The Shape of Deceit: Persistent Homology and the Topological Structure of Manipulative Communication

Every corpus of communication has a shape. Not a shape visible in any single message, but a topological structure that emerges when hundreds of statements are embedded in high-dimensional semantic space and their collective geometry is analysed. Truthful communication has one shape — dense, connected, smoothly distributed across the relevant semantic landscape. Manipulative communication has a distinctly different shape — riddled with loops of circular reasoning, punctuated by voids where topics are conspicuously avoided, and clustered around rehearsed positions that connect poorly to each other. Persistent homology, the central tool of topological data analysis, provides the mathematics for detecting and characterising these structural features.

The Mathematical Foundation

Persistent homology identifies topological features in data that persist across multiple scales of analysis. The procedure begins by constructing a sequence of simplicial complexes — combinatorial approximations of the data's shape — at increasing distance thresholds. At a small threshold, only the nearest points are connected and the topology reflects local structure. At a large threshold, distant points merge and the topology reflects global structure. Features that appear at one threshold and persist across many are structurally significant. Features that appear briefly and vanish are noise.

The output is a persistence diagram: a multiset of points in the birth-death plane, where each point represents a topological feature. The birth coordinate is the threshold at which the feature first appears. The death coordinate is the threshold at which it disappears. Features far from the diagonal — those with long lifetimes — are the persistent, structurally meaningful elements of the data. Features near the diagonal are ephemeral noise. This clean separation between signal and noise is what makes persistent homology particularly valuable for the high-dimensional, noisy data that communication analysis produces.

Topological Features in Communication

Three classes of topological feature carry diagnostic significance when persistent homology is applied to communication corpora.

Loops (H₁ features) — closed cycles in the semantic graph — correspond to circular reasoning. A party who repeatedly traverses the same chain of justification, arriving back at the starting premise each time from a different entry point, creates a topological loop in the embedding space. One cycle might be coincidence. A persistent loop that survives across many scales of analysis reveals a structural feature of the reasoning: the argument is circular, and the circularity is a fundamental property of the narrative rather than an artefact of any particular phrasing.

Voids (H₂ features) — holes in the semantic coverage — correspond to topics that are conspicuously avoided. Every communicator has a natural semantic footprint shaped by their role, knowledge, and the context of the dispute. Voids within that footprint — regions of the semantic space that the surrounding context implies should contain statements but do not — indicate deliberate avoidance. In legal analysis, persistent voids correlate with undisclosed information, topics the party has been advised not to discuss, or areas where truthful statements would undermine their position.

Clusters (H₀ features) — dense connected components at small scales that remain separated at moderate scales — correspond to rehearsed talking points. A communicator who has prepared specific positions produces tight semantic clusters around those positions, with sparse connections between them. Spontaneous, truthful communication produces a more uniform distribution because genuine knowledge generates statements that transition smoothly between related topics. The ratio of intra-cluster density to inter-cluster connectivity measures how rehearsed versus spontaneous a communication pattern is.

Persistence Barcodes and Betti Numbers

The persistence barcode — a horizontal representation where each feature is a bar spanning from birth to death — provides an intuitive visual summary. Long bars represent persistent features; short bars represent noise. The Betti numbers at each scale count the features of each dimension: β₀ counts connected components, β₁ counts loops, β₂ counts voids.

For honest communicators, the barcode is simple: a small number of long bars (corresponding to the genuine topological structure of their account) and a large number of short bars (noise). For manipulative communicators, the barcode is complex: many medium-length bars that reflect the pseudo-structure of a fabricated narrative — features that are more persistent than noise but less persistent than genuine structural elements. This intermediate persistence is itself a diagnostic signature.

Scale Independence

The deepest advantage of persistent homology is its scale independence. The same structural features are detectable whether the analysis operates on individual sentences, paragraphs, or entire documents. Manipulation strategies that defeat word-level or sentence-level analysis — careful phrasing, euphemistic language, strategic ambiguity — still produce detectable topological signatures at the corpus level because topology captures the global shape of the data rather than its local content. A communicator can control what they say at each moment but cannot easily control the emergent topology of everything they have ever said.

Complementary Analysis

Persistent homology operates on a fundamentally different analytical axis from the statistical and information-theoretic metrics used by other tools in this suite. While Cauchy convergence measures sequential consistency, KL divergence measures anchor drift, and Wasserstein distance measures distributional shift, persistent homology measures the shape of the communication space itself. A narrative could pass all statistical tests while containing topologically detectable circular reasoning or conspicuous avoidance voids. The combination of statistical and topological analysis creates a multi-layered detection framework that is substantially more difficult to defeat than any single approach.