Divided Consciousness

Frontier If cutting one bundle of axons divides a mind — which is the textbook reading, and which two complete-callosotomy patients gave direct evidence against in 2017, with 7 naming what would settle it — then the unity of consciousness has a physical price a surgeon can name, charge and revoke. That is the prize on this page. Callosotomy is the only procedure in which the property philosophers call the unity of consciousness has been experimentally manipulated in a living human being rather than inferred, modelled or argued about. Every other route to the question is a correlation.

Established The magnitude is best read off two numbers from the same set of sessions. In the study Pinto and colleagues published, comparing two stimuli straddling the midline ran at 55.4% correct, not statistically better than chance, while the same comparison inside one half-field ran at 90.4%. The perceptual division is real and it is large. In those same sessions the interaction between response channel — left hand, right hand, voice — and stimulus half-field, the interaction on which the two-conscious-agents reading rests, did not approach significance anywhere: all P > 0.5 in the detection task. Two unbridged perceptual streams, and one controller that answers for both.

Frontier Who collects, and who pays. The gain is almost entirely to theory: an adjudication here constrains every account of consciousness that makes integration across a substrate do the explanatory work, and it is the only such constraint available from a human intervention. The cost falls on patients — a population the field itself describes as rare and shrinking, ageing, with drug-resistant epilepsy and surgeries that differ in extent, era and age at operation. The people who supply the evidence get nothing from it, which is unusual even for basic research and is why 11 is not a formality here.

Speculative The condition on all of it is the one 5 names as binding: the entire dispute rests on a handful of people, and the largest recent imaging study of the group scanned six. A null interaction in two patients is not evidence of unity, and a cross-cueing story that can absorb any result is not evidence of division. The honest upside is therefore not a discovery but a closure: a chance, inside a window measured in years rather than decades, to convert a fifty-year argument into a result both camps agreed in advance to accept.

Established Corpus callosotomy severs the largest fibre bundle joining the two cerebral hemispheres, and it is done to stop seizures, not to answer questions about the mind. The operation reached modern practice through the Rochester series of van Wagenen and Herren around 1940 and the California series of Bogen and Vogel in the early 1960s; the second of these produced the patients Roger Sperry and Michael Gazzaniga tested, and the tests produced the disconnection syndrome that every introductory course now teaches.

Established The classical demonstration is a lateralised flash and a crossed response. A stimulus shown briefly in one visual half-field reaches only the opposite hemisphere, because the optic pathway is crossed and, with the callosum gone, nothing carries it over. The reported result was a dissociation: a patient would name a word flashed to the right half-field but deny seeing one flashed to the left, while the left hand — controlled by the right hemisphere — retrieved the matching object.

Frontier From that dissociation the field drew a conclusion the dissociation does not by itself license: that there are now two conscious subjects. The inference has always required a bridging premise — that a hemisphere which can perceive, discriminate and act, but cannot speak, is thereby a subject of experience. That premise is not measured anywhere. It is the load-bearing step, and it is where this brief spends its flags.

Frontier Since 2017 the empirical base of the classical reading has been directly contested. Pinto and colleagues reported that two patients with radiologically confirmed complete callosotomy answered accurately about stimuli anywhere in the field with either hand or by voice, which the two-agents account says they should not be able to do. Gazzaniga’s group replied that undetectably small eye and face movements — cross-cueing — can carry exactly the low-bandwidth information the task needs, so the result is compatible with two streams after all.

Established The surgery is an epilepsy treatment with a measured indication, which constrains who is ever available to study. It is offered for drug-resistant generalised seizures, most often to stop atonic drop attacks, and the modern version is usually partial rather than complete. Speculative That clinical fact, rather than any scientific decision, is what fixes the sample size of the entire literature.

Speculative What is at stake is narrower than the popular framing and harder than it looks. Nobody disputes that perception is divided in these patients; Pinto’s own data show it. The dispute is whether divided perception plus preserved unified control is one subject with two input streams, two subjects with a thin channel between them, or a case the concept of “a subject” does not cleanly describe. 3 is the measured record; 15 deals with the claims that have outrun it.

Established Start with what is not in dispute: after complete callosotomy, information presented to one hemisphere does not reach the other in the ordinary way, and the deficits this produces are stable, large and replicated. Fabri and colleagues at Ancona measured tactile naming in three totally callosotomised patients and three partially callosotomised controls: mean correct responses in the complete group were 93% with the right hand and 30% with the left, while the partial group reached 93–100%. The same study places the critical structure in the posterior callosum. This is the disconnection syndrome as a measurement rather than an anecdote, and no camp in the current dispute questions it. Established The 63-percentage-point gap between complete and partial section in that study is also the clearest evidence that surgical extent, not diagnosis, is the variable that matters.

Established The motor dissociations are equally robust. Lausberg and colleagues tested three patients with complete callosotomy against five with partial callosotomy and nine healthy participants, and found left-hand apraxia not only to verbal command but to visual presentation of the object, implicating a failure to associate perceptual object representations with movement concepts in the disconnected right hemisphere. Tanaka and colleagues characterised diagonistic dyspraxia — the left hand undoing what the right hand does, dissociated from stated intent — and localised the critical damage to the ventral part of the posterior body of the callosum, proposing a disconnection of the right superior parietal lobule from the left.

Established Hemispheric specialisation in these patients is measured, not assumed. Gazzaniga and Hillyard reported in 1971 that in brain-bisected patients “little or no syntactic capability exists in the right hemisphere”, with comprehension limited to an affirmative-negative dimension in visual-verbal matching, and could not confirm right-hemisphere speech production. Fifty-five years on, that finding stands. The popular extrapolation from it does not; 15 takes that up.

Frontier The 2017 result is the anchor of the current dispute and it survives checking. Pinto, Neville, Otten, Corballis, Lamme, de Haan, Foschi and Fabri tested two patients, DDC and DDV, both with radiologically confirmed complete section of the callosum, normal IQ and no other brain damage. DDC was operated at 19 and tested 10 to 16 years later; DDV was operated at 22 and tested 17 to 23 years later. In a detection-and-localisation task with 120 ms presentations and 50% blank trials, DDV scored 100% hits with 0% false alarms and a mean localisation error of 2.8°; DDC scored 97.5% hits with 7.7% false alarms and 4.5° error. Crucially, the response-type by visual-half-field interaction “did not approach significance in either patient”, all P > 0.5. Established Localisation accuracy of 2.8° and 4.5° is not degraded performance; it is roughly what an intact observer does.

Established The same paper reproduces the classical split, which is why it is evidence rather than an artefact. Reporting the cross-midline comparison for the patient tested on it, the paper records 55.4% correct, all stimulus categories P > 0.25, indistinguishable from chance, against 90.4% for the same comparison within one half-field, all P < 0.001. Verbal labelling was better in the right half-field — 92.1% against 73.4%, P < 0.001 — and picture matching better in the left, 95.5% against 73%, P < 0.001. Strong lateralisation, no cross-midline integration, and still no response-channel interaction: all P > 0.12 across those tasks and all P > 0.63 on high-confidence trials.

Frontier The authors’ conclusion was narrow, and it is worth quoting rather than paraphrasing. “Severing the cortical connections between the two hemispheres does not seem to lead to two independent conscious agents within one brain.” Their model, set out at length in a companion review, is “conscious unity, split perception”: one conscious agent experiencing two parallel, unintegrated streams. That review also states the stake plainly — the model “poses a serious challenge for current dominant theories of consciousness”. Speculative The review offers a candidate mechanism as well: “subcortical areas are still unified in split-brain patients”, so if consciousness depends critically on thalamic processing, conscious unity may simply be untouched by a cortical cut. That is a hypothesis about where consciousness lives, and it has not been tested here.

Frontier The reply from Gazzaniga’s group is not a denial of the data but a rival reading of it. Volz and Gazzaniga argued that direct interhemispheric transfer is an unnecessary explanation because cross-cueing directing hemispheric attention may well account for the findings, and that Pinto and colleagues, by characterising cross-cueing as “behavioural tricks, such as touching the left hand with the right hand”, “reveal that they underestimate the potential range and subtlety of cueing behaviour”. Their analogy is the conjoined twins Abby and Brittany Hensel, who share a pair of arms and legs, have separate neural systems and are athletically coordinated. Speculative The analogy is genuinely instructive and it is also the weakest link in the argument: two people coordinating a shared body is a demonstration that behavioural cueing can be fast and fine-grained, not a demonstration that it happened in Pinto’s sessions.

Frontier In the second round the two camps converged on the one contrast that separates them. Volz, Hillyard, Miller and Gazzaniga argued that Pinto’s results show the hemispheres sharing location but not identity: “the split-brain patient was able to report the location but not the identity of a stimulus in the incongruent condition”, and that “subtle cues may be given by minimal movements of the eyes or facial muscles, which might not even be visible to an external observer”. Their conclusion is explicit — separate conscious streams in each hemisphere, communicating by cross-cueing.

Frontier Pinto’s counter is a bandwidth-and-timing argument plus a methodological complaint. The reaction-time difference between congruent and incongruent conditions was under 200 ms, which they argue is too short for overt cross-cueing; pointing errors of roughly 1–3 visual degrees are within the normal range rather than the degraded performance a low-information channel predicts; and the absence of a response-by-field interaction persisted with equiluminant stimuli, which should suppress the subcortical attentional route. Their sharpest point is about the shape of the opposing hypothesis: cross-cueing is “not clearly defined, which makes it an easy post hoc explanation, but also one that is difficult to falsify”.

Speculative A third reading accepts the data and supplies the mechanism both camps leave thin. Meador, Loring and Sathian agreed that callosotomy does not divide consciousness for basic stimulus awareness and proposed thalamic mediation through crossed corticothalamic connections, noting that Pinto and colleagues had themselves judged it “unlikely that the split-brain patients can achieve a unified consciousness via the anterior and posterior commissures or cross-cuing”. No experiment has yet tested the thalamic proposal in a callosotomy patient.

Established The 2017 challenge has a prehistory that the 2017 framing obscured, and this is the strongest single finding in this brief. Justine Sergent, in Brain in 1987, treated the split brain “as a single organism” whose hemispheres remain connected subcortically, tested two commissurotomised patients on tasks where “only by combining the initially segregated information could a correct response be produced”, and found both significantly above chance in seven of them. Her conclusion was that “even when the two disconnected hemispheres receive different information, the commissurotomized brain works as a single and unified organism”. Cronin-Golomb, a year earlier, had three complete-commissurotomy patients matching concrete concepts across the midline at 90% and 94% correct and abstract concepts at 79% and 83%, concluding that what crosses is “neither verbal nor imagic in nature” but corresponds to “contextual or connotative associates of the stimulus”.

Established The cross-cueing move is older still. Gazzaniga and Hillyard in 1971 attributed differences in verbal reaction time between the two fields to “subcortical transfer mechanisms or cross-cuing”. Both of the explanations now traded across the 2017 exchange were on the table in the founding literature, which means the argument is not about a new finding but about which of two long-available accounts the field should have adopted. Frontier The deeper prehistory is the Rochester series: Akelaitis, testing patients operated by van Wagenen and Herren around 1940, reported no discernible differences between pre- and post-surgical behaviour, a null the field set aside for twenty years and which reads differently now.

Established Residual subcortical transfer is measured, and its bandwidth is the quantity nobody has pinned down. Michael Corballis found that in four subjects lacking a corpus callosum, the redundancy gain from bilateral stimuli exceeded what probability summation predicts when the stimuli were brighter than the background, and located the neural summation at collicular level in the surgical cases when stimuli were equiluminant. Jeff Miller reported “unusually large redundancy gains and unusually large violations of the race model inequality” in the split brain, with the effect vanishing when stimulation was confined to one hemisphere. The de Haan review names spared routes including the superior colliculus, remaining commissures, and the dorsal and ventral pontine decussations. Frontier The awkward detail for the simple subcortical story is that Corballis found the excess gain for stimuli brighter than the background and collicular summation for equiluminant ones, so the route in play depends on the display — which is why 8 treats luminance as a design variable.

Speculative Estimates of how much crosses are estimates. Azenet Lopez, reviewing the same record in 2025, describes the transferred information as restricted to coarse spatial and temporal properties and puts it at “under one bit, the equivalent of an answer to a Yes/No question”, transmitted subcortically. That figure is a reading of behavioural ceilings, not a channel measurement, and 6 asks for the measurement.

Frontier In 2020 the two camps published together, which is the most useful fact in this field. “Split-Brain: What We Know Now and Why This is Important for Understanding Consciousness” carries de Haan, Paul Corballis, Hillyard, Marzi, Seth, Lamme, Volz, Fabri, Schechter, Bayne, Michael Corballis and Pinto as co-authors — principals from both sides plus philosophers of mind. They did not converge. The paper records disagreement on three mechanisms: subcortical information transfer, ipsilateral motor control (whose role it calls “currently not settled”), and strategic cross-cueing. What they agree on is the urgency: “Full split-brain surgery is rare these days, and it is important that we try to answer the central questions while these patients are still available for study.” Established A joint paper that records disagreement rather than papering over it is the precondition for the adversarial protocol in 7, and it already exists.

Frontier The question is still treated as open in the current theoretical literature, not merely in historical reviews. A 2024 paper in Neuroscience of Consciousness building a multidimensional framework for developmental and disordered consciousness states flatly that whether split-brain patients house two streams of consciousness in the same body is unsettled, and proposes constructing separate profiles for each hemisphere if they do. Seven years after the anchor study, the field has not converged.

Frontier The newest empirical development points at the variable the whole dispute has been treating as binary. A group led by Tyler Santander with Michael Miller and Gazzaniga at UC Santa Barbara, with collaborators at Cologne, the Bethel Epilepsy Center and Indiana University, scanned six callosotomy patients — two partial, four described as full-split — and reported in PNAS in November 2025 that a patient whose surgeon, blocked by arterial congestion, “left about a centimeter of fibers in the posterior section” showed no detectable loss of cross-hemispheric synchronisation. Miller’s summary was that the team “thought there would be a lot of disconnection. But we were really surprised there wasn’t any.” The PNAS page refused retrieval; those figures come from the institution’s own account, and 7 treats surgical verification as the first item in any protocol because of them. Speculative If a centimetre of splenium restores synchronisation, then the classical and the revisionist camps may have been reporting different populations under the same label for decades.

Established Two comparison populations bound the interpretation. People born without a corpus callosum are not split-brain patients: a study of 26 adults with callosal agenesis, all with full-scale IQ above 78, found intact general intelligence alongside selective social-cognitive deficits. And Lassonde and colleagues, testing five children aged 6 to 16 before and after callosotomy, found that “younger children are considerably less affected by the interruption of callosal transmission than older children”, attributing the difference to continued reliance on ipsilateral pathways. Age at surgery and time since surgery are not nuisance variables in this literature; they are candidate explanations for the effect under dispute.

Established The supply of new subjects has effectively stopped. Callosotomy is still performed for drug-resistant epilepsy with drop attacks, increasingly by stereotactic laser interstitial thermal therapy: one paediatric series reported 11 procedures in 10 children with 71% free of drop attacks at longest follow-up, improvement in other seizure types in 57%, and no documented complications. But these are children, the sections are usually anterior or partial, and a thermal ablation is not the clean complete section the classical literature rests on. The cohort that can answer the question is the cohort that already exists.

Frontier Does unified agency imply unified experience? Schechter and Bayne put the tension most precisely: patient D.D.C. shows preserved crossed responses, which reads as unified agency, and fails cross-matching, which reads as divided experience. They separate an agency-based conception of unity from an experience-based one and conclude that even granting unified agency, the cross-matching failures make the case for consciousness that is only partially unified. Nothing in the 2017 data adjudicates between the two conceptions, because every measurement runs through a response.

Speculative Is unity layered rather than present-or-absent? Lopez’s 2025 model proposes phenomenal unity at global levels — scene configuration, summary statistics, average size, alignment, apparent motion, all tasks split-brain subjects can do across the midline — with disunity at local levels of shape and colour. It is a coherent reading of a record that neither camp fits comfortably, and it makes a testable prediction: cross-field performance should track the statistical grain of the judgement, not the category of the stimulus.

Frontier Is cross-cueing a hypothesis or a residual category? This is the live methodological question and both camps are partly right. Volz and colleagues can point to decades of demonstrations and to a concrete mechanism — minimal eye and facial movements below the threshold of observer detection. Pinto and colleagues can point out that a mechanism which is invoked whenever unity appears, and specified only loosely, cannot be distinguished from the hypothesis it opposes. Nobody has published a pre-registered definition of cross-cueing with a stated falsifier, and until somebody does the dispute cannot close. Established What can be said now is that the effect the cueing account has to produce is not small: accurate localisation to within a few degrees, on the first response, with a congruent-incongruent difference under 200 ms.

Speculative Is ipsilateral motor control doing the work that consciousness is being credited with? If the left hemisphere retains enough direct control of the left hand, a patient can answer about a left-field stimulus with the “wrong” hand without anything crossing at all. The de Haan review lists this among the three contested mechanisms and declines to settle it; it is the most tractable of the three and the least studied.

Frontier Does the disconnected right hemisphere have experiences? The de Haan group lists this as an open question and proposes attacking it with novel neural paradigms rather than behaviour. The obstacle is structural: the right hemisphere in most of these patients cannot speak, so any evidence of its experience must come from a marker that does not require report — which is the same missing instrument that blocks progress in Consciousness Research and in Dying Brain Activity.

Frontier How much does surgical completeness matter, and has the field been mis-grouping its subjects? The 2025 PNAS result suggests that about a centimetre of posterior fibres can restore functional synchronisation across the hemispheres. If that is right, “complete callosotomy” as an inclusion criterion has been carrying more weight than the imaging behind it supports, and disagreements between laboratories may partly be disagreements about which patients were in fact split.

Speculative Is the 2017 finding a fact about split brains or about adapted split brains? Pinto and colleagues raise this themselves: their patients were tested 10 to 23 years after surgery, and the paper acknowledges that the findings might reflect long-term adaptive reorganisation rather than the immediate consequence of the cut. Lassonde’s developmental result makes the concern concrete rather than rhetorical.

Speculative Does the split brain discriminate between theories of consciousness at all? Pinto’s review claims it does. The claim is harder to cash than it looks: as Integrated Information Theory sets out, the quantity that theory identifies with consciousness cannot be computed for any real brain, so “IIT predicts two complexes after callosotomy” is a qualitative expectation rather than a derived number. Seth and Bayne’s survey of the field states the general problem: it is unclear how current theories relate to each other, or whether they can be empirically distinguished.

Established The binding constraint is N, and it is not a soft constraint. The 2017 anchor rests on two patients; the classical literature rests on a few dozen accumulated over sixty years across Rochester, California, Dartmouth, Ohio, Italy and Germany; the largest recent imaging study scanned six. No source retrieved for this brief gives a worldwide count of surviving, testable complete-callosotomy patients, and the field’s own language — “rare these days”, “rapidly decreasing”, “while these patients are still available” — is the only estimate on offer. That absence is itself a finding about how this literature is governed. Speculative A field that cannot state its own denominator cannot state the power of its own null results.

Established The cohort is heterogeneous in ways that matter more than the sample size does. Complete against partial section; anterior-only against full; one-stage against two-stage; age at operation from childhood to the twenties; decades of post-surgical adaptation; different seizure aetiologies and different residual pathology. Fabri’s tactile data separate complete from partial by 63 percentage points on left-hand naming. Pooling across that variation is not conservative, it is a category error, and single-case reporting is the only honest format.

Frontier The statistical shape of the claim is unfavourable. The classical prediction is the presence of a response-by-field interaction; the challenge is its absence. A non-significant interaction in two patients is weak evidence for the null, and the 2017 paper reports P values rather than the equivalence bounds or Bayes factors the inference needs. This is fixable and nobody has fixed it.

Frontier Eye movements are the live confound and the current instrumentation does not settle them. If a patient makes a saccade during a 120 ms presentation, or after it, the lateralisation assumption fails. The opposing hypothesis explicitly invokes movements “which might not even be visible to an external observer”. Nothing short of high-rate binocular tracking with online fixation enforcement and trial rejection can remove this as a live explanation, and no published split-brain study in this dispute reports it. Speculative The same instrument would also test the opposite worry: that enforcing fixation on elderly patients changes the task enough to suppress the effect under study.

Frontier Anatomical verification is weaker than the categories used to report it. “Radiologically confirmed complete callosotomy” is a clinical read, not a fibre count, and the 2025 scanning result indicates that a centimetre of residual posterior fibres can change the functional picture entirely. Until every participant carries high-resolution diffusion imaging scored blind against a published criterion, group labels in this field are provisional.

Frontier There is no report-independent measurement of consciousness, so every result arrives through the channel under dispute. The question is whether the right hemisphere experiences anything; the evidence is what a hand or a voice does. That circularity is not peculiar to split-brain work — it is the field-wide constraint described at Consciousness Research — but it bites hardest here, because here the disputed subject is the one that cannot speak.

Speculative Patient burden caps the protocol length. These are elderly people with drug-resistant epilepsy, often tested during routine neurological visits. A battery that cannot be completed in a few sessions, close to where the patient lives, will not be completed at all, which constrains 8 more than any technical requirement does.

Frontier A marker of conscious experience that does not route through report. Everything else here is downstream of it. Without such a marker the disconnected right hemisphere can be shown to discriminate, to act and to prefer, and none of that settles whether it experiences. The instrument does not exist in any field; Consciousness Research and Anaesthetic Mechanism record the attempts.

Frontier A settled account of ipsilateral motor control. The de Haan review lists it among the three contested mechanisms and calls its role “currently not settled”. If the left hemisphere can drive the left hand directly to a useful degree, a large part of the 2017 result is explained without unity and without cross-cueing. This is a tractable neurophysiological question that the consciousness dispute has left to one side.

Speculative A measured bandwidth for the residual subcortical channel. Lopez’s “under one bit” is inferred from behavioural ceilings. Corballis located summation at the colliculus; Cronin-Golomb found connotative rather than verbal or imagic content crossing. What nobody has is a bits-per-trial figure with a stated method, and without one the cross-cueing debate is an argument about an unmeasured quantity. Handwave The step from “a thin channel exists” to “a thin channel suffices for phenomenal unity” is made by assertion in every source reviewed for this brief, on both sides of the dispute.

Frontier Theories that make divergent, pre-registrable predictions about this case. The Cogitate collaboration showed what that takes: proponents writing predictions before data exist, and accepting in advance what would count against them. Neither the two-agents reading nor the conscious-unity reading has been written down in that form. Until they are, the split-brain literature cannot be adjudicated by any amount of new data.

Established Pre-registration methods adapted to single-case neuropsychology. The field’s statistical toolkit was built for groups. Case-series designs with pre-specified per-patient analyses, equivalence bounds for null interactions and blinded scoring of video and physiological confound channels all exist; none of them is standard in this literature.

Frontier Comparison cohorts that are actually comparable. Callosal agenesis, hemispherectomy and partial callosotomy each bound the interpretation differently, and each has a literature that does not use the same tasks as the split-brain literature. A harmonised battery across all four populations would be worth more than any single new experiment on the two patients everybody already tests. Established The agenesis cohort is already large enough to support one — 26 adults in a single published comparison, against two in the anchor study.

Frontier The decisive experiment is a pre-registered, adversarially designed protocol whose predictions both camps agree in advance, run on every remaining callosotomy patient who can consent, with enforced fixation, high-rate binocular eye-tracking and blinded scoring of every confound channel. It is decisive because it removes the two things that keep this dispute open: the freedom to reinterpret a result after seeing it, and the freedom to attribute an unwelcome finding to a cueing mechanism nobody has defined. Nothing in it needs hardware that does not exist.

Established The template is already published and already ran. The Cogitate consortium tested global neuronal workspace theory against integrated information theory in 256 participants across fMRI (120), MEG (102) and iEEG (34), each modality at a minimum of two independent laboratories, with the design, the divergent predictions, the expected outcomes and their interpretation all pre-registered and peer-reviewed before data collection, theory proponents separated from acquisition and analysis, and analyses optimised on one third of the sample before final testing on the independent two thirds. Reported in Nature in April 2025, it concluded that the results “align with some predictions of IIT and GNWT, while substantially challenging key tenets of both theories”. The instrument works, including on its sponsors.

Frontier Item one is surgical verification, because the field’s inclusion criterion is weaker than its conclusions. Every participant gets high-resolution diffusion imaging of the callosum and the anterior commissure, scored blind against a published criterion for residual fibres, before any behavioural data is collected. The 2025 PNAS result indicates that roughly a centimetre of posterior fibres is enough to abolish the functional disconnection signature; a protocol that does not measure this is comparing groups it has not defined. Established This is also the cheapest item on the list: the imaging is routine, the patients are already in clinical follow-up, and the scoring criterion is the only thing that has to be invented.

Frontier Item two is fixation enforcement, which converts cross-cueing from a rhetorical move into a testable one. Binocular tracking at 500 Hz or better, gaze-contingent aborts, trial-by-trial rejection, plus facial electromyography and synchronised multi-camera video of hands and body scored by raters blind to condition. Volz and colleagues named eye and facial movements as the mechanism; this measures them. A response-by-field null that survives clean-trial-only analysis is evidence the cross-cueing account cannot absorb.

Frontier Item three is the identity-versus-location contrast, which both camps have already agreed matters. Volz and colleagues read Pinto’s data as showing location crossing and identity not crossing. That is a concrete, pre-registrable prediction: cross the response channel with the half-field for location judgements and for identity judgements separately, in the same session, in every patient, and state in advance which pattern each camp accepts as defeat. Speculative On the published record the classical camp should predict that identity never crosses while location sometimes does; the unity camp should predict that neither crossing is required, because the single agent already has access to both streams.

Established Item four is the statistics, stated before the data exist. The challenge is an argument from a null, so it needs equivalence testing or Bayes factors with pre-specified bounds rather than a non-significant P value, analysed per patient rather than pooled, with the heterogeneity in 5 treated as a covariate of record rather than noise.

Frontier Item five is coverage and harmonisation, which is the de Haan group’s own recommendation. Their paper asks that the field “test as many of the available patients as possible with the same tests”, extend into the somatosensory domain where research has been sparse, and use event-related potentials as markers of concurrent conscious processing in the two hemispheres. A travelling rig and one battery does all three at once.

Speculative Item six is the covert-response arm, which is the only part that could speak to experience rather than control. The same review proposes testing whether the hemispheres make independent conscious decisions by having them respond covertly and decoding mental imagery from each side separately. If a disconnected right hemisphere sustains a decodable, stimulus-specific state while the left hemisphere reports a different content, that is the closest this natural experiment can come to evidence of two streams. Handwave It would still not be proof, because a decodable stimulus-specific state is not by itself an experience, and nothing in this protocol closes that gap.

Frontier Item seven is the arbitration structure, without which the rest is a collaboration rather than an adversarial one. A neutral data steward, a Registered Report accepted before results, a public analysis container, and a signed statement from each camp naming the outcome it would accept as losing. Daniel Kahneman’s formulation of adversarial collaboration — agree the design, appoint an arbiter, publish jointly whatever happens — is the governing model, and the Cogitate preregistration deposited by Mudrik and colleagues in December 2022 is the worked example.

Speculative What would change this brief’s assessment fastest, short of the full protocol. A single patient, tested under enforced fixation with clean electromyography, showing a robust response-by-field interaction that Pinto’s tasks did not find would restore the classical reading at once. A single patient showing cross-field identity transfer at above-chance accuracy under the same controls would end the cross-cueing defence. Both are one well-instrumented session away, and neither has been reported.

Handwave What the field cannot do, and should stop implying it might. No new complete-callosotomy cohort will be created for research purposes, no animal model reproduces the verbal-report asymmetry that makes the human case informative, and no simulation can settle a question about whether a physical system has experiences. The window is the surviving patients, and after it closes this brief’s question becomes an exercise in re-analysing an archive.

Established The rig has to travel, because the patients will not. The surviving cohort is distributed across at least Italy, Germany and the United States and is tested during routine clinical visits. That sets the engineering target: a transportable, calibration-stable apparatus that can be set up in a neurology clinic in under a day and produce data identical in format to every other site’s.

Established Display timing has to be verified, not assumed. The anchor experiments used 120 ms presentations, with 100 ms in one shape-detection task. On modern displays a nominal 120 ms is frequently something else; photodiode verification of onset and duration on each installed system, logged per trial, is the minimum, and it is the kind of detail that decides whether a multi-site null means anything. Speculative A 20 ms drift in stimulus duration across sites is enough to move a saccade-contaminated trial into or out of the analysed set.

Frontier Luminance control is a scientific requirement here, not a nicety. Corballis found collicular summation appearing for stimuli brighter than the background and not for equiluminant ones, and Pinto’s fifth experiment used both equiluminant and bright conditions precisely to probe the subcortical route. Per-patient photometric equiluminance calibration, repeated at each session, is therefore part of the design rather than part of the setup.

Established Eye-tracking specification follows from the confound it exists to kill. Binocular, 500 Hz or higher, with drift correction, gaze-contingent trial abort inside the presentation window, and stored raw traces for blind re-scoring. Paired with facial electromyography and time-locked video, this is the only part of the protocol that could retire the cross-cueing dispute rather than continue it.

Frontier Imaging requirements are heavier than the behavioural ones. High-angular-resolution diffusion imaging to resolve residual callosal fibres at millimetre scale, resting-state and task fMRI to replicate the 2025 synchronisation result, and where an epilepsy centre already has intracranial coverage, electrophysiology at the temporal resolution the decoding arm needs. The instrumentation overlaps almost entirely with Neural Interfaces and Brain-Computer Interfaces, which is where the decoding methods come from.

Established Data engineering is the part most likely to be skipped and most likely to matter. A standard format, a pre-registered analysis container that runs unchanged at every site, per-patient identifiers held under a governance scheme that acknowledges a cohort this small cannot be anonymised, and deposition of raw traces including rejected trials. 13 assumes this archive outlives the patients, because it will have to.

Established Epilepsy surgery is the upstream technology and it has moved away from producing split brains. Vagus nerve stimulation, responsive neurostimulation and laser interstitial thermal therapy have absorbed much of the indication that used to lead to complete callosotomy. The paediatric laser series cited in 3 — 11 procedures in 10 children, 71% free of drop attacks at longest follow-up — is what the operation looks like now: partial, thermal, and in children.

Established Diffusion tractography is the adjacent measurement that changes the inclusion criteria. The difference between “complete” and “a centimetre of splenium remaining” is invisible to the clinical read that most of this literature relied on and visible to modern tractography.

Frontier Decoding methods carry the only arm of the protocol that bypasses report. Multivariate pattern analysis and imagery decoding developed for Brain-Computer Interfaces are what make a covert-response experiment thinkable, and their limits are that literature’s limits.

Frontier The theoretical neighbours are the theories this case is supposed to constrain. Integrated Information Theory, the global-workspace family surveyed at Consciousness Research, and the assessment problem set out at Machine Consciousness all take the split brain as a datum; none of them derives a quantitative prediction from it. Speculative That is the adjacency worth fixing: a theory that cannot say what a split brain should look like is not being tested by one.

Speculative The distant neighbours are the interventions that would divide or join minds deliberately. Whatever this case establishes about how much connectivity unity requires is the empirical floor under Mind Uploading and under any proposal to couple nervous systems; a unity that survives cutting the largest cortical bridge is a cheaper property than those proposals usually assume.

Frontier The first requirement is a registry, and its absence is why nobody can state the sample size. Surviving complete-callosotomy patients are held by a small number of centres — the Ancona group in Italy that supplied DDC and DDV, the Dartmouth and UC Santa Barbara lineage in the United States, the Bethel Epilepsy Center in Germany, and the residue of the California series. No standing multi-centre registry exists, so the denominator of every claim in this literature is unknown to everyone including the people making it.

Established The second is a funding instrument that pays for adversarial designs. One exists: the Templeton World Charity Foundation’s structured adversarial collaboration programme funded the Cogitate collaboration and the successor comparison of integrated information against predictive-processing accounts published in 2026. Split-brain work has not been commissioned through it, and this brief cannot find a reason beyond the fact that nobody proposed it. Speculative The cost profile is favourable compared with what that programme already funds: no new scanner cohort of 256 participants, a travelling behavioural rig, and imaging that mostly exists in clinical records already.

Frontier The third is a publication route that accepts results before it sees them. Registered Reports remove the incentive that has shaped this field most: a two-camp dispute in which each camp publishes its own patients on its own tasks produces exactly the literature that exists. The 2020 joint review shows the principals will sit at one table; what has not been built is the instrument that binds them to an outcome. Established Both of the principals have already co-signed a paper stating the disagreements precisely, which is the hardest part of an adversarial collaboration and the part most projects never reach.

Speculative The fourth is cross-jurisdictional ethics approval for a travelling protocol. One battery, four or five countries, identifiable participants, and a data-sharing regime that has to be honest about the fact that a cohort of this size cannot be de-identified. Nothing about this is novel; everything about it is slow, and the biological window in 13 is short enough that the slowness is itself a risk to the result.

Frontier The fifth is an archive commitment that outlives the cohort. Once the last participant is gone, the only thing that can be re-analysed is what was recorded, in the format it was recorded in, with the confound channels that were stored. Decisions made now about what to keep determine what the question can be asked of in 2060.

Frontier Consent capacity is not uniform across this cohort, and the studies that report it report it thinly. DDC and DDV are described as having normal IQ and no other brain damage, which is not the general case for people who reach complete callosotomy for drug-resistant epilepsy. A protocol that travels to every remaining patient will encounter capacity questions that a two-patient study did not have to face, and the assessment of capacity has to be independent of the investigators who want the data.

Speculative If the two-agents reading is right, the consent question becomes strange rather than difficult. A left hemisphere that speaks cannot consent on behalf of a right hemisphere that is, on that reading, a separate subject with no channel to refuse. The classical picture, taken seriously, implies that sixty years of split-brain research was conducted with the agreement of one of the two parties present. The field has largely not said this out loud, and it is a consequence of its own textbook position.

Established The cohort is irreplaceable, which changes the calculus on research burden. Every session consumed by an underpowered study on a non-harmonised task is a session unavailable to the protocol in 7. Testing elderly patients with epilepsy carries fatigue and seizure risk, and there is no argument from scientific value that justifies spending the last of a finite resource on designs that cannot settle anything. Speculative Triage of research access, normally an administrative matter, is here a substantive ethical judgement about who gets the last sessions.

Established De-identification is not achievable and should not be claimed. With a handful of participants worldwide, a surgical history and an institutional affiliation identify a person. Data governance here has to be consent-based and access-controlled rather than anonymity-based, and papers should stop using language that implies otherwise.

Frontier The stakes beyond the cohort are the reason the question is worth the burden. If unity of consciousness survives the loss of the largest cortical connection, the threshold for attributing unified experience to other systems drops, with consequences at Machine Consciousness and Digital Minds. If it does not, then a surgical procedure creates a subject, and the moral arithmetic of every disconnection syndrome changes.

Speculative There is also a duty not to instrumentalise. These are patients who consented to surgery to stop seizures and then spent decades as the evidence base for a philosophical dispute they did not choose. A protocol designed to close that dispute owes them a clear account of what it is for and a result they are told about.

Frontier One body, one responsible agent is an assumption, and this is the only place it has been experimentally stressed. Criminal law, contract, medical consent and the concept of a person all presume that a human being is a single locus of intention. The split-brain case is the one human intervention that puts a number on how much physical connection that presumption requires — and the number, if the 2017 reading holds, is smaller than the largest fibre tract in the brain.

Speculative A cheap unity and an expensive unity imply different futures. If unity is preserved by thin subcortical channels, then distributed and artificial systems clear the bar more easily than integration-based theories suggest, which raises the probability that the assessment problems at Machine Consciousness and Synthetic Consciousness arrive sooner and with weaker tools. If unity requires the callosum and the 2017 result is a cross-cueing artefact, the opposite follows.

Speculative Division as an available operation is the uncomfortable corollary. If callosotomy divides a subject, then a mind is something that can be split by a scalpel, and the question of what may be done to a nervous system acquires a class of harm that has no current legal name. Proposals to couple or partition nervous systems, and the continuity arguments at Mind Uploading, inherit whichever answer this case returns.

Established The methodological lesson is already banked, regardless of the outcome. A textbook consensus about the structure of consciousness was built on a few dozen patients, held for fifty years, and was challenged in 1986 and 1987 by results the field did not absorb. Cronin-Golomb and Sergent reported cross-midline integration three decades before the 2017 paper. That is a fact about how disciplines maintain their canonical cases, and it generalises well beyond this one. Frontier The corrective is not more scepticism about canonical cases but the specific instrument in 7: predictions written down before the data, by the people who will be embarrassed by them.

Handwave The largest claim occasionally made for this literature — that split-brain data show the self to be a narrative construction with no unified referent — does not follow from any measurement reported here. It is a philosophical position that borrows the prestige of a neurosurgical result, and the result does not reach it.

These horizons track patients, protocols and adjudications rather than hardware, because the instruments this question needs already exist and the cohort that can answer it does not renew:

• 10 yr: Frontier The decisive window. If the harmonised, pre-registered, fixation-controlled battery is run on every surviving patient, it happens here or not at all. Established Expect the behavioural dissociations — left-hand tactile naming near 30%, chance-level cross-midline matching, left-hand apraxia — to keep replicating in whoever is tested. Speculative Expect anatomical verification by diffusion imaging to become a condition of publication following the 2025 synchronisation result.
• 25 yr: Speculative The complete-callosotomy cohort is gone. The question is settled on the archive built in the preceding decade or it is not settled at all, and the substitutes — callosal agenesis, hemispherectomy, reversible interhemispheric inactivation in animals — answer adjacent questions rather than this one. Frontier A report-independent marker of experience, if one arrives from elsewhere in consciousness science, would re-open the archive rather than replace it.
• 50 yr: Speculative Either the question has been absorbed into a general theory that makes the split-brain case a worked example, or it has been retired as underdetermined by the evidence that was collected while it could be. This brief treats the second outcome as at least as likely as the first, on the strength of how the 1986 and 1987 results were handled.
• 100 / 250+ yr: Handwave Deliberate division and recombination of conscious systems as an experimental method, with the unity of consciousness as an engineered parameter. Nothing on any current path leads here, and the step from a surgical accident to a controlled instrument is exactly the step this brief cannot flag any higher.

• Depends on This brief waits on adjudications owned elsewhere on the map. Whether any neural index tracks experience rather than processing is the open problem of Consciousness Research; whether the right integrated structure suffices, and whether the quantity it names can be computed at all, is contested at Integrated Information Theory; the best-characterised transitions into and out of unconsciousness come from Anaesthetic Mechanism; and the general assessment problem is stated at Machine Consciousness. No typed depends-on edge is claimed, because none of those produces the specific instrument this brief is blocked on.
• Requires (not on this map) First, a marker of conscious experience that does not route through report, without which a mute hemisphere’s discriminations can never be distinguished from its experiences. Second, a standing multi-centre registry of surviving callosotomy patients, without which nobody can state the denominator that every claim in this literature implicitly uses. Third, a binding pre-registration instrument for single-case neuropsychology — predictions, falsifiers and per-patient analyses fixed before data exist — without which each camp keeps reinterpreting the other’s results. Fourth, a measured bits-per-trial bandwidth for the residual subcortical channel, without which the cross-cueing argument is a dispute about an unmeasured quantity. Fifth, portable eye-tracking and timing-verified presentation rigs that can be carried to a neurology clinic, without which the surviving patients cannot all be tested on the same battery.
• Enables A settled reading here would constrain every theory that makes integration across a substrate do the explanatory work, and would give the assessment of unity in unresponsive patients, in non-human animals and in artificial systems its only human intervention-based anchor. No typed enabling edge is claimed, and the reason is the finding: the adjudication does not exist, so nothing downstream is yet enabled.
• Adjacent Epilepsy surgery and its replacements on the clinical side; diffusion tractography and multivariate decoding on the measurement side via Neural Interfaces and Brain-Computer Interfaces; and on the interpretive side Consciousness Research, Integrated Information Theory, Dying Brain Activity and Synthetic Consciousness.

Speculative “Split-brain patients have two minds in one skull.” This is the strongest claim in circulation and the evidence does not carry it. The measured facts are dissociations of perception and of motor control; the step to two subjects of experience requires a premise nobody has tested. Notably, Volz and Gazzaniga’s own fiftieth-anniversary review reports that “patients do not report any difference in the nature of their personal experience — despite the fact that their hemispheres are separated, they report that they experience a single consciousness”. The explicit two-streams claim appears in the 2018 reply, as an inference from cross-cueing, not as an observation.

Frontier “The 2017 study overturned the textbook.” It did not, and the same paper is the evidence. Cross-midline matching in those patients ran at 55.4%, statistically indistinguishable from chance, against 90.4% within a half-field; verbal labelling was 19 percentage points better in the right field and picture matching 22 points better in the left. The disconnection is intact. What the study contests is one specific prediction — the response-by-field interaction — and one specific inference from it. Frontier A fair summary is that the textbook account of divided perception survived in those patients and the textbook account of divided control did not.

Established “This is the left-brain/right-brain thing.” No. The lateralisation these patients demonstrate is narrow and specific: speech output and syntax overwhelmingly left, visual matching better on the right-hemisphere side. Gazzaniga and Hillyard established the language asymmetry in 1971 and it has held. The popular extension to logical versus creative personality types has no support in this literature and never had any.

Frontier “Cross-cueing explains it.” Cross-cueing is real, it is old, and Volz and colleagues give it a concrete mechanism in eye and facial movements too small to see. It is also, as currently specified, unfalsifiable: Pinto and colleagues’ complaint that it is “not clearly defined, which makes it an easy post hoc explanation, but also one that is difficult to falsify” is fair on the published record. Both things are true at once, which is why 7 makes defining it a deliverable rather than an assumption.

Established “A unified mind needs a corpus callosum.” People born without one are the counterexample. In a study of 26 adults with callosal agenesis, all with full-scale IQ above 78, general intelligence was intact, with deficits selective to social cognition. Absence of the callosum from birth is a different condition from its surgical removal in adulthood, and that difference is itself informative — Lassonde’s finding that younger children are far less affected by callosotomy than older ones says the compensating pathways are developmental.

Frontier “Callosotomy is obsolete, so the question is closed.” The operation is still performed, increasingly as laser interstitial thermal therapy — one paediatric series reports 11 procedures in 10 children with 71% free of drop attacks at longest follow-up and no documented complications. But those are partial and anterior ablations in children, not the complete sections in adults that the classical literature rests on. The question is not closed; the supply of new evidence for it is. Speculative A thermal ablation also leaves a different lesion boundary from a scalpel, so even a complete laser callosotomy would not straightforwardly join the classical series.

Speculative “The split brain refutes integrated information theory.” The argument is usually that a theory identifying consciousness with a maximum of integration over a substrate must predict two conscious complexes after the substrate’s main bridge is cut, and that Pinto’s data show one. Two things block it: the quantity in question cannot be computed for any real brain, as Integrated Information Theory sets out, so the prediction is qualitative; and what Pinto measured is unified control, which Schechter and Bayne argue is agency-based unity and not the experience-based unity the theory is about.

Established “Nothing much happens to these patients in daily life, so the surgery is functionally silent.” The everyday impression is genuine — verbal IQ and problem-solving are typically unaffected — and the laboratory record contradicts the inference. Left-hand tactile naming at 30% against 93% for the right hand, apraxia on pantomime to visually presented objects, and diagonistic dyspraxia in which the left hand undoes the right are all measured in complete-callosotomy patients. Both facts are true, and the gap between them is the phenomenon.

Handwave “Enough philosophy and the case resolves itself.” Three serious positions now sit on the same data — two agents with a thin channel, one agent with split perception, and partial or layered unity — and all three are consistent with everything published. The dispute is not underdetermined by argument; it is underdetermined by measurement, taken on two people, with no fixation control, twenty years after the operation.