HPPD from DSM-IV. The closest general medical term is palinopsia, Greek for "seeing again." It comes in two forms that most online discussion conflates, which is why reported severity is so split:

Across roughly fifty papers and case series it's been reported after LSD, psilocybin, mescaline, ayahuasca, DMT, salvia, iboga, datura, ketamine, DXM, MDMA, cannabis and synthetic cannabinoids, and in one case on risperidone. One review frames HPPD as subtle over-activation of visual pathways that worsens with anxiety in predisposed people after arousal-altering drugs, including non-hallucinogens.³¹ If that's right, the name is wrong: it points at a drug class rather than a mechanism, and it has shaped what we look for ever since.

What's actually happening

• Loss of visual filtering. The leading account: the inhibitory filtering that keeps low-level visual activity out of awareness doesn't fully reset.³²
• Cortical hyperexcitability. Consistent with abnormal visual evoked potentials, and the rationale for trials of brain stimulation and anticonvulsants.
• A learned percept (my hypothesis). In two-dial terms, a prior was written during a period of extremely high plasticity and now has enough confidence to generate a percept on its own. This predicts things the other accounts don't: that context and arousal should trigger it, not just dose; that it should be more common in people with prior perceptual sensitivity; and that paying attention to it should keep it going. All three match the clinical picture. The evidence is indirect.

There's a clinical parallel worth naming. Persistent postural-perceptual dizziness (PPPD) is the balance-system version of exactly this story: an acute insult, then a maladaptive perceptual habit that outlives it. It responds to habituation and attention-directed therapy. Nobody has run that trial in HPPD. If HPPD is a maintained perceptual prior rather than a fixed lesion, the literature we've been ignoring is vestibular rehabilitation, not anticonvulsants.

What people report, and what the criteria list

The DSM lists flashes of color, trails, halos, geometric figures, motion changes at the edge of vision, things looking too small or too large, and auras. People also report floaters, visual snow, pattern repetition, unusually vivid color and sharpness, seeing faces in things, superimposed geometry, fragmented images, distorted distance, a color-filtered field, synaesthesia and image-like flashes. That's eleven phenomena people describe that aren't in the criteria. If a study's case report form uses the DSM list, those eleven aren't recorded as zeros. They're silence.

Then there's visual snow syndrome, which now has its own criteria in the headache classification's appendix: continuous tiny flickering dots across the whole visual field for more than three months, plus at least two of afterimages, enhanced floaters and other "entoptic" phenomena, light sensitivity and poor night vision.³⁴ It's strongly linked to migraine, and most people with it have never taken a hallucinogen. So either HPPD is a drug-triggered case of a broader visual-processing trait that also arises on its own, or they're separate conditions that look alike. The literature can't yet say, partly because neurology and psychiatry study them with different instruments. That's a naming problem, not a biology problem.

Who gets it, what sets it off, what helps

HPPD has been linked to depressive and anxious traits, bipolar disorder and psychotic illness.³¹ In one exploratory series of 20 persistent cases, most had dissociative symptoms and negative mood, and many had perceptual symptoms before any hallucinogen use, which does real damage to a simple drug-causes-it story.³³ Reported triggers are stress and arousal, cannabis or further psychedelic use, dark environments, sexual arousal, fatigue and sleep loss, and cues linked to the original experience. Every one is an arousal or salience trigger, not a pharmacological one. That is itself a clue.

Treatment evidence is thin, all small open-label series and case reports: clonidine, clonazepam (with real dependence risk), low-dose older antipsychotics (mixed, and one case of HPPD emerging on risperidone), lamotrigine or levetiracetam, and low-frequency rTMS.³¹ Nothing here would survive regulatory review. What I find most consistently useful is reducing arousal and reducing attention to the phenomenon, and that's the least studied of all.

Part 6What do we call this?

Parts 1 to 5 were neuroscience. This part is about measurement, and it's the one that changes what researchers do on Monday. In this field, language is the instrument: the resolution of a study is capped by the resolution of its vocabulary.

Hour four

"I understand now that I have been treating my own life as a rehearsal."A trial participant, hour four, eyeshades briefly off. Then she put them back on.

I'm the physician on that study. Within the next day, my job is to open the case report form and decide whether anything happened in that session that needs recording, and if so, under what name. This isn't a philosophical problem. It's a data-entry problem. There's a field, and the field has a controlled vocabulary. I can't type what she said. I have to pick from a list.

There's no term for what just happened. There's no code for "understood something structural about her own life under a serotonergic drug." What the dictionary I'm required to use does offer is euphoric mood, abnormal thinking, dissociation, and a whole family of psychiatric disorder terms. Whichever I pick becomes the permanent record. It pools across sites and trials and eventually becomes a line in a regulatory submission: "In the treatment arm, X% of participants experienced abnormal thinking." That sentence will be true. It will also be one of the least informative sentences ever written about what happened in that room.

Naming is not decoration

Every number in medicine is a claim about a category. If the category is badly drawn, the count isn't noisy. It's meaningless. Imagine a category called "chest discomfort" used to count both heartburn and heart attacks. That dataset could never tell you anything about heart attacks, however many patients you enrol. No one would do that in cardiology. We do it constantly with consciousness:

• Anxiety covers both the participant worried about the IV and the participant spending ninety minutes confronting her own death.
• Hallucination covers both a shimmer at the edge of vision and a sustained encounter with what seems to be an autonomous intelligence.
• Dissociation covers both a mild sense of unreality and complete dissolution of the boundary between self and world.

These aren't degrees of one thing. They're different things wearing one label. And here's the part that should worry us: coarse categories make summary statistics look better (more events per category, tighter confidence intervals, cleaner tables). Bad taxonomy is statistically flattering.

What happens to the words

Adverse events in trials are coded with MedDRA, the Medical Dictionary for Regulatory Activities. Here's one verbatim on its way through:

Standardization is a real achievement. MedDRA is why a regulator can pool a safety signal across fifty trials on three continents, and nothing here argues against having a dictionary. But it is also a compression. Two trained coders disagree on the preferred term in about 12% of cases, and about 8% of codes deviate from expert judgment on review.³⁵ MedDRA holds no formal definitions of the events it names. Coding isn't mechanical, either: in the adolescent paroxetine trial known as Study 329, suicidality-related events were coded as "emotional lability."³⁶ The dictionary didn't do that; people did, using the dictionary. And specific to psychedelic research, a 2025 analysis presented to the CNS trials society ISCTM flagged systematic under-reporting of psychoactive effects, because investigators and sponsors don't regard them as adverse, even though FDA guidance on abuse potential expects them to be captured, and FDA’s 2023 draft guidance on psychedelic trials asks for careful monitoring of these effects.^37,50

When I do this live with rooms of experienced investigators, they disagree on almost every item. That disagreement is the finding.

Excellent altimeters, no map

We do have good instruments. The altered-states scales (5D-ASC and its 11-factor version) separate simple imagery, complex imagery and synaesthesia.³⁸ The Hallucinogen Rating Scale has six clusters.³⁹ There are scales for mystical experience (MEQ30), challenging experience (CEQ), emotional breakthrough (EBI) and psychological insight (PIQ), plus dissociation and ego-dissolution scales.^40–43 But they're outcome measures, not a taxonomy. Each returns a score on a dimension; none returns a name for a phenomenon. A participant who scores 78 on the MEQ30 and 41 on the CEQ has told you the altitude. You still don't know what she saw, whether she knew it wasn't there, whether it addressed her, whether it came back at week four, or whether the distressing hour was the therapeutic one.

The frustrating part is that the mapping tools exist, in a different building. The International Consortium on Hallucination Research spent a decade on exactly this problem across diagnoses. The Questionnaire for Psychotic Experiences characterizes hallucinations by experience rather than diagnosis.⁴⁴ PSYRATS rates them on eleven dimensions, from frequency to controllability.⁴⁵ The North East Visual Hallucinations Interview was built for eye disease and neurodegeneration.⁴⁶ They'd need adapting (the QPE's psychosis framing doesn't sit comfortably in a psilocybin protocol), but the alternative right now is a dropdown. And an interview like this runs alongside MedDRA coding, preserving the structure the coding step throws away.

Why mild, moderate, severe is the wrong ruler

Three boxes work well for a headache. A mild and a severe headache are the same thing at different volumes. Now try geometry. At the low end: faint structure behind closed eyes, gone if you look straight at it. At the high end: an intricate self-transforming architecture that seems to have more dimensions than the room, which the person later calls more real than the room. That isn't a louder version of the first. It's a different thing. Effects behave in three ways:

• Continuous: heart rate, pupil size, nausea. The ruler works.
• Emergent: an encounter with an apparently autonomous being. There's no 10% of that. It has a threshold and arrives fully formed.
• Non-monotonic: detailed visual hallucination rises, peaks and then disappears at the highest doses, because the capacity to form and remember structured experience is itself failing.

So at the very top of the range people often come back with almost nothing to report, and on a three-box scale the most intense experience a person can have is recorded as the least eventful. That doesn't lose precision. It inverts the order. The fix isn't exotic: stop asking for an intensity rating and ask which effects were present. Did geometry appear? Two- or three-dimensional? Within the visual field or replacing it? Was there an apparent other being? Did the sense of self stay intact? The pattern of yeses gives you the level, an approach with precedent in Shulgin's rating scale.⁴⁷

Adverse to whom?

A participant spends about an hour in acute distress: crying, frightened, at one point saying she wants it to stop. The monitor sits with her. It passes. Two days later she calls it the most important thing that's happened to her in twenty years. At six months she's in remission. Did an adverse event occur? The formal definition is any untoward medical occurrence in a participant given a product, whether or not related to it.⁵¹ It doesn't require causation, or that the event be bad for the person on balance. "Untoward" does enormous work, and it's nowhere defined.

I want to be careful here, because "it was supposed to hurt" is the oldest bad argument in medicine and can be used to excuse anything. The narrow claim is this: two participants can both be coded "anxiety, severe." One spent ninety minutes oriented and engaged, confronting an unprocessed loss, and found it coherent and valuable afterwards. The other was disoriented, couldn't be reassured, tried to leave, and was left with no framework to make sense of it. Same row, same contribution to the safety profile. One tells you how the treatment works and the other tells you how it fails, and the dataset has thrown the difference away.^48,49

It fails in the other direction too, and this is the half that protects participants: a persistent sense that the world is subtly unreal, a new metaphysical conviction that starts reorganizing relationships and finances, visual changes that outlast the acute window. These are harder to code than a fast heart rate, so they drift to the nearest generic term, or to nothing.

What I'd build

None of this requires a new validated instrument or a regulatory negotiation. It's an add-on module for the case report form, and a scheduled follow-up question.

1. Keep the verbatim.Coding is lossy and permanent. The participant's words are the only place lost information can be recovered from, so they need to be stored in a way someone can search later.
2. Ask, don't wait.Persistent perceptual change is missed mainly because no one schedules the question. Ask at day 30 and day 90. That alone would give us an HPPD incidence figure worth having.
3. Separate distress from harm at entry.Record feeling at onset and at integration as separate fields, so analysis can tell them apart later.
4. Inventory, not amplitude.For anything emergent, the presence or absence of nameable phenomena beats a three-box severity rating.

This is additive. We can't stop coding to MedDRA, and shouldn't. We should stop only coding to MedDRA.

Part 7Monday morning

For clinicians, here's the differential in five questions, in this order.

1. Is the person clear-headed?Clouded, fluctuating or disoriented means delirium until proven otherwise: anticholinergic load, withdrawal, infection, metabolic causes, low oxygen, medication review. That's a time-critical medical workup, not a psychiatric one.
2. Which sense, and is it formed?Simple visual phenomena point to the early visual pathway (eye disease, migraine, occipital seizures, drugs), so examine the eyes. Formed visual images with insight point to Charles Bonnet, peduncular hallucinosis or Lewy body disease. Personally referential voices point to a primary psychiatric illness.
3. What's the time course?Seconds and stereotyped: seizure. Twenty minutes with a march: migraine. Evenings: Lewy body or peduncular. Hours, tied to a dose: drug. Months of constant background: visual snow, HPPD or lost input.
4. What's the sensory status?Formally check vision and hearing. Lost input is the single most under-diagnosed cause of hallucination in medicine, and the diagnosis is often made with an audiogram rather than a psychiatric interview.
5. What's on the medication and substance list?Anticholinergic load, dopamine agonists, recent benzodiazepine or alcohol cessation, stimulants, steroids. And sleep: ask how many hours, then ask again, because people under-report.

The conversation

Ask in a way that's safe to answer. Permission has to come before the question, or the honest answer is too expensive to give:

Explain the mechanism plainly. People accept a mechanism far more readily than reassurance, because a mechanism tells them the world still makes sense. Then name the boundary honestly: say what would concern you, such as voices that address or command them, content that follows them from place to place, losing the sense that it isn't real, or new confusion. A person told what to watch for reports it early. A person told "don't worry about it" reports it never.

Back to Tuesday

Three lesions. One machine. One catalog of shapes. Hallucination isn't a malfunction of perception. It's perception running with one of its checks removed, and it's the best window we have onto how the whole thing is assembled.

We can now say a great deal about the machinery. We still can't reliably write down what happened to a person, and until we can, our datasets will keep looking complete while answering the wrong questions. Better instruments won't fix that. Better words will.

This post is educational and isn't a substitute for an assessment by a licensed clinician. If you or someone you care about is having new or distressing hallucinations, please talk with a doctor.