External Reference — Archived 2026-03-27

Inside the Markets Aggregating Political Reality

How Kalshi and Polymarket moved prediction markets from niche financial instruments to mainstream political analysis — and the governance challenges that scale creates

Andy Hall · Published November 19, 2025 · Free Systems (Substack) · Stanford GSB / Hoover Institution
Archived Reference. This is a summary and annotation of an external article by Andy Hall, archived for Kyanos research. Original published at Free Systems on Substack, November 19, 2025. Andy Hall is a professor at Stanford GSB and Senior Fellow at the Hoover Institution. Disclosure noted in the original: Hall receives consulting income from a16z crypto (a recent Kalshi investor) and Meta Platforms; his writing is stated as independent of those roles.

Core Argument

Prediction markets have moved from niche financial instruments to mainstream political analysis infrastructure. Kalshi and Polymarket now regularly appear in major media coverage, their data surfaces in Google results, and their prices move markets in real time based on turnout data and social media commentary — often before official results are known. Hall observed this firsthand during a Virginia election night.

The potential is significant: Hall cites Rhode and Strumpf's research showing early 20th-century betting markets were historically accurate at predicting outcomes until modern polling emerged. At scale, prediction markets can function as public goods, aggregating dispersed information into a shared picture of political reality. But that potential depends on solving three governance challenges that scale creates: manipulation, discovery, and the definition of truth for contested outcomes.

Scale Context

Combined Kalshi and Polymarket trading exceeded $7 billion in the month preceding this article. The 2024 presidential election generated $1+ billion in combined volume. A November 2025 off-cycle election — by definition a low-attention, down-ballot moment — generated $400+ million in volume during the final week alone. These are not niche instruments.

Market Scale Data

$7B+
Combined Kalshi + Polymarket monthly volume, Nov 2025
$1B+
Combined volume, 2024 presidential election
$400M+
Off-cycle election volume, final week Nov 2025
$300M
Early 20th-century election betting peak (2025 dollars)

Three Governance Challenges

Challenge 1
Manipulation and Unintended Consequences

Three manipulation vectors: (1) Narrative influence — when markets shape public perception, actors have incentives to move them. The Virginia race Hall observed shifted before official results, driven by turnout data and X posts rather than votes. (2) Insider manipulation — subjects of markets can, in some contexts, determine outcomes. Election markets face lower risk since altering actual elections is difficult; Kalshi's policy excluding vote counters from betting provides a relevant safeguard. (3) Self-fulfilling prophecies — evidence exists that polls affect turnout; markets increasingly function like polls at scale.

Challenge 2
Discovery and Resolution at Scale

As contract proliferation accelerates, attention becomes the bottleneck. Hall observed traders discovering thousands of contracts they wished to monitor only after the election concluded. Solutions include better ranking algorithms, AI agents surfacing relevant markets, and social discovery features.

Resolution is harder. Two cautionary examples: Venezuela's 2024 election created chaos when the UMA oracle resolved contrary to initial Polymarket pricing — the market rules about "official information" versus "consensus of credible reporting" conflicted when the official result was disputed. A U.S. government shutdown contract resolved incorrectly because the resolution source (the OPM operating status page) wasn't updated until the day after the actual event. Contract resolution is an epistemology problem, not just a data problem.

Challenge 3
Defining Truth for Contested Outcomes
When markets function as information sources, defining what constitutes a valid outcome becomes politically fraught. The New York mayoral race involved questions about whether an elected official would actually be sworn in. Prediction markets can only resolve against verifiable facts — but many political outcomes are contested at the level of what the relevant facts are. The governance challenge is building resolution frameworks robust enough to handle genuine ambiguity without becoming manipulation surfaces.

Key Findings

Finding 1
Markets Now Move Before Official Results
Hall's election night observation: prediction market prices shifted based on turnout data and X commentary before any official results were certified. At sufficient scale, markets may function as faster-than-official information aggregators — but this speed advantage is also an attack surface. What moves the market moves before the facts arrive.
Finding 2
Historical Accuracy Precedent Exists
Rhode and Strumpf's research showed that early 20th-century election betting markets reached over $300 million in inflation-adjusted volume and were historically accurate at predicting outcomes — until modern polling supplanted them. Current volumes already exceed that historical peak. The information aggregation function of markets is empirically established; whether modern implementations can achieve equivalent accuracy given the governance challenges is the open question.
Finding 3
Media Integration Is Already Complete
Prediction market data is no longer a niche signal. Major media outlets regularly report market movements. Google surfaces market probabilities in search results. This media integration means market prices are now part of the information environment that AI systems retrieve — they appear in the crawlable, structured content that shapes AI political responses.

Kyanos Annotation: Applicability to Our Work

This article introduces a dimension of the AI political information environment that the Hall series addresses only indirectly elsewhere: prediction market data as an AI information source.

Polymarket prices appear in AI retrieval. Because Google and major media outlets surface Kalshi/Polymarket data, these prices are now part of the crawlable information landscape that AI systems retrieve when answering questions about political races. When a voter asks an AI chatbot "who is likely to win the Montana Senate race?", the AI may retrieve market probability data alongside news coverage and official sources. This is not speculative — it is a direct consequence of media integration at scale. AUDIT methodology should account for whether market probabilities appear in AI-retrieved context for a given candidate's race.

The resolution failure cases are directly relevant to AUDIT. The Venezuela and shutdown resolution failures illustrate a general principle: when official sources conflict with consensus reporting, AI systems that retrieve "authoritative" sources may get the answer wrong. This is the same structural problem Hall documents in "AI Is A Shitty Political Advisor" — what is officially crawlable does not always reflect reality. For AUDIT, the relevant implication is that AI responses about election outcomes or candidate records may be anchored in sources that resolved incorrectly.

Market narrative influence is an AIO attack surface. If actors can move prediction markets and those market movements are then surfaced by AI systems as political information, prediction markets become an indirect AIO channel. A candidate's opponent investing in negative market signals could — in theory — cause AI systems to retrieve and relay market-derived pessimism about the candidate's chances. This is a downstream concern, not an immediate action item, but it belongs in the threat model for AI political information systems.

Archived for Kyanos research use  ·  Original article: Andy Hall, Free Systems (Substack), November 19, 2025  ·  All content is the author's own  ·  Annotation by Kyanos team