States, contents and neural correlates

The ladder that hides the question

The usual picture of consciousness resembles a lift. Ordinary wakefulness sits near the top. Drowsiness is lower. Dreamless sleep and general anaesthesia are near the bottom. Coma appears below them. A researcher can then place a neural measurement beside this ordering and ask whether the value rises and falls with the “level” of consciousness.

This picture is useful for urgent bedside summaries. It is a poor causal model. Dreaming can contain vivid imagery while the sleeper is disconnected from the room. Some anaesthetised people can remain internally conscious despite behavioural unresponsiveness. A minimally conscious patient may show narrow, intermittent capacities that do not fit neatly between two neighbouring rungs. Global states differ in the range, structure and functional availability of their contents, rather than only in amount. Bayne, Hohwy and Owen therefore argued that global conscious states are better treated as multidimensional than as points on a single scale.³

The hidden assumption in the lift model is that three things move together:

state, the organism’s overall conscious condition;
content, the particular experience present now;
access, the ability to use, remember or report that experience.

They often covary in ordinary life, which is why the assumption feels natural. Consciousness experiments become informative precisely where they dissociate.

A condition name is compression, not explanation

“Awake”, “REM sleep” and “propofol sedation” are names for experimental or clinical conditions. They are not direct readings of consciousness. Each name compresses several variables: sensory connection, spontaneous imagery, arousal, response capacity, memory formation, autonomic regulation and the drug or sleep physiology itself. Two people assigned the same label can occupy different regions of the state-content atlas. The same person can move between internally rich and sparse periods without crossing a recognised stage boundary.

This matters because reverse inference begins as soon as a condition label is treated as the phenomenon. If a neural feature separates wakefulness from deep sedation, the data show sensitivity to that condition bundle. The feature might follow experience, environmental connection, neuromodulatory tone, cortical responsiveness or a combination. Recent work that separates content-related from state-related markers makes the same point empirically: different measurements can be informative along different dimensions rather than ranking every condition on one hidden consciousness scale.¹³

A useful discipline is to ask three reverse questions. Can the target vary inside this named condition? Can the same target occur in a different condition? Can the condition change while the target is approximately preserved? If any answer is yes, the condition label cannot stand in for the target. It must become one variable in the design.

Figure 1. A state-content atlas, not a consciousness ladder

Conceptual field

The atlas prevents a common category error: moving left towards disconnection is not automatically moving down towards absent experience. Experimental conditions occupy regions with uncertainty, not exact points. Illustrative synthesis informed by multidimensional state accounts and state-content marker research.³¹³

Three variables before one brain map

Before choosing electrodes, scanners or statistics, write down three variables. Let S denote the global state being compared. Let C denote the target content, such as seeing a red square. Let B denote the evidence bridge by which the experimenter decides whether that content occurred. A button press, an eye movement, a later dream report and a command-following imagery pattern are different bridges. They do not become interchangeable merely because each is labelled “awareness”.

A minimal observation model is:

ΔY = αΔS + βΔC + γΔB + δΔP + ηΔQ + κΔI + ε

Y is the recorded neural feature. P contains prerequisites such as arousal, expectation and attention. Q contains consequences such as decision, report preparation and memory encoding. I contains direct effects of the intervention, including pharmacology or stimulus energy. The coefficients say how strongly each changing cause influences the measured feature. The equation is not a theory of consciousness. It is an accounting device for the contrast.

The smallest useful example is a red square presented near threshold. On “seen” trials the participant attends more, accumulates enough evidence, presses “seen” and later remembers the item. On “unseen” trials these events are less likely. The observed brain difference could contain all of them. A contrast is clean only relative to a declared target and a declared set of blocked paths.

Why more trials cannot identify a bundled cause

Randomisation is powerful, but its guarantee is narrower than is often assumed. Randomly assigning a mask, task instruction or anaesthetic dose can estimate the total effect of that assignment on the neural measurement. It cannot, by itself, divide that total among altered experience, altered attention, altered report, altered memory and direct physiology. Those pathways are all downstream of the assignment.

Return to the synthetic 6.5-microvolt difference. One causal world may contain 2 microvolts of content, 3 of report and 1.5 of attention. A second may contain no content contribution, 4 of decision confidence and 2.5 of motor preparation. Both worlds generate the same mean contrast. Ten thousand additional trials narrow the confidence interval around 6.5, but they do not tell us which world produced it. Sampling noise shrinks; structural ambiguity remains.

Adding covariates is not an automatic cure. Attention recorded after stimulus onset may already have been changed by the emerging percept. Confidence and memory are consequences of seeing. Statistically controlling a consequence can remove part of the target effect or open a spurious path through selection. The design therefore needs interventions that vary suspected passengers independently, or at least a family of contrasts in which their movement differs. Identifiability comes from changed causal geometry, rather than only from increased statistical precision.

The argument in one pass

Chart · flow

flowchart TD
  A["The ladder that hides the question"] --> B["Every contrast carries passengers"]
  B --> C["Build contrasts that survive<br/>interpretation"]
  C --> D["What the evidence can honestly buy"]
  D -. "re-test" .-> A
  classDef first fill:#0a1424,stroke:#5B5CF6,stroke-width:2px,color:#e6f3ff
  classDef step fill:#0a1424,stroke:#3f7da0,stroke-width:1px,color:#e6f3ff
  class A first
  class B,C,D step

A structural route through States, contents and neural correlates: from The ladder that hides the question to Unbundle the contrast before naming the…. Reading use: follow the relation, then return to the surrounding argument for evidence and limits.

Every contrast carries passengers

The classical neural-correlate strategy compares conditions in which an experience is present with conditions in which it is absent. Chalmers described an NCC as a minimal neural system whose state maps directly onto a conscious state, while later experimental work emphasised the need to distinguish the correlate proper from its prerequisites and consequences.¹²⁴ The distinction sounds tidy on paper. In an experiment, the three are temporally and causally braided.

Suppose attention makes the red square more likely to be seen. Attention is then a cause of both experience and some neural activity. It can correlate with seeing without constituting the experience. Suppose seeing triggers confidence, working memory and a button press. Those are consequences. They also correlate with seeing. A classifier trained on the whole trial may predict conscious reports perfectly by reading prerequisites and consequences alone.

Predictive success does not identify the causal role of the predictor. This is the central mechanism behind the contrast bundle.

Figure 2. The contrast bundle and its travelling paths

Causal intervention

The recorded difference receives causal contributions from several branches. Dotted paths mark the inferential bridge from an unobserved experience to an observable label. The experimental task is to block or measure non-target routes, rather than only to obtain a significant contrast. Practitioner synthesis extending the prerequisite/consequence distinction.²¹⁸

Worked example 2 · Synthetic component mixture

A numerical contrast that lies by addition

Assume a synthetic electrode carries three effects. Conscious visual content adds 2 microvolts. Preparing an explicit report adds 3. Increased attention adds 1.5. In the standard “seen and report” condition, all three are present. In the “unseen” condition, none is present. The observed contrast is 6.5 microvolts.

Now remove overt report while using a validated involuntary eye proxy. The content effect remains 2. Attention still differs slightly, contributing 0.6. The contrast becomes 2.6. Nothing about experience need have disappeared. The larger original value was simply the sum of content, report and attention.

This example exposes two mistakes. The first is localisation by magnitude: the strongest signal is assumed to be the most important NCC. The second is disappearance by subtraction: a signal that vanishes in no-report is assumed irrelevant to consciousness. It may instead support access, metacognitive monitoring or stable use of the experience. Those are genuine cognitive functions, but they are different claims.

Figure 3. The same experience, a smaller observed contrast

Illustrative chart

A smaller no-report contrast does not imply weaker experience. It can show that the original signal included report preparation and attention. The numbers are synthetic and expose the additive logic rather than estimate any particular component. Illustrative synthetic data.

Four paradigms, four different bundles

Near-threshold perception appears to hold the stimulus constant, yet tiny sensory fluctuations, prestimulus excitability, attention and decision criterion can sort trials into “seen” and “unseen”. Masking gives the experimenter tighter control over timing, but the mask changes sensory processing directly. A later neural difference may therefore reflect the physical manipulation as well as the target experience.

Binocular rivalry keeps retinal input relatively stable while perception alternates. That is a major advantage, but adaptation, eye movements, switch dynamics and the method used to detect a switch still travel with the contrast. No-report variants exchange the manual decision for a physiological proxy. They remove some consequences while adding a new measurement model.

Sleep and anaesthesia create an even broader intervention. They allow state comparisons that threshold paradigms cannot provide, while simultaneously changing arousal, connectivity, spontaneous activity, responsiveness and memory. A result recurring across all four families deserves more weight precisely because the bundles differ. A result found in only one family may still be useful, but it should inherit that family’s most plausible alternative explanation. The unit of interpretation is therefore not “the seen-unseen contrast”. It is the target plus the particular bundle used to reveal it.

No-report is a trade, not a purification ritual

No-report paradigms were developed because verbal and manual reports recruit decision, working memory and motor systems.⁵ Binocular-rivalry studies have used optokinetic eye movements and pupil changes to estimate perceptual switches without asking for a button press. Removing active report reduced some frontal activity that had previously been attributed to perceptual transitions.⁶ Other no-report work has still found visual, cortical and subcortical activity linked to perceived stimuli after overt report and measured eye contributions were reduced.⁷¹⁰

The gain is real. So is the cost. Experience is now inferred through a proxy. The proxy might fail under fatigue, sedation, pathology or a different stimulus class. It can be affected by attention or low-level image properties. No-report also removes direct trial-by-trial testimony, which can make misclassification invisible. The right comparison is not report versus no-report. It is one evidence bridge versus another, each with different error paths.

Worked example 3 · State-dependent misclassification

Proxy drift is itself a state variable

Suppose an eye-based proxy identifies the dominant percept with 92 per cent sensitivity and 90 per cent specificity in alert participants. In a block of 100 trials with equal numbers of the two percepts, the proxy would still mislabel about nine trials. That error may be tolerable if it is measured and propagated into the neural estimate. It becomes dangerous when the same calibration is carried into fatigue or sedation without testing.

Now imagine that sedation weakens the eye response so sensitivity falls to 65 per cent while specificity remains high. Trials with the target experience are preferentially placed in the “absent” bin. The estimated neural content effect shrinks even if the underlying relation is unchanged. Because the proxy error depends on state, it can manufacture an apparent interaction between state and consciousness. This numerical example is synthetic, but the logic applies to any pupil, eye-movement, autonomic or decoder bridge.

Calibration should therefore be treated as part of the main model. Report proxy performance by state, content class and time, rather than as one global accuracy. Where direct report is intermittently possible, use it to estimate drift. Where it is impossible, include negative controls that drive the proxy without the target percept and positive controls known to preserve the target. A silent participant does not remove measurement error. It moves the error into the bridge.

Figure 4. The evidence bridge from experience to a dataset label

Measurement anatomy

Every consciousness label is the end of a measurement chain. Validation must cover the same state, task and timing used in the neural analysis. A proxy validated in alert wakefulness is not automatically valid under anaesthesia. Practitioner synthesis.

Anaesthesia changes the instrument and the phenomenon

A perceptual mask changes visual input and processing. An anaesthetic changes far more. It alters arousal systems, neuromodulation, cortical dynamics, sensory connectivity, memory formation and motor responsiveness. These are not nuisance variables added after the fact. They are part of the intervention’s mechanism.¹² Therefore an awake-versus-anaesthetised brain difference is bundled by construction.

Behavioural unresponsiveness is especially dangerous as a state label. Reviews of anaesthesia distinguish connected consciousness, in which a person remains aware of the environment; disconnected consciousness, in which internal experience continues without environmental connection; and absent experience.¹¹ The same immobile body can occur in all three.

A repeated-awakening study during deep propofol sedation obtained positive reports of experience in 82 per cent of awakenings with interpretable responses. Yet commonly used complexity measures distinguished wakefulness from sedation rather than experience versus no experience within the sedated condition.¹⁶ This is a productive null. It suggests that a measure can be a state marker without being a moment-to-moment content marker. It also shows why state and content cannot be inferred from each other.

Figure 5. One propofol session can cross several latent conditions

State trajectory

Responsiveness, experience and later reportability are separate trajectories. A single post-recovery question compresses the entire interval into one uncertain label. Immediate repeated sampling reduces, but does not abolish, that temporal ambiguity. Illustrative trajectory informed by anaesthesia and repeated-awakening research.¹⁶¹⁷

Dream reports contain a time problem

Dream research gains leverage by waking people from both rapid-eye-movement and non-rapid-eye-movement sleep and asking immediately whether experience was present. Dream reports have been associated with reduced low-frequency activity in posterior cortex, and reported content has been linked to activity in relevant sensory regions.⁸ The result is important because sleep stage alone does not determine whether experience occurred.

The strongest boundary is equally important. Neural activity before awakening is compared with a report after awakening. The bridge assumes that the report reflects the sampled interval, that memory survived, and that “no experience” differs from “no recall”. Critics have argued that serial-awakening results can therefore track dream recall or reportability rather than dreaming itself.⁹ The right response is not to discard dream science. It is to type the claim: a pre-awakening neural pattern predicts a later report under a stated sampling protocol.

Build contrasts that survive interpretation

A good experiment does not seek one perfect contrast. It builds a contrast family: several comparisons that share the target phenomenon but alter the likely confounds. The candidate signal earns a stronger interpretation only when it survives the family in the predicted way.

The proposed contrast-bundle method has four operations:

Target. State exactly whether the claim concerns global state, specific content, access, reportability or a mechanism proposed by a theory.
Trace. Draw paths from the intervention to the neural measure and to the experience label. Mark prerequisites, consequences, direct physiology and selection effects.
Trade. Replace one bridge or manipulation with another. Report becomes an eye proxy; masking becomes rivalry; propofol becomes sleep; retrospective recall becomes repeated awakening.
Test. Ask whether the candidate signal generalises where the target is preserved and changes where the target is intervened on.

This is not a scoring scheme for consciousness. It is a discipline for making smaller, better-typed claims.

Name the estimand before naming the correlate

An estimand is the exact difference the study intends to learn. For content, a useful target is the change in a neural feature when red experience is present rather than absent, while global state and evidence bridge are held fixed. For state, the target is the change produced by moving between two global conditions while content class and bridge are matched as closely as possible. For report, the target is the change caused by requiring an explicit report while the percept is preserved.

Using the earlier notation, these are different quantities. The content estimand asks for ΔY when ΔC changes but ΔS and ΔB are approximately zero. The state estimand changes S while constraining C and B. The bridge estimand changes B while preserving S and C. A factorial design estimates all three and their interactions rather than forcing them into one binary contrast.

The interaction terms matter. A content code need not have identical amplitude during wakefulness, dreaming and sedation. The state may change how the same content is implemented, amplified or made available. Conversely, a global state marker may respond differently when rich internal content is present. Averaging over those interactions can create a marker that describes no actual condition. The practical rule is to preserve heterogeneity until the design demonstrates that pooling is justified. A universal NCC is a conclusion to earn, not a default model specification.

Figure 6. A minimal factorial design that separates three questions

Configured experiment

The cube does not make the labels true. It makes their dependencies testable. A content candidate should distinguish perceived from unperceived trials in both bridge modes; a state candidate should vary with state after content and bridge effects are modelled. Illustrative experimental specification.

Worked example 4 · A configured contrast family

One study, three linked assays

Assay A presents threshold-level images during alert wakefulness and collects immediate visibility reports. Assay B repeats the stimuli without trial reports, using an eye-based proxy calibrated in separate report blocks. Assay C repeats the no-report design during graded propofol sedation, with intermittent command-following checks and brief awakenings for immediate experience sampling. High-density EEG, eye tracking and drug concentration are time-aligned. The preregistration declares separate predictions for content, global state and proxy validity.

A posterior signal that distinguishes perceived from unperceived trials in A and B, and preserves content information in responsive sedation, becomes a stronger content candidate. A frontoparietal signal that appears only in A is typed as report, decision or access-related until further intervention. A complexity measure that separates wakefulness from sedation but not experience reports within sedation is typed as a state marker. No result is called “the NCC” without its contrast family.

Design the bridge before collecting the brain

For each cell, write a bridge contract. It should state what observation labels the target, how accurate that observation is under the same state, and which failure modes produce false positives or false negatives. An eye proxy may be compared with simultaneous reports during wakefulness and lighter sedation. A retrospective dream report should include awakening latency, uncertainty and temporal scope. A motor-imagery command should include sensory controls and a pre-specified classifier.

Then use negative controls. If an eye measure is the bridge, test stimuli that drive the eye response without the target percept. If a decoder identifies imagined tennis, test whether arousal or instruction onset alone produces the pattern. If a state marker tracks propofol concentration, test a different state-changing intervention. Bridge validation is part of the consciousness experiment, not a preliminary technical detail.

Figure 7. Evidence strengthens by crossing contrast families

Evidence matrix

The matrix rewards a coherent causal pattern, not a single large effect. The symbols are illustrative. A candidate can be useful as a state, access or motor marker even when it fails as a content correlate. Illustrative decision instrument.

Intervention without self-deception

Causal perturbation is stronger than passive correlation, but it also creates new bundles. Disrupting a cortical region can alter attention, confidence or response selection. Stimulation can spread. A global perturbation can change the entire state. Therefore a causal claim needs an intervention fingerprint: what changed, when, where, and which non-target capacities changed with it.

A credible necessity test should show that altering the candidate process changes the target experience more than matched control experiences, while preserving sensory input and the evidence bridge as far as possible. A restoration test is stronger: perturb the process, observe the predicted loss, then restore the process or compensate through a distinct route and recover the target. Even then, the result identifies a role within the tested system, not a metaphysical identity between neural activity and experience.

Necessity, sufficiency and replacement are different tests

A necessity experiment asks whether disrupting a candidate process removes or alters the target. Its strongest control is not another brain region chosen for convenience, but a manipulation matched for discomfort, attention, sensory disturbance and task difficulty. If both the target experience and the evidence bridge fail, the result cannot distinguish loss of experience from loss of expression.

A sufficiency experiment asks whether activating the process can produce the target under conditions where it would otherwise be absent. Evoking a phosphene, image fragment or report is stronger than observing a correlate, but the evoked event may recruit a wider network. The causal unit is the stimulated process plus the response it triggers, unless spread is measured and constrained.

A replacement experiment is more demanding. First perturb the candidate and observe the predicted deficit. Then restore the relevant organisation through a different route while keeping the original process impaired. Recovery would support the preserved causal role rather than a unique anatomical location. Failure to recover could reflect an incorrect role specification, an inadequate replacement or biological constraints. These tests move a claim upward for different reasons. None licenses the jump from “causally contributes in this preparation” to “is identical with experience”.

How the claim changes under pressure

Chart · state

stateDiagram-v2
  [*] --> Claim
  state "A condition name is<br/>compression, not explanation" as Claim
  state "Turn off response, not experience" as Test
  state "Figure 1. A state-content<br/>atlas, not a consciousness<br/>ladder" as Revision
  Claim --> Test: expose mechanism
  Test --> Decision: evidence survives
  Test --> Revision: rival survives
  Revision --> Claim: narrow or rebuild
  Decision --> [*]

The revision loop begins with A condition name is compression, not… and keeps counterexamples inside the reasoning path. Reading use: follow the relation, then return to the surrounding argument for evidence and limits.

What the evidence can honestly buy

Recent adversarial work illustrates the value of precise contrasts. A large preregistered, multimodal collaboration tested selected biological predictions of integrated information theory and global neuronal workspace theory using clearly visible stimuli whose category, identity, orientation and duration could be decoded. Some findings challenged sustained posterior synchronisation predicted by one implementation; others challenged the expected prefrontal ignition pattern at stimulus offset and the representation of certain contents predicted by the other.¹⁴

The result was not a ceremonial winner. The study tested particular predictions under a particular content-rich design. It did not compare conscious with unconscious trials, and the authors explicitly separated challenges to biological implementations from the theories’ broader computational cores. This is a model of good claim typing: the study says exactly which prediction met which observation and where alternative interpretation remains.

Human single-neuron recordings provide another useful boundary. Activity in lateral occipital cortex has tracked visual awareness across masking, flash suppression and binocular rivalry, with population decoding favouring the consciously perceived stimulus.¹⁵ That cross-paradigm convergence strengthens the content link. It does not by itself prove that those neurons are sufficient for experience, nor that lower-level feature differences have been eliminated in every condition.

Evidence status

What the paper inherits and what it proposes

Published evidence. Visibility, rivalry, no-report, awakening and anaesthesia studies show that neural features can track content, condition, report demand or broad state under specified protocols. Intracranial and multimodal work strengthens localisation and temporal claims without removing the measurement bridge.

Contested interpretation. Researchers disagree about which recurrent, posterior, frontal and global patterns belong to experience itself, its prerequisites or its consequences. Dream absence, unresponsiveness and proxy-labelled perception remain especially bridge-sensitive.

Practitioner synthesis. The contrast bundle, explicit estimands and five-step claim ladder are decision tools developed here from those methodological constraints. They are not a validated consciousness scale.

Open hypothesis. A content correlate should retain content-specific structure across contrast families and evidence bridges, then fail in a target-specific way under controlled perturbation.

A five-step claim ladder

Condition-linked marker. The feature differs between two experimental conditions.

Bridge-validated correlate. The feature tracks a target state or content under an explicitly validated evidence bridge.

Contrast-family invariant. The feature follows the target across several manipulations while predicted confounds change.

Causal component. Targeted perturbation and restoration show that the process contributes to the target within the tested system.

Constitutive or sufficient mechanism. A theory explains why the complete mechanism realises the conscious property and survives serious alternatives. Neural correlation alone does not reach this step.

Most published findings belong on the first three steps. That is not a failure. State markers can guide diagnosis or monitoring. Report markers can reveal access and metacognition. Content decoders can test representation. Problems arise when a result climbs the ladder in the abstract while the experiment stayed on a lower rung.

Figure 8. One null result, four incompatible explanations

Failure paths

A null becomes informative only after sensitivity, bridge validity and temporal alignment are demonstrated. Otherwise “no NCC” is one of several explanations, not the result itself. Practitioner synthesis.

Optional depth: what the same NCC result means under rival metaphysical views

View	Strong interpretation of an NCC	What remains unpaid
Physicalism	The neural process is identical with, realises or fully determines the experience.	Why that process has the claimed phenomenal character.
Biological naturalism	Consciousness is a higher-level biological feature caused by organised neural activity.	Which biological properties are necessary and why artificial realisation may fail.
Functionalism	The NCC instantiates a causal role that could be realised in another substrate.	A complete role specification and a preservation test across realisers.
Emergentism	Consciousness appears when network organisation crosses relevant systemic conditions.	A non-arbitrary emergence mechanism and intervention-sensitive threshold.
Idealism	Neural measurements are regular structures within experience, not its ultimate producer.	Why these structures lawfully track private and shared experience.
Neutral monism	Brain and experience are two organisations or descriptions of a more basic neutral process.	The mapping from neutral structure to both descriptions.
Panpsychism	Neural organisation combines or structures more basic experiential properties.	The combination problem and a test distinguishing combination from ordinary function.
Dual-aspect / process views	Neural and experiential patterns are inseparable aspects or phases of one evolving process.	A precise account of aspect relations, temporal individuation and causal intervention.

Executable lab

The contrast-bundle audit

The instrument below does not calculate a consciousness score. It checks whether the proposed claim is licensed by the design. Choose the target and answer the design questions. The output lists unresolved causal passengers, the strongest permitted claim and the smallest redesign likely to improve interpretation.

Figure 9. Interactive NCC contrast designer

Executable decision tool

Select the closest description of the planned comparison. “Partial” should be used when a factor is measured but not matched.

Target claim

Physical input matched?

Report and decision demand matched?

Bridge validated in this state?

Memory and timing controlled?

Intervention has direct neural effects?

Within-participant or well matched?

Independent contrast family included?

Targeted perturbation or restoration?

Sensitivity positive control?

Result will appear here

The audit reports interpretability, not whether consciousness is present.

A design passes only relative to its target claim. The tool prioritises unresolved paths over a single numerical score and exports the reasoning as structured JSON. Executable practitioner synthesis. Runs locally in this page.

How to read the audit

A positive result means the design supports a better-typed claim, such as “bridge-validated content correlate across two contrast families”. It does not establish consciousness, metaphysical identity or universal necessity. A negative result means the current contrast cannot separate the target from one or more travelling causes. The result is a redesign instruction, not evidence that the target phenomenon is absent.

The tool encodes several assumptions: matching or modelling a cause reduces its confounding path; a proxy must be validated in the target state; direct intervention effects require convergent manipulation; causal mechanism claims require perturbation and suitable controls; and null claims require a positive sensitivity control. These assumptions are deliberately conservative.

Python · contrast_bundle_audit.py

from dataclasses import dataclass, asdict
from typing import Literal

Status = Literal["yes", "partial", "no", "proxy"]
Target = Literal["content", "state", "access", "mechanism"]

@dataclass(frozen=True)
class ContrastDesign:
    target: Target
    input_match: Status
    report_match: Status
    proxy_valid: Status
    memory_control: Status
    direct_effect: Status
    within_subject: Status
    contrast_family: Status
    perturbation: Status
    positive_control: Status


def audit(d: ContrastDesign) -> dict:
    issues: list[str] = []
    redesign: list[str] = []

    if d.target == "content" and d.input_match != "yes":
        issues.append("Physical input can explain the content contrast.")
        redesign.append("Add a stimulus-matched visibility manipulation.")

    if d.report_match == "no":
        issues.append("Report, decision and motor consequences differ.")
        redesign.append("Match report demand or add a no-report contrast.")

    if d.report_match == "proxy" and d.proxy_valid != "yes":
        issues.append("The no-report proxy is not validated in the target state.")
        redesign.append("Calibrate the proxy against reports in the same state.")

    if d.memory_control != "yes":
        issues.append("Negative reports may reflect forgetting or temporal smear.")
        redesign.append("Sample experience immediately or model report timing.")

    if d.direct_effect == "yes":
        issues.append("Direct intervention physiology is bundled with the target.")
        redesign.append("Use a convergent intervention or dose-matched control.")

    if d.within_subject == "no":
        issues.append("Between-person differences can drive the neural contrast.")
        redesign.append("Use within-participant contrasts or stronger matching.")

    if d.contrast_family == "no":
        issues.append("A single contrast cannot test invariance to confounds.")
        redesign.append("Add an independent manipulation preserving the target.")

    if d.target == "mechanism" and d.perturbation != "yes":
        issues.append("A mechanism claim lacks controlled causal perturbation.")
        redesign.append("Add perturbation, negative controls and restoration.")

    if d.positive_control == "no":
        issues.append("A null result would be uninterpretable for sensitivity.")
        redesign.append("Add a positive control known to produce a detectable effect.")

    if not issues:
        permitted = "Causally informative, contrast-family-supported candidate"
    elif len(issues) <= 2 and d.contrast_family != "no":
        permitted = "Bridge-validated or condition-linked candidate with named limits"
    else:
        permitted = "Condition difference only; not yet an NCC claim"

    return {
        "design": asdict(d),
        "permitted_claim": permitted,
        "unresolved_paths": issues,
        "smallest_redesigns": list(dict.fromkeys(redesign)),
        "cannot_establish": [
            "that consciousness is present or absent",
            "metaphysical identity between neural activity and experience",
            "universal necessity outside the tested system",
        ],
    }

Save the code as contrast_bundle_audit.py, construct a ContrastDesign, and print audit(design). The browser version uses the same rule family. Extended tests belong in a separate harness because the article’s main contribution is the reasoning exposed by the rules, not interface code.

Optional depth: compact test harness

Python · minimal tests

def test_single_report_contrast_is_bundled() -> None:
    d = ContrastDesign(
        target="content", input_match="yes", report_match="no",
        proxy_valid="no", memory_control="yes", direct_effect="no",
        within_subject="yes", contrast_family="no",
        perturbation="no", positive_control="yes",
    )
    result = audit(d)
    assert "Condition difference only" in result["permitted_claim"]
    assert any("Report" in x for x in result["unresolved_paths"])


def test_controlled_mechanism_design_can_reach_top_tier() -> None:
    d = ContrastDesign(
        target="mechanism", input_match="yes", report_match="yes",
        proxy_valid="yes", memory_control="yes", direct_effect="no",
        within_subject="yes", contrast_family="yes",
        perturbation="yes", positive_control="yes",
    )
    result = audit(d)
    assert result["unresolved_paths"] == []
    assert result["permitted_claim"].startswith("Causally informative")

The evidence conversation

Chart · sequence

sequenceDiagram
  participant Q as Question
  participant M as Mechanism
  participant E as Evidence
  participant D as Decision
  Q->>M: The ladder that hides the question
  M->>E: A condition name is compression, not explanation
  E-->>M: compare rival explanations
  E->>D: Unbundle the contrast before naming the neural correlate
  D-->>Q: revise scope or act

Question, mechanism, evidence and decision exchange the burdens developed in States, contents and neural correlates. Reading use: follow the relation, then return to the surrounding argument for evidence and limits.

Unbundle the contrast before naming the neural correlate

Do not begin a consciousness study by asking which brain signal separates two named conditions. Begin by asking what exactly changes between them. Is the target the organism’s global state, a particular content, access to that content, the ability to report it, or a mechanism predicted by a theory? Which observation makes the target visible to the experimenter? Which other causes move with the intervention?

The practical decision is simple: replace the search for a single decisive NCC contrast with a programme of contrast unbundling. Use at least one within-state content contrast, one alternative evidence bridge, one convergent state manipulation where relevant, and one perturbation or negative control matched to the claim. Type every result at the rung it has earned.

This changes how nulls are read, how biomarkers are described and how theories are tested. Unresponsiveness becomes a behavioural fact, not a verdict about experience. A later “no dream” becomes a memory-dependent measurement, not a transparent window into the prior interval. A no-report signal becomes proxy-conditioned evidence, not purified consciousness. A decoder becomes a predictor until intervention reveals its role.

The science advances when the contrast becomes narrower than the phenomenon and the claim becomes narrower than the data. That restraint does not weaken consciousness research. It gives each neural difference a job it can actually perform.

Glossary

Global state: The organism’s overall conscious condition, described across several functional and content-related dimensions rather than assumed to be one scalar level.
Conscious content: A particular experience, such as seeing a red square, hearing a tone or having a dream image.
Evidence bridge: The measurement chain used to infer an unobserved experience from report, behaviour, physiology or a trained classifier.
NCC proper: A candidate neural process directly associated with the experience rather than merely preceding or following it.
Contrast bundle: The complete set of causal changes that differ between experimental conditions, including target and non-target paths.
Contrast family: Several experiments that preserve the target while varying likely confounds, bridges or interventions.
Disconnected consciousness: Experience that continues without reliable awareness of or response to the external environment.

References

Chalmers, D. J. (2000). What is a neural correlate of consciousness? In T. Metzinger (Ed.), Neural Correlates of Consciousness. MIT Press. Author manuscript.
Aru, J., Bachmann, T., Singer, W., & Melloni, L. (2012). Distilling the neural correlates of consciousness. Neuroscience & Biobehavioral Reviews, 36(2), 737–746. doi:10.1016/j.neubiorev.2011.12.003.
Bayne, T., Hohwy, J., & Owen, A. M. (2016). Are there levels of consciousness? Trends in Cognitive Sciences, 20(6), 405–413. doi:10.1016/j.tics.2016.03.009.
Koch, C., Massimini, M., Boly, M., & Tononi, G. (2016). Neural correlates of consciousness: progress and problems. Nature Reviews Neuroscience, 17, 307–321. doi:10.1038/nrn.2016.22.
Tsuchiya, N., Wilke, M., Frässle, S., & Lamme, V. A. F. (2015). No-report paradigms: extracting the true neural correlates of consciousness. Trends in Cognitive Sciences, 19(12), 757–770. doi:10.1016/j.tics.2015.10.002.
Frässle, S., Sommer, J., Jansen, A., Naber, M., & Einhäuser, W. (2014). Binocular rivalry: frontal activity relates to introspection and action but not to perception. Journal of Neuroscience, 34(5), 1738–1747. doi:10.1523/JNEUROSCI.4403-13.2014.
Hesse, J. K., & Tsao, D. Y. (2020). A new no-report paradigm reveals that face cells encode both consciously perceived and suppressed stimuli. eLife, 9, e58360. doi:10.7554/eLife.58360.
Siclari, F., Baird, B., Perogamvros, L., et al. (2017). The neural correlates of dreaming. Nature Neuroscience, 20, 872–878. doi:10.1038/nn.4545.
Ruby, P. M. (2020). The neural correlates of dreaming have not been identified yet. Frontiers in Neuroscience, 14, 585470. doi:10.3389/fnins.2020.585470.
Kronemer, S. I., Aksen, M., Ding, J. Z., et al. (2022). Human visual consciousness involves large scale cortical and subcortical networks independent of task report and eye movement activity. Nature Communications, 13, 7342. doi:10.1038/s41467-022-35117-4.
Montupil, J., Cardone, P., Staquet, C., et al. (2023). The nature of consciousness in anaesthesia. BJA Open, 8, 100224. doi:10.1016/j.bjao.2023.100224.
Mashour, G. A. (2024). Anesthesia and the neurobiology of consciousness. Neuron, 112(10), 1553–1567. doi:10.1016/j.neuron.2024.03.002.
Pérez, P., Manasova, D., Hermann, B., et al. (2024). Content–state dimensions characterize different types of neuronal markers of consciousness. Neuroscience of Consciousness, 2024(1), niae027. doi:10.1093/nc/niae027.
Cogitate Consortium. (2025). Adversarial testing of global neuronal workspace and integrated information theories of consciousness. Nature, 642, 133–142. doi:10.1038/s41586-025-08888-1.
Vanhoyland, M., Janssen, P., & Theys, T. (2025). Single-neuron correlates of visual consciousness in human lateral occipital complex. Nature Communications, 16, 11008. doi:10.1038/s41467-025-67077-w.
Bajwa, I. J., Nilsen, A. S., Skukies, R., et al. (2025). A repeated awakening study exploring the capacity of complexity measures to capture dreaming during propofol sedation. Scientific Reports, 15, 32746. doi:10.1038/s41598-025-12695-z.
Huang, Z., Tarnal, V., Fotiadis, P., et al. (2026). An open fMRI resource for studying human brain function and covert consciousness under anesthesia. Scientific Data, 13, 127. doi:10.1038/s41597-025-06442-2.
Stockart, F., Robin, A., Blumenfeld, H., et al. (2026). Neural correlates of perceptual consciousness from within: a narrative review of human intracranial research. eLife, 15, RP109604. doi:10.7554/eLife.109604.4.