How Hedgebook is built
Hedgebook is a descriptive catalogue: every company in the S&P 500, mapped to the live Kalshi event-contract markets whose underlying state references that company's documented business. This page records how the catalogue is compiled, what its sentences claim, and what they deliberately do not.
What this artifact is, and is not
Hedgebook is a map. For each of the 500 index constituents it lists five public Kalshi markets — 2,500 company-market links across 49 markets — together with a short account of why each market touches the company's reported business, and a note on the traditional financial instruments that have historically expressed the same class of risk.
It is not an analytical research tool, a screening product, or a source of recommendations. No page measures an exposure, sizes a position, or expresses a view on whether any contract is cheap or rich. The catalogue records that a connection exists and where it is documented; it goes no further.
Snapshot and refresh cadence
The catalogue is frozen at a matching snapshot: the date the company-to-market graph was last recomputed. Market quotes refresh on a faster cadence — the pipeline re-reads the live Kalshi tape, re-exports the public dataset, and redeploys the site, so quote timestamps run ahead of the matching snapshot. Every page carries the compile date; nothing on the site is computed at request time.
- Matching snapshot
- 2026-06-01
- Issuer universe as of
- 2026-06-01
- Latest quote timestamp
- 2026-06-02T20:28:04.685660Z
- Quote source
- kalshi_live
- Dataset compiled
- 2026-06-02
The fixed-five selection method
Each company carries exactly five markets. The number is a design constraint, not a measurement: five is enough to show the breadth of state-contingent claims that touch a single issuer, and small enough that every admitted edge can meet the same evidentiary bar. For each issuer, a dedicated matching pass reads the company's SEC disclosures — Item 1A risk factors and Item 7 MD&A — against the live Kalshi universe and admits a contract only when a traceable business reference can be cited: regulatory exposure, customer concentration, commodity input, geographic exposure, or named corporate-event risk.
Nothing is matched by statistical similarity, price correlation, or thematic association. When fewer than five company-specific contracts clear the bar, the remaining slots are filled by broad-market contracts — rates, recession, index levels — whose relevance to any large constituent is structural rather than idiosyncratic, and the connective text says so plainly.
What a “why this company cares” line claims
Every company-market pair carries a short passage explaining the connection. The passage makes one claim: that the company's own filings document a business interest in the state variable the contract prices. It cites the kind of evidence (a disclosed customer, a named regulator, a commodity input, a geography) and stops there.
The line does not claim materiality, magnitude, or direction. It does not claim that the company hedges the risk, that it ought to, or that the contract's price implies anything about the company's value. A reader who wants the underlying evidence will find it in the company's 10-K, which is where every line originates.
Data sources
Three sources feed each compile. The Kalshi live tapesupplies bid, ask, last probability, status, category, and resolution date for every contract in the coverage set; quote data is Kalshi's. The index constituent set supplies sector, sub-industry, and market-cap weight for all 500 issuers. The EDGAR filings cache holds parsed Item 1A and Item 7 sections for every name, refreshed against SEC EDGAR. Alongside these, a category taxonomy buckets contracts into roughly thirty themes, and a curated hedge-class file records which traditional instruments express which categories of risk.
The descriptive register as a design constraint
The artifact is written in the third person, descriptive throughout. That is a method choice, not a legal afterthought. A catalogue that described the same markets in advisory language — what a treasurer might do, what a position might return — would be making claims the underlying evidence cannot support. The evidentiary bar set in the matching pass only licenses statements of record: this filing names this risk; this contract prices this state; these instruments have historically expressed this class of exposure.
Nothing on the site is investment advice, a recommendation, or a solicitation to buy or sell any contract or security. That sentence appears here as a description of what the method can produce, which is the same thing a disclaimer would require.
The compile, step by step
The shape of the system
Hedgebook is a compile pipeline that runs end-to-end over the entire S&P 500. Each compile ingests the live Kalshi contract universe, the index constituent set, and a parsed cache of SEC filings for every name. A per-issuer matching pass decomposes each company into the small set of state-variables that actually drive its business, composes the connective text against the resulting bipartite graph of roughly 2,500 company-to-contract edges, and freezes the entire corpus to static HTML before the page is ever requested. No state persists between compiles; every page on the site is a fresh deterministic projection of that compile’s inputs.
Inputs
Three live inputs and two static ones. The live tape pulls bid, ask, last probability, status, category, and resolution date for every active and recently-resolved Kalshi event contract — tens of thousands of instruments, filtered down to the few dozen that currently sit inside the artifact’s coverage envelope. The constituent set carries sector, sub-industry, and market-cap weight for all 500 issuers. The filings cache holds the parsed Item 1A risk factors and Item 7 MD&A for each name, kept fresh against EDGAR. Alongside these, a category taxonomy buckets contracts into roughly thirty themes, and a hedge-class metadata file records which traditional financial instruments express which categories of risk.
Issuer-level decomposition
For each of the 500 issuers, a dedicated matching process reads the company’s SEC disclosures alongside the live Kalshi universe and selects the five contracts whose underlying state references the company’s documented business. The match criteria are explicit and structural: regulatory exposure, customer concentration, commodity input, geographic exposure, named corporate-event risk. A contract is only admitted to a company’s set when a traceable business reference can be cited; nothing is matched by statistical similarity, correlation, or thematic association. Each admitted edge is written to the graph with the citation recorded.
Traditional-derivative analog
Each Kalshi contract is also linked to the traditional financial instruments that have historically expressed the same risk: FX forwards and options for currency exposure, single-name equity options for company-specific event risk, sector ETF overlays for industry-wide moves, interest-rate swaps for duration, commodity futures for input cost. This mapping is curated against a fixed taxonomy of hedge classes rather than generated, and refreshes when the contract universe shifts. Each market on the site displays both its live event-contract quote and the legacy-derivative analog adjacent to it.
Generation discipline
Every passage on the artifact is composed against the validated graph, never against a blank page. A language model is invoked at three granularities per compile (per-issuer exposure syntheses, per-category rationales, per-edge notes) and at each granularity it receives a typed, validated record as structured input. The prompt operates as a style contract; the input operates as a fact contract. The model has no path to introduce an edge, claim an exposure that EDGAR does not support, or reference a market outside the selected set. A validation pass after generation rejects anything that drifts from either contract.
Compile-time, not request-time
Inference runs at compile time. Nothing runs at serve time. Once a compile finishes, the matching graph, the generated text, and every page derived from them are frozen and shipped as static HTML. The output is served from edge cache; no language model touches a request, no per-request inference cost is incurred, no client-exposed credentials exist, and no real-time fragility — Kalshi API latency, rate limits, model outages — can break a page in front of a reader. The “as of” timestamp on every page reflects the compile, not the request.
Theoretical frame
State-contingent claims
An event contract is a financial instrument that pays a fixed amount if a specific state of the world obtains and zero otherwise. This is the primitive Arrow (1964) introduced to formalize risk transfer under uncertainty: a contract indexed not to an asset’s price but to an outcome of the world. A market trading such contracts is quoting the price of a state. Under standard normalization, that price equals the probability the market assigns to the state.
Strips and completeness
Ross (1976) showed that a rich enough collection of state-claims spans the outcome space, so every payoff contingent on those states can be replicated as a portfolio of claims drawn from the strip. For variables that move continuously, Breeden and Litzenberger (1978) closed the loop: across a continuum of strikes, the slope of the call-price curve equals the digital price, and its curvature equals the implied probability density. A dense enough strip of event contracts does not merely host bets on individual outcomes; it quotes the full implied distribution of the underlying state.
The binary-option detour
The traditional instrument for a binary payoff is the binary option, replicated from continuous derivatives. Binary options trade over-the-counter, require dealer construction, and price the implied probability indirectly through a fitted volatility surface built from sparse vanilla strikes under model assumptions. They cover only a narrow set of underlyings and cannot natively express outcomes that lack a continuous strike axis.
The Vega Wedge
The cost gap between options replication and a direct event-contract purchase decomposes into three components: the Vega Wedge. The variance risk premium prices the entire volatility path, while a binary payoff depends only on the endpoint. The dealer balance sheet cost is the capital rented whenever a hedger trades against a constructed volatility surface. The replication friction is the cost of synthesizing a binary payoff from vanilla options: multiple legs, strike discreteness, oversized margin. Three structural costs that apply to options replication and do not arise in a direct digital purchase.
The direct expression
An exchange-listed event contract is a direct expression of the Arrow primitive. It pays one dollar if the outcome occurs and zero otherwise, and none of the three components of the Vega Wedge applies to it: variance risk premium does not apply because there is no path uncertainty, balance sheet rental does not apply because there is no intermediation, replication friction does not apply because nothing is being replicated. The quote equals the probability, with no implied surface and no model extraction in the way. The payoff is standardized and the contract is exchange-cleared, removing bilateral counterparty construction. The contract is indexed to the state itself rather than to the continuous price of an asset, which means it can carry state-contingent outcomes that traditional binary options cannot: corporate events, macro outcomes, regulatory decisions, policy thresholds.
The complete state-contingent machine
Three primitives together compose the complete state-contingent machine. The perpetual contract is the linear backbone, delta-one exposure with no strike. The event contract is the state-claim layer, a digital payoff indexed directly to outcomes. The volatility surface is the convex layer, continuous-payoff options whose curvature encodes the implied density. By Breeden-Litzenberger, any two of these primitives contain the third: a dense digital strip plus a linear instrument already implies the option surface. The state-contingent machine in its complete form is three primitives on one substrate, pricing the entire Arrow-Debreu state space directly.
What the map shows
Hedgebook surfaces the state-contingent claims available for every S&P 500 company. For each issuer, it records the event contracts whose underlying state references the company’s documented exposures, alongside the traditional financial instruments that have historically expressed similar risks. The map is descriptive: it records where a state-contingent expression is now available on a listed venue, and where the traditional derivative remains the only path. The contract universe is recomputed daily.
The dataset
The full public dataset behind the site — endpoints, schemas, download links, and a suggested citation — is documented on the data page.