Moving the Goalposts: Wasserstein Distance as a Measure of Narrative Manipulation

When a party to a dispute gradually shifts their narrative — reframing blame, withdrawing previously offered support, redirecting attention from accountability to conflict — the manipulation can be almost invisible at the level of individual messages. Each communication may appear reasonable in isolation. The deception lives not in what is said at any single point but in the aggregate redistribution of what gets talked about over time. The Wasserstein distance, also known as the Earth Mover's Distance, provides a mathematically rigorous framework for detecting and quantifying exactly this kind of structural narrative drift.

The Mathematical Foundation

The Wasserstein distance measures the minimum cost of transforming one probability distribution into another. Imagine two piles of earth of equal volume but different shapes. The Earth Mover's Distance calculates the least amount of work — mass multiplied by distance moved — required to reshape one pile into the other. When the distributions are similar, the cost is low. When they differ substantially, the cost is high.

Applied to communication analysis, the "distributions" in question are the relative proportions of semantic themes in a party's messages across different time periods. A workplace communication corpus might be decomposed into themes such as workload, blame, deadlines, support, accountability, credit, process, and conflict. Each time window produces a distribution over these themes — a snapshot of what the party is actually talking about and in what proportions.

The Wasserstein distance between a party's early and late distributions then answers a precise question: how much did this person's narrative have to move, and how far, to get from where it started to where it ended?

The Diagnostic Power of Transport Cost

Unlike simpler measures of distributional difference, the Wasserstein distance captures not just that distributions differ but how they differ — the geometry of the shift matters. A party who gradually increases discussion of deadlines while slightly reducing discussion of process produces a small transport cost, because the thematic movement is short-range and proportionate. A party who redirects narrative mass from support and accountability into blame and conflict produces a large transport cost, because the thematic movement crosses long semantic distances.

This geometric sensitivity is what makes the metric particularly suited to detecting manipulation. Goalpost-moving, by definition, involves relocating the terms of a dispute. The Wasserstein distance measures the cost of that relocation directly. It does not require a predefined taxonomy of manipulation tactics. It does not depend on keyword detection or sentiment classification. It measures the structural reorganisation of a narrative at the distributional level, making it resistant to sophisticated manipulation where individual statements are carefully worded but the aggregate pattern is deceptive.

The Optimal Transport Map

Beyond producing a single distance metric, the Wasserstein framework generates an optimal transport plan — a detailed accounting of which themes lost mass, which gained it, and the specific flows between them. This transport map is analytically rich and evidentiarily powerful.

In a workplace harassment case, the transport map might reveal that 15% of a respondent's communication mass shifted from "support" to "blame" over a six-month period, while 12% shifted from "accountability" to "conflict." These are not subjective characterisations. They are computed redistributions of measured thematic proportions, reproducible by any analyst using the same embedding and clustering methodology.

Presented visually, the transport map makes the narrative restructuring legible to non-technical audiences. A barrister can point to the diagram and state that the respondent's own communications demonstrate a measurable, systematic pivot from collaborative to adversarial framing — and the mathematics behind that statement are defensible under cross-examination.

Acceleration as Signal

A further diagnostic emerges from tracking cumulative Wasserstein drift over time. Organic topic evolution — the natural tendency for conversations to shift focus as circumstances change — produces roughly linear cumulative drift at a low rate. Deliberate narrative manipulation tends to produce accelerating drift, because the manipulation intensifies as the dispute escalates.

This acceleration signature distinguishes intentional reframing from innocent conversational evolution. A party whose cumulative drift curve is flat is talking about the same things in the same proportions. A party whose curve is steep and accelerating is progressively restructuring their narrative, and the rate at which they are doing so is increasing. The shape of the curve, not just its endpoint, carries diagnostic information.

Complementary Measurement

The Wasserstein distance operates on a different analytical axis than sequential consistency metrics such as Cauchy convergence. Where convergence analysis asks whether successive statements are internally consistent, Wasserstein analysis asks whether the overall distribution of topics is stable. A communicator could produce individually consistent messages while systematically shifting the terrain of the conversation — passing a Cauchy test while failing a Wasserstein test. Combining both metrics creates a multi-dimensional credibility profile that is substantially harder to defeat through careful wording alone. The manipulation must be invisible not just at the statement level but at the distributional level, across multiple independent mathematical measures simultaneously.