Latest
2026-07-22 · Preprint · bioRxiv
A re-analysis of the COGITATE brain recordings finds the frontal 'flare-up' at picture offset that the original analysis missed.
What we do with it: This is the 'July 2026 reanalysis' the plan could not name. It goes into our multiverse on COGITATE (task #11) as one of the analysis paths, and shows why we hash the analysis set first. Planned
2026-07-07 · Paper · Neuroscience of Consciousness 2026(1), niag029
Testing 246 ways of measuring brain connectivity on the COGITATE data gives tentative support to the workspace theory.
What we do with it: A second independent reanalysis of the same data, from a different angle. Its 246-measure sweep is a template for our pre-registered multiverse; we would hash the measure list first. Planned
2026-06-09 · Preprint · medRxiv/bioRxiv
Resting brain signals could stand in for the stimulation-based consciousness index in some brain-injured patients, but not across all of them.
What we do with it: Tempers the 2024 claim that resting EEG can replace PCI. Our out-of-sample test of that claim should report results by aetiology. Planned
2026-06 · Paper · Trends in Cognitive Sciences 30(6), 488-501
A method for judging whether an AI could be conscious by checking its inner workings against what theories of consciousness require, not by what it says.
What we do with it: The rubric for our AI stress tests: we use it to find where theories disagree or give absurd answers, never as a verdict on AI and never scoring an AI's self-report. Planned
2026-05-14 · Paper · Neuroscience & Biobehavioral Reviews 187, 106742
The second big adversarial collaboration has published its plan: IIT against two predictive-brain theories. No results yet.
What we do with it: If its data appears after we seal our predictions, it is a clean prospective test. Its evidence-pooling idea is a model for our own log-score comparison of theories. Planned
2026-02-14 · Paper · Translational Psychiatry 16, 160
With a sedative given alongside psilocybin, brain-signal complexity still rose as the psychedelic effects came on.
What we do with it: A case where complexity and the spectral exponent move together under a drug that changes experience. Our test asks whether complexity adds anything once the exponent is known. Planned
2026-01-28 · Paper · Nature Human Behaviour 10, 777-802
Across humans, monkeys, marmosets and mice, anaesthetics break down the brain's integration of information, and waking a monkey with deep-brain stimulation brings it back.
What we do with it: Shows that whether a Phi-like proxy rises or falls under anaesthesia depends on the modality and the estimator. Our known-Phi network test (task #9) must check each proxy before we trust its direction. Planned
2025-09-24 · Paper + dataset · Scientific Reports (data: OpenNeuro ds005620)
Brain complexity dropped under sedation, but did not differ between people who said they had been dreaming and people who said they had not.
What we do with it: One of our L2 datasets. The full set is far over our 1.5 GB budget; we fetch only the pre-awakening sedation files per subject. The 5 no-experience awakenings make it weak alone for dream vs no-dream; pool with DREAM. Planned
2025-09-16 · Paper · Cerebral Cortex 35(9), bhaf254
In 303 unresponsive patients, a popular complexity measure mostly repeated what the slope of the EEG spectrum already showed; once corrected for that, its diagnostic value vanished.
What we do with it: The direct source for our surrogate gate (task #8) and the LZ vs slope test (task #4). It is the PMC12448740 paper the plan cites. We also stratify by aetiology wherever patient data allow. Planned
2025-08-13 · Dataset · Nature Communications 16, 7495
Twenty sleep studies pooled their brain recordings with what people said they were dreaming when woken up: 505 people, 2,643 awakenings.
What we do with it: The key resource for our dream vs no-dream test (task #6). Their AUCs are the bar to beat, but we split by subject and lab, so our numbers may come out lower. Planned
2025-04-30 · Paper · Nature 642, 133-142
The first big head-to-head test of two theories of consciousness found that each got some predictions right and each got key claims wrong.
What we do with it: The data is public (Experiment 1: MEG, fMRI, iEEG), so it is the first place we test whether the published conclusions survive other reasonable analyses. Note the paper's iEEG n is 34; the plan said 38. Planned
Landmarks
Older papers that set the terms of the question.
2024-08-05 · Paper · Communications Biology 7, 946
In 15 healthy adults under three anaesthetics, resting EEG predicted the stimulation-based consciousness index closely.
What we do with it: If true, the best-validated measure could be tested on many more open datasets. Caution: 15 subjects, 5 per drug, and the same cohort as Colombo 2019 and Sarasso 2015, so these are not independent replications. We test it out of sample. Planned
2024-05-16 · Paper · Neuroscience of Consciousness 2024(1), niae021
In rats, integrated-information estimates from spontaneous EEG went up under anaesthesia, opposite to the stimulation-based index.
What we do with it: A warning that Phi proxies may not track what IIT says Phi tracks. Our known-Phi network test checks this before any proxy touches brain data. Planned
2023-10-17 · Paper · PLOS Computational Biology 19, e1011465
The current formal version of integrated information theory.
What we do with it: The reference for the IIT row on our scoreboard and for which toy networks have exactly computable Phi in our planted-answer tests. Planned
2023-09-16 · Letter · PsyArXiv (open letter); follow-up in Nature Neuroscience 28, 689-693 (2025)
A group of researchers called IIT pseudoscience, which set off a public dispute over what counts as a test of a theory of consciousness.
What we do with it: It is why we judge every theory only by predictions sealed in advance. Planned
2017-04-19 · Paper · Scientific Reports 7, 46421
Psychedelics pushed brain-signal diversity above normal waking levels, the first time such measures went higher than wakefulness.
What we do with it: Tests whether a measure rises when contents intensify but level does not obviously change. We re-check the spectral control with phase-randomised surrogates. Planned
2017-04-10 · Paper · Nature Neuroscience 20, 872-878
Less slow activity at the back of the brain marked dreaming in both kinds of sleep, and could predict a dream report in real time.
What we do with it: The sharpest within-stage test of a measure against report. The DREAM database (2025) lets us test whether it replicates across labs. Planned
2016-11-02 · Paper · Annals of Neurology 80, 718-729
A cutoff on the stimulation-based complexity index, set in people who could report, separated conscious from unconscious perfectly and flagged some 'vegetative' patients as possibly conscious.
What we do with it: The strongest published L2 result. No open TMS-EEG set of this kind exists, which is the biggest data gap in our plan. Planned
2015-11-19 · Paper · Current Biology 25, 3099-3105
Under ketamine, people stopped responding but their brain's response to a magnetic pulse stayed wake-like, and afterwards they reported vivid dreams.
What we do with it: Defines our ketamine dissociation case. The same group's data underlies Colombo 2019 and Maschke 2024, so we need an independent ketamine dataset. Planned
2013-08-14 · Paper · Science Translational Medicine 5(198)
The paper that introduced the perturbational complexity index: tap the brain with a magnetic pulse and measure how complex the echo is.
What we do with it: The measure every spontaneous-EEG candidate is judged against. Planned