Concept · information-theory
Probability Infrastructure
The concept of prediction market mechanisms as a general-purpose layer that embeds live probability signals into decision surfaces beyond trading · attention, credibility, demand. The "endgame" framing where PMs become invisible plumbing, not destination products.
Key insights
- Forecasting accuracy has outpaced product design. A decade inside the US intelligence community produced zero complaints about forecast quality, yet forecasting firms remain niche while simulation startups (Aaru, Simile) raise nine-figure rounds. Gap: failure to embed forecasts into institutional workflows. Recommendation: forecasting companies hire deployment managers to transform probabilities into artifacts clients can act on (Mohamed Elrashid, "Good Forecasts, Bad Products").
- Dan Schwarz analyzes ~13,500 Polymarket contracts: >80% volume in sports/crypto/elections; accuracy on "useful" markets hasn't improved since early 2025. AI chatbots may supersede PMs as the primary forecasting interface, leaving markets to serve an epistemic role as common-knowledge infrastructure · a Schelling-point shared probability, not a personal-use tool (Asterisk).
- Probability layers thesis: PMs are a proof-of-concept for a broader shift · probability as infrastructure. Three layers beyond trading: attention markets (price content virality forward), credibility markets (trust as continuously updated score), demand markets (consumer intent before production). Endgame: probability signals embedded invisibly into every decision surface on the internet (Aggie).
- Elrashid specifics: Cultivate Labs ran a decade-long PM inside US IC, decommissioned in 2020 for institutional reasons. "In 10+ years, I can't remember a single time that someone told us they had issues with the forecasts not being accurate enough." Open Philanthropy / Coefficient Giving has put $50M+ across 30+ grants into forecasting field; the field has not changed how important decisions get made. Good Judgment Inc., Hypermind (Paris, 2014), INFER/RAND Forecasting Initiative · all respected, all "muted" impact.
- Elrashid on the simulation alternative: Aaru built a synthetic panel of 3,600 respondents that replicated a 6-month wealth research survey in a day for EY, sometimes more accurately than the original surveys. Aaru raised a $50M Series A in Dec 2025 at ~$1B valuation (customers: Accenture, EY, IPG, political campaigns). Simile raised $100M Series A Feb 2026 from Index, building on Joon Park's generative-agents work (customers: CVS · 9,000 stores shelf placement, Gallup, Wealthfront). Mantic, focused on probability briefs, raised $4M pre-seed in same period. Implication: the format of the output determines the market, not the underlying forecast quality.
- Elrashid's prescription: forecasting companies need to hire "Deployment Managers and Forward Deployed Engineers" (Palantir-style) · turn probabilities into the shape the customer's decision loop accepts. Target buyer profiles: scenarios teams at energy majors, hedge funds, Lloyd's-style specialty insurance, foreign ministries, policy planning staffs.
- Schwarz on the substitution risk: ChatGPT/Claude/Gemini already serve as primary forecasting interfaces for casual users, providing all 5 use-categories (risk monitoring, news interpretation, policy outcomes, accountability, novel info) implicitly. Schwarz's bet: "before [PMs deliver Vitalik's info-finance vision], Claude will be the only forecaster anyone will ever want to ask about the future." But there's an institutional channel for PMs: "By accelerating the adoption of probabilities by mainstream media, prediction markets help build common knowledge." PMs win not on accuracy but on visibility.
In their words
Probability signals embedded invisibly into every decision surface on the internet.· Aggie
Forecasting accuracy has outpaced product design.· Mohamed Elrashid
AI chatbots may supersede prediction markets as the primary forecasting interface, leaving markets to serve an epistemic role as common knowledge infrastructure.· Dan Schwarz
Where it matters
"Probability infrastructure" is the most ambitious end-state framing for the PM industry. It says the product isn't the trading UI · it's the probability output surfaced inside other products (a price ticker, a CMS, an ad-buying tool, a content recommender). Three things follow: (1) PM platforms become picks-and-shovels for other apps, not destination apps themselves; (2) the moat shifts from UX to API + data + canonicalization; (3) accuracy stops being the only KPI · adoption and embeddability matter more. For Dekant, this framing supports a thesis where the distribution-market output (a full belief curve) is more valuable to embed than a single binary price, because downstream apps can integrate the whole shape.
Connections
- Info finance · Vitalik's umbrella term for the same idea
- Credibility markets · one of three named probability layers
- Demand markets · another named probability layer
- Attention markets · third named probability layer
- Nowcasting · embeds PM prices into econ data workflows
- Forecasting accuracy · the input that probability infrastructure exposes
- Decision markets · adjacent embedding (probabilities → governance choices)
- Platform competition · moat shifts from app to API
Platforms linked to this concept
- Dekant · thesis · Argues for/positions around Probability Infrastructure
- Good Judgment Open · addresses · Proposes a solution to issues raised under Probability Infrastructure
- Polymarket · implements · Mentioned in Probability Infrastructure content as an implementing platform
Related concepts
- Info Finance
- Credibility Markets
- Demand markets
- Attention Markets
- Nowcasting
- Forecasting Accuracy
- Decision Markets
- Platform competition
Sources
- Good Forecasts, Bad Products · Mohamed Elrashid · Apr 24, 2026
- Are Prediction Markets Good for Anything? · Dan Schwarz · Apr 1, 2026
- The Probability Layers Are Coming · Aggie · Mar 30, 2026