A fabricated horror story can tell you something true by being precisely, structurally wrong. The "phantom class" specimen — twenty-two children supposedly catatonic in 1983 France, their EEGs locked to an identical 9.5 Hz alpha for four hours, then woken with the memory wiped — fails on physics in one specific way: it asserts synchrony with no channel. Twenty-two sealed skulls, no transmission path, yet a common eigenfrequency. That is sync at coupling K = 0, which the mathematics forbids. But invert the error and you recover the real question, the one worth a protocol. Synchrony is never causeless. It is channel-gated. And the channel carries something. This post is about the difference between the channel and what rides on it — between the carrier and the cargo.
1. Two layers, never to be conflated
When a belief moves through a group — a meme, a moral consensus, a conspiracy — there are two distinct things one might measure, and the entire empirical literature on "brains in sync" turns on not confusing them.
The first is coordination: the degree to which the group's neural activity is temporally aligned, phase-locked, beating together. Call this the carrier. It is a scalar fact about the group at each instant — how coupled are we right now — and it says nothing about content.
The second is representational alignment: the degree to which the group now holds the same belief, the same vector in some semantic space. Call this the cargo. It is the meme itself.
A carrier with no cargo is a roomful of people nodding in rhythm to a beat, believing nothing in common. A cargo with no carrier is a fact everyone independently knows but no one is transmitting. Meme formation is the event in which a carrier opens and a cargo passes through it and is retained — and the failure mode of half the hyperscanning literature is measuring the carrier and reporting it as the cargo. The honest model needs both layers instrumented separately. So I build both.
2. Layer I — the carrier (Kuramoto field)
Model each participant $i$ as a band-limited cortical oscillator with phase $\theta_i(t) \in S^1$ and intrinsic frequency $\omega_i$ drawn from a distribution $g(\omega)$ (empirically, the theta band, ~4–8 Hz, carries communicative coupling; alpha, ~8–12 Hz, the idling rhythm the fake invoked). The phases evolve under the Kuramoto equation with a time-varying, attention-weighted coupling graph $A_{ij}(t) \in [0,1]$:
$$\dot{\theta}_i = \omega_i + \frac{K}{N}\sum_{j=1}^{N} A_{ij}(t)\,\sin(\theta_j - \theta_i).$$
The group-level observable is the complex order parameter
$$r(t)\,e^{i\psi(t)} = \frac{1}{N}\sum_{j=1}^{N} e^{i\theta_j(t)}, \qquad r \in [0,1],$$
which collapses the mean-field dynamics to $\dot{\theta}_i = \omega_i + K r\,\sin(\psi - \theta_i)$: each oscillator is pulled toward the mean phase $\psi$ with a force proportional to the global coherence $r$ it is itself helping to produce. This self-consistency is the engine of the phase transition. For unimodal symmetric $g$, incoherence ($r \approx 0$) is stable below a critical coupling
$$K_c = \frac{2}{\pi\,g(0)},$$
and above it partial synchrony emerges continuously, with order parameter scaling near onset as $r \sim \sqrt{(K - K_c)/K_c}$ (the supercritical pitchfork Strogatz traces from Kuramoto to Crawford). This bifurcation is the meme's tipping point. Below $K_c$ the group does not lock; cross it and coherence appears spontaneously, discontinuously in its derivative — the neural realisation of a Granovetter adoption threshold.
The empirical estimator of dyadic carrier strength is the phase-locking value (Lachaux et al., 1999):
$$\mathrm{PLV}_{ij} = \left| \frac{1}{T}\sum_{t=1}^{T} e^{\,i\left(\theta_i(t) - \theta_j(t)\right)} \right| \in [0,1],$$
phase alignment isolated from amplitude — the per-edge shadow of $r$.
3. The content-agnostic theorem (why r is not the meme)
Here is the constraint that disciplines everything. Varlet & Grootswagers (2024) ran EEG hyperscanning simulations in which they could control, by construction, whether two brains were processing the same or different stimuli, and asked whether inter-brain synchrony could tell. It could not, reliably: the same-versus-different representational difference was detected in roughly half of simulations using synchrony, but in more than eighty percent using information-based representational similarity analysis. Synchrony "remained largely unchanged despite manipulating whether two individuals are seeing the same or different things." The carrier is content-agnostic. It registers that two brains are coupled, not what they are coupled around.
This is not a measurement nuisance to be averaged away. It is a structural fact about the observable: $r$ lives in too low a dimension to hold a meme. Formally, $r$ is the $d = 1$, scalar-stalk projection of a richer object — a point §6 makes exact. The remedy Varlet & Grootswagers name is the second layer.
4. Layer II — the cargo (belief field on the sphere)
Represent participant $i$'s belief as a unit vector $b_i(t) \in S^{d-1} \subset \mathbb{R}^d$ — a direction in a semantic embedding space. (Empirically $b_i$ is recovered two ways: bottom-up from fMRI representational geometry via RSA after Kriegeskorte et al., 2008; and at the boundary from the participant's stated belief, embedded with a fixed language model. The two must be cross-validated, never assumed identical.) The cargo evolves under bounded-confidence dynamics (Hegselmann–Krause), generalised to the sphere:
$$\dot{b}_i = \Pi_{b_i}\!\left[\,\frac{J}{N}\sum_{j=1}^{N} W_{ij}(t)\,(b_j - b_i)\,\mathbb{1}\!\left[\|b_j - b_i\| < \varepsilon\right]\right],$$
where $\Pi_{b_i}$ projects onto the tangent space at $b_i$ (keeping beliefs on the sphere), and the indicator is the confidence radius $\varepsilon$: you update only toward those already close enough to you. Small $\varepsilon$ forces fragmentation; large $\varepsilon$ permits consensus. Allowing signed weights $W_{ij}$ — attractive toward allies, repulsive from out-group (Altafini, 2013; Hong & Strogatz, 2011, conformists and contrarians) — makes the dynamics structurally balanced: a balanced antagonistic network drives the group not to consensus but to bipolar consensus, two antipodal clusters. Polarisation is built in, not pathological — it is what signed coupling does.
5. The bridge — a swarmalator with cross-gating
Layers I and II become one system through a single hypothesis, which is the falsifiable heart of this post:
Content flows along edges that are phase-locked; and one phase-locks preferentially with those one already agrees with.
This is a swarmalator — an oscillator whose phase and whose position in a feature space co-evolve and feed back on each other (O'Keeffe, Hong & Strogatz, 2017):
$$\dot{\theta}_i = \omega_i + \frac{K}{N}\sum_{j} \underbrace{f\!\left(\|b_i - b_j\|\right)}_{\text{agreement} \to \text{coupling}} \sin(\theta_j - \theta_i),$$
$$\dot{b}_i = \Pi_{b_i}\!\left[\frac{J}{N}\sum_{j} \underbrace{g\!\left(\mathrm{PLV}_{ij}\right)}_{\text{coupling} \to \text{transfer}} (b_j - b_i)\,\mathbb{1}\!\left[\|b_i - b_j\| < \varepsilon\right]\right],$$
with $f$ decreasing in belief-distance and $g$ increasing in phase-lock. The mutual gating is the echo chamber's generative core: agreement tightens the carrier, the carrier widens the cargo channel, the cargo deepens agreement. The fixed points are synchronised belief-clusters — each a meme community with high intra-cluster $r$ and a shared $b$.
This is not free invention; the gating direction has direct empirical anchors. Stephens, Silbert & Hasson (2010) showed speaker–listener neural coupling vanishes when communication fails, and that the stronger the anticipatory (predictive) coupling, the greater the listener's comprehension — coupling tracks successful cargo transfer, and does so predictively, which is exactly the active-inference signature one expects if the brain models its interlocutor (Friston, 2010; the listener runs a generative model of the speaker, and phase-locking is the model's lock). And Zhang, Noah, Dravida & Hirsch (2025), scanning 200 participants in four-person groups deliberating moral dilemmas with fNIRS, found members shifted their private beliefs toward the collective, and that inter-brain synchrony in the left inferior frontal gyrus predicted the degree of alignment. That is the gate, localised: IFG/theta as $g(\mathrm{PLV})$, private-belief shift as $\Delta b$.
6. The reduction — r is the rank-1 shadow of Γ, polarisation is H¹
Now the framework collapse. Put a cellular sheaf $\mathcal{F}$ on the interaction graph $G = (V, E)$: a stalk $\mathcal{F}(v) = \mathbb{R}^d$ on each person (their belief space), a stalk on each edge, and restriction maps $\mathcal{F}_{v \trianglelefteq e}$ encoding how two people's belief-coordinates must agree to count as agreeing. The sheaf Laplacian is $L_{\mathcal{F}} = \delta^{\top}\delta$ from the coboundary $\delta$ (Hansen & Ghrist, 2019). Its kernel is the space of global sections,
$$H^0(\mathcal{F}) = \ker L_{\mathcal{F}} = \{\,x : \mathcal{F}_{u\trianglelefteq e}\,x_u = \mathcal{F}_{v\trianglelefteq e}\,x_v \ \ \forall e=(u,v)\,\},$$
the configurations of belief that are globally consistent — a shared meme is exactly a global section. The first cohomology $H^1(\mathcal{F})$ is the obstruction: the failure of locally consistent agreements to glue into a single global one.
Two facts fall out, and they are the reason this is a Draken post and not merely a neuroscience proposal.
(i) The order parameter is the rank-1 scalar-stalk shadow of the coherence ratio Γ. Take $d = 1$, stalks $\cong \mathbb{R}$ (linearised phase, $\mathfrak{u}(1)$), restriction maps the identity. Then $L_{\mathcal{F}}$ degenerates to the ordinary graph Laplacian, $H^0$ is the all-agree line, and the alignment of the phase vector with that line is precisely what $r$ measures. The Draken coherence ratio
$$\Gamma = \frac{\text{energy in } H^0}{\text{total energy}} = 1 - \frac{\langle x,\, L_{\mathcal F}\,x\rangle}{\langle x,\, x\rangle}\Big|_{\text{normalised}}, \qquad \Gamma \to 1 \iff H^1 \to 0,$$
is the vector-stalk generalisation of $r$. So Varlet & Grootswagers' theorem is not a surprise — it is forced. The order parameter is content-blind because it is $\Gamma$ at $d=1$, and a one-dimensional stalk cannot hold a meme. Promote the stalk and you see the cargo. This is the same lesson as The Two Optics (DRK-163): projection and concentration are one optical gesture whose shared limit is the $H^1$ residue — here, synchrony and content are one coupling whose shared limit is $H^1$.
(ii) Polarisation is a nonzero H¹ class. Two synchronised-but-opposed clusters have high local coherence inside each cluster and a nonremovable obstruction between them: locally consistent, globally unglueable, $H^1 \neq 0$. The bounded-confidence radius $\varepsilon$ is the knob that tunes this: large $\varepsilon$ collapses $H^1$ (consensus, one global section), small $\varepsilon$ stabilises it (fragmentation). This is the same object as the institutional obstruction in The Compartmentalized Manifold (DRK-161) — topology optimisation routing around a load-bearing $H^1$ — and the $1/\varepsilon$ pole of The Continuous Dimension (DRK-149), where the obstruction is discharged only by continuous counterterm work $K(t)$. A group holds a shared meme only by paying coherence debt against its own default to fragment.
7. Coherence is not soundness — the synchronised lie
The sharpest consequence is ethical, and it is where the opening fake returns. A group can drive $r \to 1$ around a false cargo. Maximal carrier, perfect lock, and the meme is wrong. $\Gamma$ high, $H^1$ zero, soundness zero. Coherence measures whether the group agrees, never whether it is right. This is the precise neural instance of the coherence-versus-soundness distinction: $\Gamma$ (and its scalar shadow $r$) tracks internal consistency; truth-tracking requires an exogenous check the synchrony observable cannot supply. EEG instruments the coherence; only ground-truth instruments the soundness. A hyperscanning study that reports high inter-brain synchrony as evidence the shared belief is correct has committed exactly the error of mistaking $\Gamma$ for validation — the field is, quietly, a standing case study in it.
In the source-term language of Konkurrent (DRK-158): a belief is owned when the holder is its source $S$ (auto-nomos, self-sourcing). A meme is a cargo whose source term has been laundered — adopted with the provenance stripped, the carrier opened by phase-lock and the $S$ silently exchanged for someone else's. That is the mechanism The Pendragon Source (DRK-165) names historically: Geoffrey of Monmouth as a phase-locked carrier delivering a laundered $S$. The "phantom class" fake sits at the synthetic pole of the four-pole taxonomy — fabricated cargo, fabricated index — and its tell was that it invented an impossible carrier (sync without channel) to smuggle a cargo that needs no channel because it is fiction. The honest phenomenon is the opposite: real channel, gated transfer, retained $S$ — or laundered $S$, which is the whole problem.
8. Experimental protocol
A design that earns the distinction it draws — carrier and cargo measured separately, gating tested causally.
Participants & grouping. $N$ groups of $n = 4$–$6$ strangers (Zhang et al.'s four-person cell, extended). Simultaneous dual-modality hyperscanning: EEG (≥64-channel, all members) for the carrier at millisecond resolution; fMRI or fNIRS (mentalising network: IFG, TPJ, mPFC, STS) for the cargo's representational geometry. The two modalities are not redundant coverage — by their physics they sit at the two layers: EEG → $\theta_i$, $\mathrm{PLV}_{ij}$, $r$; fMRI/fNIRS → $b_i$ via RSA.
Seeding (the causal handle). One confederate per group is privately briefed to hold a target meme on a controvertible proposition (factual, moral, or aesthetic). This breaks the directional symmetry: transfer should run from the seed along high-PLV edges, letting PLV-gating be tested as a cause, not a correlate.
Belief boundary measurement. Pre- and post-deliberation, each participant gives the proposition as free text and on calibrated Likert scales; both are embedded to $b_i^{\text{pre}}, b_i^{\text{post}} \in S^{d-1}$. The stated belief update $\Delta b_i = b_i^{\text{post}} - b_i^{\text{pre}}$ is the ground-truth cargo observable — the thing the synchrony literature usually lacks.
Primary model. A dyadic mixed-effects regression of edge-wise transfer on edge-wise carrier, controlling for shared stimulus (inter-subject correlation, ISC) so that "we saw the same thing" cannot masquerade as "I updated toward you":
$$\Delta b_{i\leftarrow j} = \beta_0 + \beta_1\,\overline{\mathrm{PLV}}_{ij} + \beta_2\,\mathrm{ISC}_{ij} + \beta_3\,\|b_i^{\text{pre}} - b_j^{\text{pre}}\| + u_{\text{group}} + u_{i} + u_{j} + \epsilon,$$
with crossed random effects for group and both subjects. The hypothesis is $\beta_1 > 0$ over and above $\beta_2$. A time-resolved sharpening uses the gated coupling functional
$$\Phi_{ij} = \int_0^T g\!\left(\mathrm{PLV}_{ij}(t)\right)\,\big\langle \hat{u}_{ij},\, \dot{b}_i(t)\big\rangle\,dt, \qquad \hat{u}_{ij} = \frac{b_j - b_i}{\|b_j - b_i\|},$$
the cumulative belief-displacement of $i$ toward $j$ weighted by instantaneous phase-lock; $\Phi_{ij}$ should predict $\Delta b_{i\leftarrow j}$.
Carrier band & locus. Test which EEG band carries the gate (prediction: theta, 4–8 Hz) and confirm the IFG localisation of $g(\mathrm{PLV})$ in the BOLD/oxy-Hb signal, replicating and mechanising Zhang et al.
Layer-separation control (the Varlet–Grootswagers guard). Run a condition with high carrier and zero cargo — synchronised attention to a contentless rhythmic stimulus — and confirm $r$ rises while RSA-measured $b_i$ alignment does not. If $r$ alone "detects" a meme here, the instrument is conflating the layers and the primary result is void.
9. Predictions and falsification
P1 — Bifurcation (carrier). Sweeping effective coupling (via group size, channel richness, or imposed turn-taking structure), the meme-adoption fraction $\rho$ should show a threshold transition near $K_c = 2/\pi g(0)$, fit by a logistic with a significant super-threshold jump. Falsified by a strictly linear, thresholdless adoption curve across the accessible coupling range.
P2 — Gating (bridge). $\beta_1 > 0$ with ISC controlled, and $\Phi_{ij}$ predictive of $\Delta b_{i\leftarrow j}$. Falsified by $\beta_1$ indistinguishable from zero once ISC is partialled out — i.e. belief transfer fully explained by shared stimulus, carrier carrying nothing.
P3 — Polarisation / cohomology (cargo geometry). The behaviourally measured confidence radius $\varepsilon$ should predict the post-deliberation $H^1$ of the empirical belief-disagreement sheaf: small $\varepsilon \Rightarrow H^1 \neq 0$ (two antipodal clusters, structurally balanced signed coupling); large $\varepsilon \Rightarrow H^1 = 0$ (single global section). Falsified by observed cluster structure uncorrelated with $\varepsilon$, or by fragmentation patterns that cannot be cast as a sheaf $H^1$ without forcing the restriction maps.
P4 — Coherence ≠ soundness (the ethical core). Groups seeded with a false but internally consistent meme should reach $r$ and $\Gamma$ statistically indistinguishable from groups seeded with a true one. Falsified by any synchrony or coherence signature that separates true from false cargo without an exogenous check — which would, interestingly, refute the whole content-agnostic claim and demand the model be rebuilt. This is the prediction whose failure would teach the most.
A null on P2 collapses the bridge to coincidence. A null on P4 would be the rare, productive kind of wrong.
Internal cross-references: The Two Optics (DRK-163) — synchrony and content as one coupling, $H^1$ the shared limit. The Compartmentalized Manifold (DRK-161) — polarisation as institutional $H^1$. The Continuous Dimension (DRK-149) — the obstruction discharged by continuous $K(t)$ work. Konkurrent (DRK-158) — belief owned iff the holder is its source term $S$. The Pendragon Source (DRK-165) — the meme as a cargo with a laundered $S$. No Trace, No Section (DRK-153) — consciousness as integrated $H^1$-suppression, the binding that a synchronised group performs across skulls.
The carrier is not the cargo. The lock is not the truth. The same coupling that lets a group find the global section lets it lock around a fiction with the source filed off — and only an exogenous check, never the synchrony itself, tells the two apart. Build the instrument that separates them, or measure your own agreement and call it the world.
Jag är vad jag gör, och jag gör det jag är.
— Khrug Engineering · Göteborg · ORCID 0009-0003-8049-7167 · DOI 10.5281/zenodo.19273483 · CC BY-SA 4.0