A person prepares to reach for a glass, then stops before the muscles move. A motor plan was forming, the neural state was changing, and perhaps the probability of action was rising. A fraction of a second later, the glass remains untouched. The stop might have been another physical event that began before the person felt they had decided. It might have expressed an old habit of restraint. Or the person might, at that moment, have authored a different future.
We can study parts of this sequence by recording the brain’s preparation for movement, measuring inhibition, perturbing electrical rhythms, and modeling the accumulation of evidence toward an action. None of these measurements alone settles the question of free will. Neural causation and metaphysical authorship are distinct questions.
This article proposes a research program called Conditional Ontological Agency Theory (COAT). It separates the debate into models with stated assumptions, identifiable predictions, credible null hypotheses, and limits on what evidence can establish. Its starting hypothesis is that, if stronger-than-ordinary free will exists, it may occur at only a few pivotal moments. On this account, most of a human life unfolds through physical processes and habits shaped by the past.
The proposal also considers two less conventional possibilities. A subject might have some authority over which life it experiences, even if events within that life are fixed. Conscious agency, if causally active, might change the probabilities of physical events without guaranteeing particular outcomes. Both remain conditional hypotheses. Stating them explicitly makes their assumptions available for scrutiny.
The near-term program concerns measurable neural processes. We propose prospective EEG experiments on voluntary veto, computational tests of intermittent control, a field-mediated neural-control study, and a longitudinal study of how deliberate inhibition becomes habitual. These experiments would test how decisions are redirected. They could not establish the existence of an immaterial soul. The stronger ontological claims remain open until a proposed mechanism yields predictions that distinguish them from ordinary physics.
The program asks where, if anywhere, authorship enters the causal account, what it can change, and how much of what follows is the physical consequence of that change. These questions extend the usual debate over whether people have free will.
The research question
Most debates begin with the legitimate philosophical question of whether freedom is compatible with determinism. The word “freedom,” however, can refer to several things: a person’s capacity to deliberate, the existence of more than one physically possible future, an experiencing subject’s authority over which future occurs, or an earlier choice’s influence on the probability of present behavior. These need to be considered separately.
COAT separates four dimensions:
| Dimension | Question | Scientific status |
|---|---|---|
| Existential selection ($E$) | Could a subject authentically select which life or experiential trajectory it occupies? | Metaphysical and presently without a discriminating empirical test |
| Local intervention ($I_t$) | Could an authentic choice alter the unfolding physical probabilities or causal trajectory at a particular time? | Conditional; some proposed neural dynamics are testable, ultimate authorship is not yet identified |
| Inherited authorship ($L_t$) | Could an earlier genuine choice continue shaping later habits when no fresh choice occurs? | The learning and persistence mechanism is testable; genuine prior authorship is not directly observable |
| Functional agency ($G_t$) | Can an embodied system deliberate, inhibit, learn, and cause outcomes through its own mechanisms? | Empirically tractable, independent of a commitment to libertarian free will |
Freedom over a life, freedom within a life, and effective agency within a life are logically distinct. COAT provides terms for comparing models that are often treated as competing explanations even when they address different questions. It does not propose a single mechanism to explain all four dimensions.
The main empirical question concerns cancellation:
When a developing action is canceled, do prospectively measured brain dynamics support only familiar stochastic preparation and inhibition, or is there a reproducible additional pattern of brief, state-dependent control that improves out-of-sample prediction?
A positive answer would establish a finding about neural control. Establishing the source of the controlling event would require a separate argument.
Earlier work and the proposed contribution
The philosophical claims have substantial precedents. Robert Kane’s work on self-forming actions explains how comparatively rare undetermined choices might help shape a person’s later character.1 Timothy O’Connor’s agent-causal account addresses the more demanding requirement that an action originate in an agent, beyond its place in a chain of events.2 Christian List’s higher-level account allows for meaningful alternatives and control at the agential level; critics question whether levels of description can supply the relevant metaphysical openness.3 Mark Balaguer argues that some questions about free will may be scientific questions that science has yet to settle.4
Belnap, Perloff, and Xu provide formal tools for discussing action, branching histories, and the conditions under which an agent sees to it that something happens.5 The idea of choosing one’s life appears in Plato’s Myth of Er and later Neoplatonic thought. It provides a philosophical precedent for distinguishing the choice of an existence from control over every event within it.6 Contemporary work on self-locating belief represents a first-person location within a world without implying that souls select bodies.7
Peter Ulric Tse’s criterial causation offers a physicalist account of brain-level control in which neuronal conditions, information, and indeterminism may support meaningful mental causation.8 Causal-emergence research and information-theoretic empowerment provide other operational accounts of effective organization and control.910
This literature does not prove the COAT hypothesis. COAT’s proposed contribution is to combine an explicit conditional ontology, a distinction between present and inherited authorship, an identifiable neural-control testing program, and a competing electromagnetic pathway. A published interpretation matrix would set limits on the metaphysical conclusions drawn from experimental results.
Conditional ontologies
A model of free will must state what kinds of things it allows to exist. COAT leaves open whether consciousness reduces to physics and whether a subject can have an independent causal role.
Let $X_t$ denote the complete relevant physical state of the world at time $t$, including the brain. In models that require a hypothetical subject-level state, let $S_t$ denote that state. We distinguish three ontology classes:
$$ \begin{aligned} \mathcal O_0 &: \text{physical processes only},\\ \mathcal O_1 &: \text{physical processes and a noncausal experiencing subject},\\ \mathcal O_2 &: \text{physical processes and a causally active subject}. \end{aligned} $$These classes describe possible models; they are not discoveries. Irreducible experience alone would not imply the power to change matter, since subjecthood and authorship are distinct. A soul would also require an account of authorship. If its choices were determined by unchosen conditions, the same explanatory problem would remain.
“Could have done otherwise” also has three distinct meanings. A physically open alternative means that the complete previous physical state and laws permit different continuations. A counterfactual alternative means that a change in a reasons-state, instruction, or other input would change the action. A trajectory-level alternative means that more than one life or experiential center could have been occupied under the stipulated ontology. The latter two can be coherent even without a physically open alternative, because they answer different questions.
A determinist and a libertarian can therefore agree that a person has powerful executive control while disagreeing about whether the person could have made a fundamentally different choice with the entire physical past held fixed.
The formal core of COAT
The mathematics specifies what each proposed mechanism must explain. Formalizing a metaphysical possibility does not establish it experimentally.
Physical histories and experienced histories
Let $\mathcal H$ be the set of physical histories allowed by a model’s laws and boundary conditions. A history $h$ is a sequence of physical states. A deterministic model specifies a function $F$, and a stochastic model specifies a transition kernel $K_0$:
$$ X_{t+1}=F(X_t) \quad\text{or}\quad X_{t+1}\sim K_0(\cdot\mid X_t). $$The kernel $K_0$ describes the baseline physical account. It can include synapses, inhibitory control, neuromodulators, learning, electrical fields, and ordinary stochasticity. The account must represent these mechanisms adequately; the brain’s behavior cannot be reduced to following the path of least resistance.
We represent first-person perspective separately through a centered history:
$$ c=(h,i,t),\qquad c\in\mathcal C, $$where $i$ identifies an observer-position within history $h$ at time $t$. This representation comes from the logic of self-location and does not establish a subject existing independently of biology.7
For any life, the agency state has four coordinates:
$$ \boxed{\mathbf F_t=(E,I_t,L_t,G_t).} $$Here $E$ denotes authentic existential selection if it exists, $I_t$ a presently authored intervention, $L_t$ the later causal effects of an earlier authored intervention, and $G_t$ functional control by the embodied organism. The configuration $I_t=0$ with $L_t>0$ and $G_t>0$ allows an old decision to shape an automatic present response. If one accepts the pre-embodiment premise, $E=1$ can also coexist with $I_t=0$ throughout a deterministic life.
Where an intervention would enter
Ordinary evolution and a proposed local intervention differ operationally as follows:
$$ \begin{aligned} K_\theta(\cdot\mid X_t,U_t=\varnothing)&=K_0(\cdot\mid X_t),\\ K_\theta(\cdot\mid X_t,U_t=u)&\ne K_0(\cdot\mid X_t) \quad\text{for at least one }u. \end{aligned} $$$U_t$ is an intervention variable postulated by the theory, not an observed EEG signal. One measure of its possible causal influence over a future action $A$ is the greatest difference between action distributions generated by two candidate interventions:
$$ \Delta_t(x)=\sup_{u,v}D_{\mathrm{TV}}\!\left[ P(A\mid x,\operatorname{do}(U_t=u)), P(A\mid x,\operatorname{do}(U_t=v))\right]. $$$D_{\mathrm{TV}}$ is total-variation distance, and $\operatorname{do}(\cdot)$ denotes an idealized causal intervention as used in structural causal models.11 If $\Delta_t=0$, changing $U_t$ has no action-level efficacy in the model. A positive value means that interventions affect modeled outcomes. Neither result alone establishes whether the intervention variable is authentically chosen.
Authentic choice remains the central problem. Define $\operatorname{Auth}(S_t,U_t)$ to mean that the subject is the ultimate author of an intervention. A complete libertarian theory would need to explain this relation without equating it with randomness alone, a post-hoc feeling, or a prior determining chain the agent never owned. COAT retains this requirement; a stochastic differential equation does not resolve it.
Conditional propositions
The definitions imply the following conditional conclusions, none of which is an empirical theorem:
- If complete physical causal closure is assumed and subject-level variables do not change the physical kernel, independent local metaphysical intervention has zero modeled physical efficacy.
- If multiple deterministic histories or experiential centers exist and a subject can authentically select one, deterministic events within the selected life are logically consistent with existential freedom. The selection assumption remains unsupported by direct evidence.
- If an earlier authored intervention changes learning and later actions, present behavior can inherit causal effects even when no further intervention occurs.
- Averaging over an unobserved intervention process can produce an outcome distribution consistent with ordinary physical dynamics. Two theories may therefore be observationally equivalent even if their proposed ontologies differ.
For example, a hidden subject might change an outcome distribution to 60/40 in some trials and 40/60 in an equal number of others. An experimenter unable to observe or classify the interventions may see only 50/50. These aggregate frequencies alone would not establish that an intervention had occurred.
Intermittent authorship and its later consequences
The original intuition is that many human actions follow learned patterns. Habits, preferences, inhibitions, fears, commitments, and skills influence later decisions without requiring a new act of conscious will each time.
The strong version of COAT asks whether some of these patterns could have originated in rare, authentically authored choices. A difficult decision might change someone’s future tolerance for risk or willingness to help another person. Years later, the person may respond almost automatically. Learned disposition would be the immediate cause; depending on the ontology, an earlier authentic intervention could be among its distant causes.
This proposal resembles Kane’s self-forming actions and adds a time-indexed distinction between current intervention and inherited consequence.1 Let $W_t$ represent learned dispositions and policy parameters:
$$ W_{t+1}=\mathcal L(W_t,X_t,A_t,R_t), $$where $A_t$ is action and $R_t$ its feedback or reinforcement. Earlier events can alter the trajectory of $W_t$. An inherited-agency measure describes how strongly a later action depends counterfactually on an earlier intervention under common prior conditions:
$$ L_t=\sup_{\tauHistorical causal dependence alone does not establish moral responsibility, blame, or desert. A choice may affect later behavior, but coercion, available information, harm, capacity, and fairness require separate ethical analysis.
Workstream 1: COAT-VETO-1 and action cancellation
The principal near-term study is a high-density EEG experiment on voluntary motor inhibition. It asks whether someone can veto a movement already being prepared, using prospective prediction, explicit model comparison, and closer examination of self-initiated stopping.
What earlier experiments established
Libet and colleagues reported that a readiness potential preceding a voluntary movement can begin before the participant’s reported awareness of deciding to act.12 This challenged simple accounts in which a conscious command precedes the brain’s execution of it. The experiment did not establish that all choices are unconscious or that every decision is settled at the onset of an averaged EEG signal.
Schurger, Sitt, and Dehaene showed that threshold-crossing stochastic accumulation can reproduce a readiness-potential pattern without the early hidden decision often inferred from it.13 Using a real-time EEG brain–computer interface, Schultze-Kraft and colleagues showed that movements can be stopped after preparation has begun. Their task also showed a time-dependent point beyond which cancellation was unsuccessful.14 They used externally presented stop cues, so their result concerns a different condition from self-initiated veto.
Studies of intentional inhibition by Brass and Haggard, deliberate versus arbitrary choice by Maoz and colleagues, and veto paradigms by Shum, Galang, and Brass support the need for these distinctions.151617 A 2023 review cautions against using retrospective reports of the moment of intention, often called W-time, as precise clock readings of the start of consciousness.18
The proposed experiment tests whether conventional stochastic accumulation and inhibitory control adequately describe the dynamics of self-initiated cancellation. An ERP component could not establish the presence of a soul.
Design and recruitment
The initial protocol would use 128–256-channel EEG, surface EMG on the responding muscles, electrooculography, a low-latency response sensor, and synchronized event logging. A pilot cohort would establish signal quality, viable veto frequency, and decoder accuracy. We provisionally plan a main cohort of roughly 80–120 adults. Simulation-based power analysis would determine the final sample, with preregistered exclusion rules and repeated sessions to assess reliability.
EEG provides the temporal resolution needed here; it does not reveal microscopic quantum states. EMG is essential for detecting partial motor bursts when no completed button press occurs. Because the events are separated by tens to hundreds of milliseconds, event clocks and latency calibration must be accurate. Timing artifacts could produce a false version of the pattern being studied.
The experiment uses five conditions with distinct control functions:
| Condition | Task | Why it exists |
|---|---|---|
| Spontaneous action | Move at a self-chosen time within a defined window | Estimate ordinary motor preparation and execution |
| Self-initiated veto | Begin preparing but remain able to cancel without an external stop instruction | Primary condition for internally generated inhibition |
| Externally cued stop | An unpredictable cue instructs cancellation | Benchmark ordinary reactive inhibition |
| Mandatory-go / no-veto | Participants know that the prepared response must be executed | Measure how merely having a veto option changes preparation |
| Consequential choice | Choose or revise a small real allocation, donation, or reward option | Test whether arbitrary-movement results generalize to choices with consequences |
A candidate self-veto trial must show prospective evidence of movement preparation, a subsequent change in trajectory, and no completed action. The absence of a press alone is insufficient. The criteria will be defined before outcome comparison to avoid selecting only trials that look interesting.
The prospective decoder
The decoder estimates the probability of movement in a future window using only data already available at the prediction time:
$$ \hat p_t=P(\text{movement in }[t,t+H]\mid Y^{\mathrm{EEG}}_{0:t}). $$An initial prediction horizon might be $H=500$ milliseconds, subject to refinement in the pilot. Real-time preprocessing must be causal: zero-phase filters that use future data would invalidate prospective detection. Decoder training and hyperparameter selection should be nested within participant-level or session-level held-out splits. A canceled movement has no observed movement time, so trials must not be aligned to an invented estimate of when it would have occurred. Decoder state or trigger time provides the common reference.
Stimulus-triggered and yoked controls will account for the behavioral effect of a decoder-based stop cue. If a cue appears when preparation is high, any subsequent difference may reflect the cue as well as the participant’s internal state. Interpretation of the self-veto and externally cued conditions must account for this asymmetry.
Three competing explanatory models
The study should compare explanatory models. Unfamiliar scalp patterns alone would not establish a new mechanism.
The baseline model, $M_0$, includes both stochastic motor accumulation and ordinary inhibitory control. Let $z_t$ be a preparation variable and $h_t$ an inhibitory variable:
$$ \begin{aligned} dz_t &= (\mu-\kappa z_t)\,dt+\sigma_z\,dW_t,\\ dh_t &= [-\eta h_t+g(X_t,C_t)]\,dt+\sigma_h\,dB_t,\\ \text{movement} &\iff z_t-h_t\ge a. \end{aligned} $$$C_t$ represents task context. $M_0$ can generate stops through ordinary physical dynamics. Omitting inhibitory control from the baseline would risk misidentifying a known process as a new phenomenon.
The operational intermittent-control model, $M_1$, adds a sparse event process $J_t$:
$$ dz_t=(\mu-\kappa z_t-\beta J_t)\,dt+\sigma_z\,dW_t. $$Here $J_t$ represents a brief, state-dependent change in control dynamics. It may be implemented as a latent point process with preregistered duration and regularization. We would test whether it reproducibly improves held-out predictions of canceled versus completed actions, EMG bursts, and relevant EEG features.
The metaphysical model, $M_2$, adds $\operatorname{Auth}(S_t,J_t)=1$, meaning that the extra event is authentically authored. This claim yields no unique EEG prediction unless the theory specifies an additional measurable signature. COAT-VETO-1 therefore compares $M_0$ with constrained versions of $M_1$. It cannot directly adjudicate $M_2$ through EEG.
Other competitors will include flexible change-point models, state-dependent stopping thresholds, time-varying physical inhibition, and alternative accumulator architectures. Support requires outperforming these credible alternatives; beating an artificially weak baseline is insufficient.
Preregistered hypotheses and endpoints
The preregistered hypotheses concern testable observations:
- H1, late cancellation: Some actions with clear prospective preparation can be canceled before execution. We will estimate cancellation probability conditional on preparation state, without assuming a universal 200-ms boundary.
- H2, self-initiated versus reactive stopping: Self-veto and externally cued stopping produce reliably distinguishable dynamics after matching on measured preparation and task context.
- H3, sparse-control model comparison: $M_1$ improves out-of-sample joint prediction of behavior, EMG, and prespecified EEG features over $M_0$ and other physical alternatives.
- H4, decision significance: Model parameters and veto signatures differ, or demonstrably generalize, across arbitrary movement and meaningful allocation tasks.
- H5, stability: The key effects are observable in a second session and are not artifacts of post-hoc trial selection.
The primary inferential endpoint is the difference in expected log predictive density (ELPD) on held-out data:
$$ \Delta_{\mathrm{pred}}=\operatorname{ELPD}(M_1)-\operatorname{ELPD}(M_0). $$Before confirmatory data collection, we will specify the minimum practically relevant predictive gain, an uncertainty criterion, and the requirement to outperform the strongest prespecified physical competitors. Secondary endpoints include cancellation probability conditioned on decoder state, stop latency, partial EMG frequency, and time-resolved spatial/oscillatory signatures. A hierarchical model will account for trial-level dependence across participants and sessions. Discovering a neural feature after inspecting outcomes will not, by itself, give that feature evidentiary weight.
Subjective awareness reports provide supporting evidence. Sparse random probes and structured post-trial descriptions can help distinguish “I intended to move and canceled” from “I never formed that intention,” but cannot directly measure metaphysical authorship. In the deliberate task, matching neural preparation across two kinds of decisions still leaves the internal choice to veto unrandomized.
Limits on interpretation
If the sparse-control model outperforms the baseline, we could conclude that this temporally localized representation improves the physical account of voluntary inhibition. It might represent an unmodeled neural controller, attention, a change in inhibitory threshold, or a better statistical description of ordinary control. If the baseline performs as well or better, the stronger operational claim loses support for these tasks. Neither outcome proves or refutes existential selection or all forms of metaphysical freedom.
Workstream 2: COAT-EM and electrical field effects on veto
The original theory considered quantum-level probability intervention as a possible way for an experiencing subject to redirect physical events. Neuroscience offers a more immediate physical candidate for steering high-dimensional brain activity: electric fields generated by neural activity may feed back onto the neurons that generate them.
There is experimental evidence for ephaptic coupling. Anastassiou and colleagues found that weak extracellular fields can affect membrane potential and spike timing under physiological experimental conditions.19 Pinotsis and Miller investigated whether comparatively stable electric-field representations could constrain and organize variable populations of neurons despite drift in the underlying neural ensembles.20 The cytoelectric coupling hypothesis proposes further connections to cellular and cytoskeletal organization; these stronger claims remain under investigation.21 A 2026 modeling study by Pinotsis and Miller links trial-to-trial oscillatory variability to inferred field–neuron interactions in prefrontal recordings.22
These findings provide a mechanistic basis for targeted experiments on field-sensitive neural control. They do not establish field consciousness, an entity separate from the nervous system, or an ability to override physics. McFadden’s conscious electromagnetic information theory makes substantially stronger claims and requires separate evaluation.23
The candidate mechanism
Let $\mathbf n_t$ denote a population-level neural state and $\mathbf e_t$ a coarse-grained endogenous electric-field state. A minimal model of their reciprocal effects is:
$$ \begin{aligned} d\mathbf n_t &= [f(\mathbf n_t)+\gamma C(\mathbf e_t)]\,dt+\Sigma\,d\mathbf W_t,\\ \mathbf e_t &=\mathcal E[\mathbf n_{0:t},\text{tissue geometry}]. \end{aligned} $$The parameter $\gamma$ quantifies the field-feedback contribution explicitly represented in the model. It does not describe all electrical properties of neurons. Setting $\gamma=0$ removes this coupling route from a reduced model while leaving the neurons’ other electrical properties intact.
The proposed mechanism is a physical feedback loop. The network generates a field, the field alters neuronal excitability or timing, and the altered neurons produce a different field. This loop could produce complex causal organization in a deterministic system, without an external nonphysical influence on synapses.
Testing field effects through intervention
A 2024 study found that motor-inhibition performance varied with the phase of 20-Hz stimulation over the pre-supplementary motor area.24 A 2026 preprint reported a related approach using closed-loop phase-locked EEG–tACS; this remains preprint evidence.25 Another 2026 study changed frontomotor connectivity through phase-specific dual-site stimulation without a significant overall improvement in stopping performance.26 Changes in brain signals therefore need to be assessed separately from changes in behavior.
COAT-EM would use these constraints to design a secondary study arm, subject to ethics approval:
- Estimate individual beta-band phase, spatial field-related signatures, and conventional preparation/inhibition signals during self-veto and cued-stop trials.
- Randomly assign sham, preregistered phase-specific, and suitable off-frequency or phase-control stimulation conditions where safe and technically valid. Use individualized field modeling and blinding checks.
- Evaluate stopping success, EMG evidence of partial execution, response timing, and adverse sensations. Field or oscillation measures serve as mechanistic secondary outcomes; they cannot substitute for behavioral measures.
- Test phase and site specificity in an independent session. Separate effects during stimulation from those that persist after stimulation stops.
An illustrative analysis would use:
$$ \operatorname{logit}P(V_t=1)=\alpha+\beta z_t+ \gamma\cos(\phi_t-\phi_0)+\delta^\top C_t, $$where $V_t$ indicates successful cancellation, $z_t$ summarizes motor preparation, $\phi_t$ is the relevant phase, and $C_t$ contains preregistered covariates. The primary estimand for a causal phase test is the difference in stopping outcomes between randomly assigned stimulation conditions. The regression coefficient of observational phase alone cannot establish that effect.
Stimulation artifacts can contaminate oscillatory estimates, making simultaneous stimulation and EEG difficult. Interleaving stimulation with artifact-minimized recording may be preferable, provided the design includes a plan to evaluate phase estimation errors. High-gamma scalp power is especially vulnerable to facial or muscle activity. Its speed alone gives no basis for calling it a quantum-Zeno signature.
The strongest interpretable result would be reproducible phase-dependent changes in stopping behavior that survive sham, blinding, field-localization, and artifact controls. This would show that externally applied fields can influence neural control. It would not alone identify endogenous ephaptic feedback as an autonomous computer or establish that electromagnetic fields are conscious.
Workstream 3: Historical agency and habit formation
The lasting effects of earlier decisions may be the most experimentally practical and most neglected part of the original idea. An act of inhibition with small immediate consequences may still alter learning, confidence, or habitual response selection. Studying these effects requires no metaphysical assumption.
The longitudinal component has separate phases for formation and expression. First, participants repeatedly encounter a choice or stop opportunity. We randomize the availability or framing of self-directed inhibition, externally instructed inhibition, and matched reward or exposure conditions. In the second phase, potentially days or weeks later, participants encounter related situations without the original training prompt. We then test whether the prior condition changes later stopping behavior and learned action policy.
A basic learning model is:
$$ W_{t+1}=W_t+\alpha(R_t-\hat R_t)e_t, $$where $W_t$ is the learned policy, $R_t-\hat R_t$ is a prediction error, and $e_t$ is a learning trace. We will compare this model family with habit-strength and context-dependent alternatives, allowing for persistent effects that a single reinforcement rule might not explain.
Outcomes include the probability of stopping without deliberative prompts, changes in response speed, resistance to changed incentives, and persistence after task instructions are removed. Individual self-veto choices are not randomized, which limits causal claims about a particular chosen veto. Randomized opportunities or instructions provide a stronger basis for causal inference than observational comparisons between participants who choose to veto and those who do not.
A positive result would support the hypothesis that an earlier episode of control helps shape later automaticity, without establishing that the episode was metaphysically free. A null result would limit the generality of this historical mechanism under the tested conditions. It would not disprove all long-term consequences of choice.
Workstream 4: Quantum intervention and boundary conditions
The preceding studies fit within neuroscience. COAT also retains two unestablished hypotheses: subject-directed changes in physical probabilities and existential selection of a life or observer-center.
Hypothetical quantum probability biasing
In ordinary quantum mechanics, a density operator $\rho$ and a measurement projector $\Pi_k$ yield Born-rule probabilities $p_0(k)=\operatorname{Tr}(\Pi_k\rho)$. COAT could parameterize a hypothetical preference-weighted outcome distribution as follows:
$$ p_\lambda(k\mid\rho,u)= \frac{p_0(k\mid\rho)e^{\lambda v(k,u)}} {\sum_j p_0(j\mid\rho)e^{\lambda v(j,u)}}. $$Setting $\lambda=0$ recovers the ordinary distribution. Nonzero $\lambda$ changes relative outcome frequencies within the support of the baseline distribution. This normalized mathematical family describes where a hypothetical coupling might act; it is not a derived quantum law or an observed control channel. The theory would still need to specify the physical system carrying $u$, when it can operate, and why the Born rule is modified. It must also explain how entanglement, complete positivity, conservation laws, and no-superluminal-signalling are preserved.27
Henry Stapp proposed a different approach to mental causation based on quantum measurement and the quantum Zeno effect.28 Conway and Kochen’s Free Will Theorem describes a conditional relationship between assumptions about experimenter choice and the determination of particle outcomes. It does not demonstrate that human subjects possess metaphysical authorship.29 Scott Aaronson’s analysis of quantum freebit-style ideas also explores conceptual possibilities without proving that people control quantum outcomes.30
Neural quantum proposals must also explain how microscopic variations affect macroscopic behavior. Tegmark’s decoherence estimates cast doubt on certain brain-quantum mechanisms at cognitive timescales. Hagan, Hameroff, and Tuszyński disputed details relevant to proposed microtubule models.3132 Any microscopic influence would need a neural mechanism that transmits it into action-level differences. A fixed stochastic channel cannot increase total-variation distinguishability. Nonlinear dynamics may make an effect behaviorally consequential, but cannot make a physically irrelevant perturbation evidence of control.
No human EEG result will be counted as a deviation from the Born rule. A quantum-intervention study would be warranted only after the theory specifies a physically coherent coupling, a measurable signature that ordinary contextual dynamics cannot explain, and a design excluding experimenter-choice and selection artifacts. Until those requirements are met, this branch is limited to theoretical consistency analysis and offers no basis for detecting consciousness.
Existential selection
Existential selection can be modeled as a relation that selects an experienced centered history:
$$ \Sigma(S,U_E)\longrightarrow(h,i,t_0). $$One model allows a hypothetical subject to choose among already admissible experiential centers. Another claims that selection determines physical boundary conditions or brings a history into existence. These models require different explanations, and both differ from intervention within a life. A time-symmetric or block-universe formulation alone does not imply a chooser outside time. Philosophical precedents for pre-life selection also provide no empirical evidence for it.6
A model can formally represent the absence of memory of such a selection during embodied life. For example, it could set the mutual information between accessible autobiographical memory and an earlier selection variable to zero. Failure to remember a hypothetical pre-life choice provides no evidence that it occurred. Infancy, the onset of self-reflection, memory formation, and personal identity require separate representation.
This branch examines the often implicit assumption that freedom must continually redirect a life from within it. If distinct observer assignments yield exactly the same physical observations, purely third-person experiments cannot identify the true assignment. The research program should acknowledge this limit instead of proposing a test that cannot distinguish the models.
Workstream 5: Beliefs about agency and the self-model
The human brain produces behavior and models itself doing so. Beliefs about control may affect motivation, effort allocation, explanations of success and failure, and willingness to try again. These effects warrant physical and psychological study regardless of whether metaphysical freedom exists.
Let $B_t$ encode the embodied system’s beliefs about its own agency. These beliefs can influence choices through the ordinary physical transition law:
$$ B_{t+1}=f_B(B_t,X_t,R_t),\qquad P(A_t\mid X_t,B_t). $$A deterministic brain’s belief about determinism can affect its behavior because the belief is part of the causal system. Whether disbelief in free will reliably undermines self-control is an empirical question. Vohs and Schooler reported such consequences under certain manipulations.33 A later meta-analysis of 145 experiments found reliable shifts in free-will beliefs, but no reliable downstream behavioral consequences overall.34
The reflexive loop can be tested without treating it as a moral warning. A small, preregistered study could compare neutral explanatory materials with balanced accounts of mechanistic and agential perspectives. It would measure effort, persistence, and stopping outcomes, with checks for comprehension and demand effects. Materials must avoid convincing participants that they lack responsibility or capacity. A null result would be reported as such, without interpreting it as a hidden collapse of agency.
Research stages and decision criteria
The workstreams have different evidentiary standards despite their shared questions. COAT organizes physical models and conditional metaphysical interpretations within one research program. Its initial outputs should remain useful even if no claim of extra-physical authorship survives.
| Stage | Planned output | Decision gate |
|---|---|---|
| 0: Formal specification | Public ontology ledger, source-authorship definitions, model catalog, simulated counterexamples | Are the assumptions distinguishable, internally consistent, and free of hidden circularity? |
| 1: Veto pilot | Timing validation, task feasibility, EEG/EMG quality, calibrated prospective decoder | Can valid candidate veto trials be identified without outcome leakage or subjective guesswork? |
| 2: Confirmatory veto study | Preregistered $M_0$ versus $M_1$ predictions across sessions and task conditions | Does any sparse-control model outperform credible physical competitors on held-out outcomes? |
| 3: Field-control arm | Sham-controlled phase-specific stimulation with behavioral endpoints as the primary measures | Is there a repeatable stimulation effect after artifact, sensation, and preparation controls? |
| 4: Learning and self-model studies | Tests of post-training habit persistence and belief-dependent control | Do the causal effects persist, and which parts can be attributed to randomized conditions? |
| 5: Synthesis and replication | Independent replication, released models and preregistered null results, revised COAT ledger | Which branches gain support, lose support, or remain observationally undecidable? |
We provisionally estimate roughly two years for formalization, piloting, principal experiments, and initial replication. The schedule depends on ethics review, collaborations, recruitment, and equipment. The sequence is proposed; no study is claimed to have been funded, approved, or completed.
Research integrity and safeguards
Before confirmatory data collection, the registered protocol should specify primary model comparisons, exclusion criteria, stopping rules, effect thresholds, and a statistical plan for nested trials and sessions. Pilot data may inform these choices but must remain separate from confirmatory evaluation. Features selected after inspecting outcomes must not be presented as preregistered hypotheses.
Studies of consciousness and responsibility require particular care in protecting participants. Consent materials must not claim that EEG can reveal whether a person has free will. Any tACS arm requires appropriate institutional ethics approval, screening, safe stimulation parameters, adverse-event procedures, and qualified oversight. Public discussion must not treat motor-inhibition performance as evidence of moral worth, competence, criminal culpability, or the presence of a soul.
We should share analysis code, simulated datasets, model definitions, deidentified measurements where consent permits, and enough timing metadata to reproduce decoder predictions. Potentially identifying neural and behavioral records require a privacy plan and controlled access where appropriate. The planned record includes null outcomes and unsuccessful replications; these are valid research outputs.
Interpreting results
Every major result should be accompanied by a record of the conclusions it supports and the questions it leaves unresolved. This would help prevent results from being used to answer questions the experiments were never designed to test:
| Possible result | What would be supported | What would remain unproven |
|---|---|---|
| Late-stage self-veto after measurable motor preparation | Preparation is not identical to irreversible execution | An immaterial author initiated the stop |
| Distinct frontal or sensorimotor transition before self-veto | Internally generated inhibition has detectable neural dynamics | That the dynamics lie outside physical causation |
| $M_1$ consistently predicts better than the strongest physical baselines | A sparse-control description adds explanatory/predictive value | That $J_t$ is ontologically free rather than an unmodeled physical process |
| Phase-specific stimulation shifts stopping behavior | Electrical perturbation can influence inhibitory control | Endogenous EM fields are conscious or independently author actions |
| Earlier stop opportunities alter later automatic responses | Learning carries control-related history into future policy | Earlier choices were metaphysically authored |
| Belief manipulation changes effort or stopping | Self-models can causally affect behavior in the tested context | Universal passivity from believing in determinism |
| No advantage for sparse control; no stimulation effect | The tested added mechanisms lack support under the chosen models and precision | No form of free will can exist in any ontology |
Stronger claims of metaphysical authorship require additional predictions. Greater confidence in describing the same neural traces supplies no new evidence. If physical and nonphysical theories predict identical observable distributions, repeated measurement of those variables cannot distinguish them. This identifiability limit still permits research on physical control.
Failure criteria
The proposal must be narrowed when evidence or unresolved theoretical problems warrant it. Four failure criteria follow.
Predictive failure: If the additional intermittent-control process does not meaningfully improve held-out prediction after stronger physical baselines are included, COAT-VETO-1 should report a negative result and downgrade the operational claim. The model must not be made increasingly flexible simply to recover a preferred latent event.
Intervention failure: If a field-related observational signature is present but controlled stimulation produces no reliable behavioral change, claims about phase-sensitive behavioral control must be limited accordingly. An oscillatory correlate alone does not establish a mechanism.
Mechanistic failure: If a hypothetical quantum control channel cannot be reconciled with physical constraints or assigned a discriminating measurement, it must remain metaphysical speculation. Unrelated EEG gamma oscillations and the existence of quantum indeterminacy provide no support for that channel.
Conceptual failure: Defining authorship by assuming an event was authored leaves freedom unexplained. The theory must acknowledge the unresolved regress: it still needs to explain what makes an intervention the subject’s own choice, beyond determination by prior conditions or randomness.
Negative results in the first two categories would improve the scientific account of voluntary control. They would also help prevent unexplained residuals from being treated as discoveries of free will.
Purpose of the program
Debates about free will often reduce agency to general claims about matter and causation, overlooking deliberation, responsibility, and experience once nervous-system mechanisms are explained. Others infer an unmeasured controller outside physics from subjective experience. Both approaches leave substantial parts of the problem unexamined.
The proposed program asks whether neural preparation remains reversible, which physical mechanisms coordinate a veto, how earlier control reshapes later behavior, and why a self-model changes the system’s behavior. It also states the metaphysical alternatives, including pre-embodiment selection and probability-level intervention, while acknowledging that none has been established in the laboratory.
Conditional Ontological Agency Theory aims to separate questions about causal efficacy, origins, identity, and lived control. It offers no final answer to free will. Evidence favoring ordinary inhibition and physical field dynamics would contribute to our understanding of agency. Stronger models could be tested if they yield distinctive, replicable predictions. Where branches remain observationally equivalent, the framework should acknowledge that limit.
A satisfying definition of free will alone would not answer these questions. The research program seeks a causal account of a life that distinguishes current control, inherited consequences, experience alone, and authentic authorship, if any.
GTCode.com Research
References
Kane, R. (1996). The Significance of Free Will. Oxford University Press. Kane’s self-forming actions are the principal precedent for temporally concentrated and historically persistent authorship. ↩︎ ↩︎
O’Connor, T. (2011). Agent-Causal Theories of Freedom. A survey of agent-causal explanations and their central challenges. ↩︎
List, C. (2014). Free Will, Determinism, and the Possibility of Doing Otherwise. Noûs. A sophisticated higher-level account; the ontology of alternatives remains debated. ↩︎
Balaguer, M. (2010). Free Will as an Open Scientific Problem. MIT Press. ↩︎
Belnap, N., Perloff, M., & Xu, M. (2001). Facing the Future: Agents and Choices in Our Indeterminist World. Oxford University Press. Branching-time and STIT-style formal treatment of agency. ↩︎
Coope, U. (2020). Freedom and Responsibility in Neoplatonist Thought, especially the chapter Responsibility and the Myth of Er. Oxford University Press. Historical-philosophical precedent, not empirical evidence for pre-life selection. ↩︎ ↩︎
Self-Locating Beliefs. Stanford Encyclopedia of Philosophy. Introduction to centered-world and indexical frameworks. ↩︎ ↩︎
Tse, P. U. (2013). The Neural Basis of Free Will: Criterial Causation. MIT Press. ↩︎
Hoel, E. P., Albantakis, L., & Tononi, G. (2013). Quantifying causal emergence shows that macro can beat micro. Proceedings of the National Academy of Sciences, 110(49), 19790–19795. ↩︎
Salge, C., Glackin, C., & Polani, D. (2013). Empowerment—An Introduction. arXiv:1310.1863. ↩︎
Pearl, J. (2009). Causality: Models, Reasoning, and Inference, 2nd ed. Cambridge University Press. ↩︎
Libet, B., Gleason, C. A., Wright, E. W., & Pearl, D. K. (1983). Time of conscious intention to act in relation to onset of cerebral activity (readiness-potential). Brain, 106(3), 623–642. ↩︎
Schurger, A., Sitt, J. D., & Dehaene, S. (2012). An accumulator model for spontaneous neural activity prior to self-initiated movement. PNAS, 109(42), E2904–E2913. ↩︎
Schultze-Kraft, M., et al. (2016). The point of no return in vetoing self-initiated movements. PNAS, 113(4), 1080–1085. This study used a prospective EEG decoder and externally delivered stopping cues. ↩︎
Brass, M., & Haggard, P. (2007). To Do or Not to Do: The Neural Signature of Self-Control. Journal of Neuroscience. ↩︎
Maoz, U., Yaffe, G., Koch, C., & Mudrik, L. (2019). Neural precursors of decisions that matter—an ERP study of deliberate and arbitrary choice. eLife, 8, e39787. ↩︎
Shum, Y. H., Galang, C. M., & Brass, M. (2024). Using a Veto paradigm to investigate the decision models in explaining Libet-style experiments. Consciousness and Cognition, 124, 103732. ↩︎
Triggiani, A. I., Kreiman, G., Lewis, C., Maoz, U., Mele, A., Mudrik, L., Roskies, A. L., Schurger, A., & Hallett, M. (2023). What is the intention to move and when does it occur?. Neuroscience & Biobehavioral Reviews, 151, 105199. ↩︎
Anastassiou, C. A., Perin, R., Markram, H., & Koch, C. (2011). Ephaptic coupling of cortical neurons. Nature Neuroscience, 14, 217–223. ↩︎
Pinotsis, D. A., & Miller, E. K. (2022). Beyond dimension reduction: Stable electric fields emerge from and allow representational drift. NeuroImage, 253, 119058. ↩︎
Pinotsis, D. A., Fridman, G., & Miller, E. K. (2023). Cytoelectric coupling: Electric fields sculpt neural activity and “tune” the brain’s infrastructure. Progress in Neurobiology, 226, 102465. ↩︎
Pinotsis, D. A., & Miller, E. K. (2026). Ephaptic coupling can explain variability in neural activity. Cerebral Cortex, 36(6), bhag098. The study used prefrontal local-field-potential data and model-based inference rather than directly isolating a conscious field controller. ↩︎
McFadden, J. (2020). Integrating information in the brain’s EM field: The CEMI field theory of consciousness. Neuroscience of Consciousness. See also McFadden, J. (2025), Computing with electromagnetic fields rather than binary digits: A route towards artificial general intelligence and conscious AI, Frontiers in Systems Neuroscience, 19. These are theoretical proposals, not established demonstrations that EM fields are conscious. ↩︎
Fang, Z., Sack, A. T., & Leunissen, I. (2024). The phase of tACS-entrained pre-SMA beta oscillations modulates motor inhibition. NeuroImage, 290, 120572. ↩︎
Fang, Z., Sack, A. T., & Leunissen, I. (2026 preprint, manuscript initially posted in 2025). Closed-loop phase-locked EEG-tACS enables adaptive control of cortical beta synchrony and motor control. bioRxiv. Findings should not be treated as independently replicated peer-reviewed results. ↩︎
Zhu, T., Sack, A. T., & Leunissen, I. (2026). Phase-Specific Dual-Site Beta Transcranial Alternating Current Stimulation Differentially Influences Functional Connectivity Associated With Motor Inhibition Performance. Human Brain Mapping, 47(3), e70470. ↩︎
Simon, C., Bužek, V., & Gisin, N. (2001). No-Signaling Condition and Quantum Dynamics. Physical Review Letters, 87, 170405. ↩︎
Stapp, H. P. (2008). A model of the quantum-classical and mind-brain connections, and of the role of the quantum Zeno effect in the physical implementation of conscious intent. arXiv:0803.1633. ↩︎
Conway, J. H., & Kochen, S. (2009). The Strong Free Will Theorem. Notices of the American Mathematical Society, 56(2), 226–232. ↩︎
Aaronson, S. (2013). The Ghost in the Quantum Turing Machine. arXiv:1306.0159. ↩︎
Tegmark, M. (2000). Importance of quantum decoherence in brain processes. Physical Review E, 61, 4194–4206. ↩︎
Hagan, S., Hameroff, S. R., & Tuszyński, J. A. (2002). Quantum computation in brain microtubules: Decoherence and biological feasibility. Physical Review E, 65, 061901. ↩︎
Vohs, K. D., & Schooler, J. W. (2008). The value of believing in free will: Encouraging a belief in determinism increases cheating. Personality and Social Psychology Bulletin. ↩︎
Genschow, O., Cracco, E., Schneider, J., Protzko, J., Wisniewski, D., Brass, M., & Schooler, J. W. (2023; first published online 2022). Manipulating Belief in Free Will and Its Downstream Consequences: A Meta-Analysis. Personality and Social Psychology Review, 27(1). An analysis of 145 experiments reporting no reliable aggregate downstream behavioral consequences. ↩︎