The Graph Was the Weapon - Decipon Insights

What the Graph Showed

On the night of the June 2026 Los Angeles mayoral primary, betting odds for MAGA-backed candidate Spencer Pratt started falling.

This was normal. Mail-in ballots were being counted — a slow, standard California procedure. As votes accumulated, the totals were shifting toward the other candidates. The odds were shifting to reflect that.

Influencer Mila Joy posted to her 500,000 followers: “They are actually doing it. They are counting votes until SPENCER LOSES. Someone DO SOMETHING.” She attached a Polymarket graph showing Pratt’s odds declining as the count progressed.

David Freeman, who posts under the handle Gunther Eagleman, asked his followers: “Is CA cheating to get Spencer Pratt out?” alongside Kalshi odds data.

Both posts carried small “paid partnership” disclosures. Kalshi and Polymarket had paid these influencers to promote their platforms.

No one at either company had asked them to post anything about election fraud. No one had scripted the allegations. No one had coordinated the campaign. The companies’ own financial infrastructure had simply created the conditions for it — and the conditions had produced it automatically.

The Mechanism

To understand what happened in Los Angeles, you need to understand what prediction market odds actually measure.

They measure speculative betting. Thousands of people placing money on who they think will win, with prices updating in real time as bets flow in. When odds shift, it means the composition of bets has changed. It does not mean the vote count has changed in an anomalous way. It means bettors updated their expectations based on the same publicly available count everyone else was watching.

But odds graphs look authoritative. They have axes. They have lines. They move in real time. They have the aesthetic texture of data.

When Pratt was leading in betting odds while trailing in actual vote tallies — a divergence that emerged because his X-platform engagement made him look more viable than polling suggested — the graph looked like a candidate who “should” be winning. When the count caught up with reality and Pratt fell behind, the graph showing that decline looked, to someone already primed for fraud narrative, like evidence of something happening to the count.

The graph was not evidence of anything except normal ballot counting. But it was being presented, by people with large audiences, in a format that looked like evidence. And those people had been paid to post about those graphs.

The 2020 Blueprint

This is not new territory. The visual mechanics were documented in detail after the 2020 election.

The “red mirage / blue shift” phenomenon — in which early returns favored Trump because in-person votes were counted first, then shifted toward Biden as mail-in ballots were tabulated — was exploited for months as visual “proof” of fraud. A FiveThirtyEight chart showing Biden’s trajectory as votes were counted was widely shared with fraudulent context stripped away, the line’s sudden rise repurposed as evidence of ballot injection.

Election officials and journalists spent enormous effort explaining, repeatedly, why vote totals shift during counting. The explanations were accurate. They were also less visually compelling than the graphs, and they required context the graph didn’t carry on its own.

What changed in 2026 is the source of the graph and the incentive structure around it.

In 2020, the weaponized visuals were charts from news organizations and official sources, stripped of context by hostile actors. In 2026, the visuals came from prediction market companies that had paid their own marketing infrastructure to distribute them — and that marketing infrastructure had optimized for engagement, not accuracy.

The Incentive Architecture

Kalshi and Polymarket operate influencer programs that pay creators — up to $500 per post — to promote their platforms on X. The terms were loose: mention the markets, grow the audience, capture the commission. Content oversight was minimal.

This creates a specific economic condition. Influencers who cover elections, and who have audiences primed to distrust electoral institutions, now have a financial reason to post about odds — and odds graphs are their promotional content. When they post the graphs in political context, the graph carries the implicit argument: here is what the market thinks is happening. When the market’s assessment diverges from the official count, posting the graph in real time becomes posting an implicit challenge to the count.

No one needed to instruct Mila Joy to post “they are counting votes until Spencer Loses.” She was paid to promote Polymarket. Polymarket’s odds were changing during a count. She posted the graph with her own commentary. The fraud allegation was not a campaign directive. It was the natural product of financial incentive meeting political priming meeting a changing visualization in real time.

The operation, if you want to call it that, required no operator. The architecture produced the output.

Disclosure That Doesn’t Disclose

Both posts carried “paid partnership” labels.

This matters because disclosure is the standard regulatory fix for sponsored content. If audiences know a post is paid, the argument goes, they can calibrate accordingly.

But disclosure addresses one specific risk: that the audience doesn’t know the content is commercial. What happened here is not that risk. The harm was not that audiences were unaware these influencers had financial relationships with Polymarket. The harm was that audiences were misled about what a prediction market graph means as evidence.

A small “paid partnership” label at the bottom of a post claiming active electoral fraud does not tell viewers: “this odds graph reflects speculative betting behavior, not vote-counting anomalies, and the candidate it appears to favor was overrepresented on this platform because his X engagement was disproportionate to his actual polling support.” That is what needed to be disclosed. What was disclosed was the commercial relationship.

Disclosure, as a tool, is calibrated to the wrong threat.

What the Companies Did

After NPR began asking questions, both platforms moved.

Kalshi told influencers to delete the posts and announced new rules prohibiting creators from “questioning the integrity or accuracy of an election, legal ruling or official determination in connection with an election.” Polymarket cited existing guidelines against misinformation — guidelines that had not been enforced — and removed some “paid partnership” tags from posts.

Neither action addressed the underlying condition. The same financial incentives remain. The same influencer pool remains. The same real-time odds graphs remain, updating during future vote counts. The rule changes add a prohibition that influencers must now navigate — but the economic logic that produced the posts hasn’t changed.

Election expert Stephen Richer put it plainly: this is a preview of what November midterms will look like. A candidate who overperforms in prediction market betting will build an audience expectation. When ordinary counting reveals a different result, the same graph will be available to post, the same influencers will be paid to engage with market content, and the same fraud narrative will be available to apply.

The rule change is reactive. The architecture it is responding to is still operational.

Why the Graph Works

Visual disinformation doesn’t succeed because it’s technically false. It succeeds because it exploits the gap between what a visualization actually shows and what audiences understand it to show.

The FiveThirtyEight chart in 2020 was technically accurate. The odds graphs in 2026 are technically accurate. The numbers are real. The lines are real. The manipulation is in the framing — stripping the contextual information that makes the number interpretable and replacing it with a narrative frame that fills the interpretive gap with something untrue.

In 2020, the stripping was done by hostile actors taking accurate charts and reposting them without context. In 2026, the stripping happened at the posting stage: the influencer provides the graph and the fraud narrative simultaneously, without the context that would make the graph interpretable correctly. The audience gets the visual authority of data and the narrative of fraud, bundled.

The reason the “paid partnership” label doesn’t help is that it addresses the commercial relationship, not the stripped context. What would help — a requirement that posts showing real-time odds during active vote counts include explanatory language about what odds measure — would require platforms to enforce content standards that currently don’t exist, against creators they are paying to post.

The Structural Problem

The Los Angeles case is not primarily about Mila Joy or Gunther Eagleman making deliberate choices to spread disinformation. It is about what emerges when you put financial incentives, real-time changing metrics, politically primed audiences, and minimal content standards in the same system.

Any platform that pays people to post about numbers that change during observable events will produce this. If the numbers change in ways that contradict audience expectations — expectations the platform itself may have helped build by allowing overconfident interpretation of its data — some fraction of those posts will present the change as evidence of something sinister.

The companies involved are not bad actors by design. They built a marketing program using standard influencer mechanics. The marketing program generated disinformation as a byproduct, because the product being marketed — real-time odds on contested elections — is intrinsically legible as election-fraud evidence to audiences that are already suspicious of electoral institutions.

Nobody designed this. Nobody needed to.

The incentive was the operation.

This article is part of Decipon’s Manipulation Breakdowns series, examining specific influence operations through the Influence Tactics Protocol.

Sources: