A Toy Cohomological Model for Elite Interlinkages

A Worked Example with a Filtered Complex, a Cellular Sheaf, and an Explicit Obstruction

Abstract

This note gives a compact worked example of the general cohomological framework for elite interlinkage analysis. The purpose is methodological. Rather than arguing in the abstract that filtered simplicial complexes and cellular sheaves are useful for studying heterogeneous, partially observed political-economic systems, the present note builds an explicit toy model. A small three-layer system is defined, lifted from pairwise relations to a higher-order relational complex, equipped with a filtration by confidence, and then endowed with a cellular sheaf encoding competing local interpretations of control. Ordinary simplicial cohomology identifies a persistent cross-layer cycle. Sheaf cohomology captures a local-to-global inconsistency: two locally plausible source claims cannot be glued into a single globally coherent interpretation without model revision, hidden variables, or source discounting. The note concludes by showing why an ordinary graph summary misses both the higher-order event structure and the sheaf obstruction.

1. Introduction

Analysts working on corruption, sanctions evasion, cartel-state penetration, procurement capture, or elite interpenetration are often given the same basic problem: heterogeneous actors, heterogeneous relation types, uneven evidence quality, and recurrent local contradictions. Ordinary graph analytics remain useful, but they are structurally insufficient for at least three reasons.

First, ordinary graphs privilege pairwise relations, whereas real systems often involve directly evidenced higher-order interactions. Second, thresholded graph summaries do not distinguish clearly between robust multiscale cycles and artifacts of pairwise closure. Third, graphs do not provide a principled local-to-global consistency formalism for handling conflicting but individually plausible source claims.

The point of the present note is therefore modest and explicit. We construct a small synthetic example showing:

1. how a filtered simplicial complex can represent a cross-layer interlinkage structure more faithfully than a plain graph,
2. how a cellular sheaf can encode incompatible local interpretations,
3. how a nontrivial cohomological obstruction emerges, and
4. why the graph-only view misses the same structure.

2. A Three-Layer Synthetic System

Let

$$
\mathcal{L}={P,E,K},
$$

where:

• $P$: political or administrative actors,
• $E$: economic or logistics actors,
• $K$: illicit or criminal facilitators.

Let the entity set be

$$
V={p_1,p_2,e_1,e_2,e_3,k_1,k_2,b_1,c_1}.
$$

Interpret the entities as follows:

• $p_1$: deputy minister,
• $p_2$: customs official,
• $e_1$: state-adjacent holding company,
• $e_2$: logistics firm,
• $e_3$: shell importer,
• $k_1$: smuggling broker,
• $k_2$: criminal enforcement intermediary,
• $b_1$: regional bank,
• $c_1$: port terminal.

Assume the following observed weighted pairwise relations:

$$
\begin{aligned}
w(p_1,e_1)&=0.95, \cr
w(e_1,b_1)&=0.93, \cr
w(b_1,e_3)&=0.90, \cr
w(e_3,k_1)&=0.88, \cr
w(k_1,p_2)&=0.86, \cr
w(p_2,c_1)&=0.91, \cr
w(c_1,e_2)&=0.89, \cr
w(e_2,e_3)&=0.87, \cr
w(k_1,k_2)&=0.80.
\end{aligned}
$$

These ties are meant to represent a stylized chain linking political influence, financial routing, shell importation, customs access, logistics handling, and illicit brokerage.

In addition, suppose a transaction dossier provides direct evidence of a coordinated three-way shipment-clearance event involving the customs official, the logistics firm, and the smuggling broker. We therefore include the 2-simplex

$$
\sigma=[p_2,e_2,k_1].
$$

This simplex is not inferred from pairwise closure. It is introduced only because the higher-order interaction is itself observed.

3. Source Contradiction

Now impose one deliberate contradiction concerning the control status of $e_2$.

• Source A: $e_2$ is an ordinary private logistics firm with no state control.
• Source B: $e_2$ is beneficially controlled by $p_1$ through a nominee structure.

Each of these claims is locally plausible on the evidence available to the source reporting it. The analytic question is whether they can be made globally consistent in a single integrated model.

That is a local-to-global problem. It is not handled naturally by ordinary graph summaries. It is exactly the type of problem for which a sheaf-theoretic formalism is useful.

4. The Filtered Simplicial Complex

Let $K_t$ denote the simplicial complex obtained by thresholding the weighted evidence at level $t$, together with inclusion of the directly observed simplex $[p_2,e_2,k_1]$ once its own evidence threshold is satisfied.

Consider the filtration

$$
K_{0.92} \subset K_{0.87} \subset K_{0.80}.
$$

4.1 Stage $K_{0.92}$

At threshold $t=0.92$, only the strongest edges survive:

$$
(p_1,e_1), \qquad (e_1,b_1).
$$

At this stage there is no nontrivial 1-cycle.

4.2 Stage $K_{0.87}$

At threshold $t=0.87$, we additionally include

$$
(b_1,e_3), \qquad (p_2,c_1), \qquad (c_1,e_2), \qquad (e_2,e_3).
$$

There is still no closed 1-cycle spanning the three layers.

4.3 Stage $K_{0.80}$

At threshold $t=0.80$, we additionally include

$$
(e_3,k_1), \qquad (k_1,p_2), \qquad (k_1,k_2),
$$

and the directly evidenced 2-simplex

$$
[p_2,e_2,k_1].
$$

Now the 1-skeleton contains the loop

$$
\gamma = p_2 \to c_1 \to e_2 \to e_3 \to k_1 \to p_2.
$$

This loop spans the political, economic, and illicit layers:

$$
P:{p_2}, \qquad E:{c_1,e_2,e_3}, \qquad K:{k_1}.
$$

Crucially, $\gamma$ is not filled by supported higher-order structure. The simplex $[p_2,e_2,k_1]$ fills one local triangle, but there is no evidentiary support for simplices such as

$$
[p_2,c_1,e_2], \qquad [e_2,e_3,k_1],
$$

or any higher-dimensional cell filling the entire loop.

Hence

$$
\gamma \in Z_1(K_{0.80}) \setminus B_1(K_{0.80}),
$$

so that

$$
[\gamma] \neq 0 \in H_1(K_{0.80};\mathbb{F}_2).
$$

Dually, one may speak of a nonzero class in first cohomology:

$$
H^1(K_{0.80};\mathbb{F}_2) \neq 0.
$$

Analytic interpretation

The point is not that a conspiracy has been proved. The point is narrower and more disciplined. The model contains a robust mediated circuit linking customs authority, port handling, logistics, shell importation, and smuggling brokerage. That circuit does not collapse into a fully observed clique or fully filled higher-order event.

This is the sort of structural feature that a plain graph can display visually but cannot characterize with the same precision.

5. Cochains and Cohomology

Let $K$ be a finite simplicial complex over a field $\Bbbk$. For each $k \ge 0$, let

$$
C^k(K;\mathbb{k})=\operatorname{Hom}(C_k(K;\mathbb{k}),\mathbb{k})
$$

denote the space of $k$-cochains, with coboundary

$$
\delta^k : C^k(K;k) \to C^{k+1}(K;k), \qquad \delta^{k+1} \circ \delta^k = 0.
$$

The $k$th cohomology group is

$$
H^k(K;\Bbbk)=\ker \delta^k \big/ \operatorname{im}\delta^{k-1}.
$$

In the present toy model, the nontrivial first cohomology class corresponds to a robust cross-layer cycle that survives thresholding and remains unfilled by supported higher-order cells.

6. A Cellular Sheaf of Control-Status Interpretations

Ordinary cohomology sees topology of the underlying complex. It does not, by itself, encode source semantics or compatibility conditions for competing interpretations. To do that, we equip the complex with a cellular sheaf.

Let $X$ be the cell complex underlying $K_{0.80}$. Define a cellular sheaf $\mathcal{F}$ on $X$.

6.1 Stalks

For each vertex $v \in X_0$, let

$$
\mathcal{F}(v)=\mathbb{R}^2,
$$

with basis vectors

$$
\mathbf{p}=\text{private/autonomous}, \qquad
\mathbf{s}=\text{state-linked/patronage-controlled}.
$$

For each edge $e \in X_1$, let

$$
\mathcal{F}(e)=\mathbb{R}^2.
$$

For the 2-simplex $\sigma=[p_2,e_2,k_1]$, let

$$
\mathcal{F}(\sigma)=\mathbb{R}^2
$$

as well.

6.2 Restriction maps

For most face relations $\tau \le \sigma$, take the restriction maps to be the identity:

$$
\rho_{\sigma\tau}=I_2.
$$

This encodes the default requirement that local interpretations agree on overlaps.

Now focus on the contradictory neighborhoods around $e_2$. Let

$$
a=(c_1,e_2), \qquad b=(e_2,e_3).
$$

Interpret edge $a$ as a neighborhood read primarily through Source A, and edge $b$ as a neighborhood read primarily through Source B. Define

$$
\rho_{a,e_2}=I_2,
$$

but

$$
\rho_{b,e_2}=M,
\qquad\text{where}\qquad
M=
\begin{pmatrix}
0 & 0 \cr
0 & 1
\end{pmatrix}.
$$

The interpretation is straightforward:

• along $a$, the control-status assignment at $e_2$ may remain private,
• along $b$, only the state-linked component is admissible.

This is a deliberately simple sheaf. Its purpose is not full realism. Its purpose is to show explicitly how locally credible assignments can fail to glue globally.

7. The Sheaf Obstruction

The sheaf cochain spaces are

$$
C^k(X;\mathcal{F})=\bigoplus_{\dim \sigma = k}\mathcal{F}(\sigma),
$$

with coboundaries induced by the incidence structure and the restriction maps. Sheaf cohomology is then

$$
H^k(X;\mathcal{F})=\ker d^k / \operatorname{im} d^{k-1}.
$$

A (0)-cochain is an assignment

$$
x\in C^0(X;\mathcal{F})=\bigoplus_{v\in X_0}\mathcal{F}(v)
$$

of a local state vector to each vertex.

The obstruction can be seen already in the two neighborhoods incident to (e_2). Recall that

$$
a=(c_1,e_2), \qquad b=(e_2,e_3).
$$

Let Source A induce the local requirement that $e_2$ be interpreted as private/autonomous:

$$
x_{e_2}=\mathbf{p}.
$$

Along edge $a$, the restriction map is the identity:

$$
\rho_{a,e_2}=I_2.
$$

Thus

$$
\rho_{a,e_2}\mathbf{p}=I_2\mathbf{p}=\mathbf{p},
$$

so Source A’s local interpretation is compatible with the $a$-neighborhood.

Now let Source B induce the local requirement that the $b$ -neighborhood interpret $e_2$ as state-linked or patronage-controlled. In this neighborhood, the restriction map is

$$
\rho_{b,e_2}=M,
\qquad\text{where}\qquad
M=
\begin{pmatrix}
0 & 0 \cr
0 & 1
\end{pmatrix}.
$$

This map suppresses the private component and preserves only the state-linked component. Therefore

$$
\rho_{b,e_2}\mathbf{p}=M\mathbf{p}=0.
$$

If the $b$-neighborhood requires a nonzero state-linked interpretation, then the private assignment from Source A cannot satisfy the compatibility condition on $b$. It does not restrict to the section required by Source B.

Conversely, suppose we assign

$$
x_{e_2}=\mathbf{s}.
$$

Then

$$
\rho_{b,e_2}\mathbf{s}=M\mathbf{s}=\mathbf{s},
$$

so the assignment satisfies the Source B neighborhood. But it no longer represents Source A’s claim that $e_2$ is an ordinary private logistics firm.

Hence the two local interpretations cannot both be restrictions of a single global section under the present semantics. The contradiction is not merely that two sources disagree. The stronger point is that, once the source claims are encoded as local sections with compatibility maps, the model itself has no global assignment that preserves both interpretations simultaneously.

Equivalently, the obstruction appears as a failure of gluing. The locally plausible sections over the $a$ - and $b$-neighborhoods agree with their own evidentiary contexts, but they do not extend to a single coherent global section on $X$.

In a fully specified finite cellular sheaf, this failure can be detected by the coboundary operator

$$
d^0 :C^0 (X;\mathcal{F})\to C^1 (X;\mathcal{F}),
$$

whose components measure disagreement between vertex assignments and edge-level compatibility requirements. A global section is precisely a $0$-cochain in

$$
\ker d^0.
$$

The Source A and Source B assignments define locally valid sections, but their attempted amalgamation has nonzero coboundary on the edge $b=(e_2,e_3)$. Thus the obstruction is witnessed by the fact that the combined local data do not lie in $\ker d^0$.

If the sheaf is extended so that the incompatible local source claims are treated as fixed edge- or neighborhood-level observations, then the obstruction may also be represented as a nontrivial cohomology class: the disagreement is not removed by changing a vertex assignment without violating one of the local source constraints. In that sense, the obstruction is naturally cohomological. The precise group in which it appears depends on the final sheaf specification, but the analytic fact is already visible here: the present model cannot glue the two locally plausible control-status claims into a single globally coherent interpretation.

Analytic interpretation

This does not mean one source must simply be false. It means the current model cannot simultaneously accommodate all locally plausible claims without at least one of the following moves:

1. introducing a hidden variable, such as an unobserved nominee, intermediary, or temporal regime switch,
2. weakening or discounting one of the source claims, or
3. refining the semantics of control so that legal ownership and operational direction are not forced into the same binary category.

This is the central local-to-global lesson. The contradiction is not merely “messy data.” It is represented as a mathematically explicit gluing failure. With a fully specified cellular sheaf, that gluing failure can be studied cohomologically rather than treated as an informal source disagreement.

8. Why the Plain Graph View Is Insufficient

Suppose the same evidence is collapsed into an ordinary weighted graph $G$ on vertex set $V$.

Standard graph analysis will recover some useful facts:

• $k_1$ will likely have high betweenness,
• $p_2$ will appear as a bridge between administrative and illicit neighborhoods,
• $e_3$ will sit between financial and logistics substructures,
• $e_2$ will look like a moderate-degree intermediary.

None of that is wrong. It is simply incomplete.

8.1 The graph misses the distinction between an unfilled cycle and a dense subgraph

In the graph, the subgraph on

$$
{p_2,c_1,e_2,e_3,k_1}
$$

looks like a connected component with a loop. But graph summaries do not natively distinguish between:

1. a loop that persists because there is no supported higher-order closure, and
2. a loop that is merely one face of a filled higher-order structure.

The simplicial model does distinguish these cases. The 2-simplex $[p_2,e_2,k_1]$ fills one triangle, but it does not kill the larger cycle $\gamma$.

8.2 The graph loses the provenance of the higher-order event

The transaction dossier directly supports a three-way coordinated event:

$$[p_2,e_2,k_1].$$

In the graph, this becomes only its pairwise shadows unless one adds special annotations. The simplicial model records that the event was jointly evidenced as a multi-actor interaction, not reconstructed post hoc from edge co-occurrence.

8.3 The graph cannot express the local-to-global contradiction intrinsically

The contradiction about the control status of $e_2$ can be stored in a graph as two source notes or competing node attributes. But the graph itself provides no natural mechanism for asking whether the local claims can be made globally coherent across overlaps.

The sheaf does.

That is not a cosmetic distinction. It is a genuinely different inferential capability.

9. What the Toy Model Shows

The toy model supports four narrow but important claims.

First, a filtered simplicial complex can retain higher-order event structure that an ordinary graph discards.

Second, first cohomology can identify a cross-layer cycle that remains structurally meaningful because it is not filled by supported higher-order cells.

Third, a cellular sheaf can encode competing local interpretations in a disciplined way.

Fourth, nontrivial first sheaf cohomology can represent an explicit obstruction to gluing locally plausible reports into a globally coherent account.

None of this proves causation. None of it proves criminal intent. None of it repairs weak collection or poor entity resolution. What it does provide is a mathematically explicit language for structural features and consistency failures that analysts often discuss informally but rarely model rigorously.

10. Conclusion

The value of the toy model is not that it is realistic in every operational detail. It is useful because it forces the abstractions to cash out.

The filtered complex captures a persistent cross-layer loop spanning administrative, logistics, financial, and illicit interfaces. The sheaf records a contradiction that survives local plausibility but fails global integration. The resulting obstruction is analytically meaningful: it signals that the current model cannot support every local claim simultaneously without adding hidden structure, weakening evidentiary commitments, or refining the semantics of interpretation.

That is the main methodological point. In elite interlinkage analysis, the relevant object is often not merely a graph. It is a filtered, typed, higher-order relational complex endowed with compatibility constraints. In that setting, cohomology and sheaf cohomology become useful because they measure precisely what ordinary graph summaries leave out.