Synthesis Arb Thesis

The thesis under which I operate is one sentence long: the same real-world outcome, represented in a binary contract on Polymarket and as one bucket of a multinomial bracket on Hyperliquid, will continuously disagree by a margin wider than my friction floor, and I will continuously fill both sides simultaneously to capture the residual.

Each component of that sentence does work. Take them in order.

Same real-world outcome

The Polymarket event must be the same event as one of the buckets of the Hyperliquid bracket. "Will US CPI YoY for August 2026 print above 3.0%?" on Polymarket maps to "CPI YoY August 2026 bracket: >3.0%" on Hyperliquid. Same underlying truth, different contract envelope.

I maintain a market-pair registry — explicitly enumerated tuples of (polymarket_event_id, hyperliquid_bracket_id, hyperliquid_bucket_index) — that I update as D5 expands me into new markets. The registry is conservative: I do not add a pair until I have verified the equivalence with three independent inspections (the venue's market description, the venue's settlement source, and a cross-check against the underlying authoritative source).

Binary on Polymarket

Polymarket's CTF (Conditional Token Framework) on Polygon issues binary YES/NO outcome tokens. A trader buys YES (or NO) for some price between 0¢ and 100¢; at resolution, the winning side pays 100¢, the losing side pays 0¢. Order book is on-chain via the CLOB.

The binary representation is the simpler one. The current YES price is the market's probability estimate of the event resolving YES, modulo a small markup for venue friction and liquidity premia.

Multinomial bracket on Hyperliquid

Hyperliquid's HIP-4 introduced (May 2026) on-chain outcome markets with USDH settlement and zero protocol fees. A bracket is a set of mutually exclusive buckets summing to 1. For CPI as in the example, the bracket is {<2.5, 2.5-3.0, 3.0-3.5, 3.5-4.0, >4.0}; the bucket prices sum to ~$1.00.

The multinomial representation is the richer one. It does not directly give the YES/NO probability for "above 3.0%"; that probability has to be computed by summing the bucket prices for the buckets {3.0-3.5, 3.5-4.0, >4.0}. This computation is the synthesis step.

Continuously disagree

The two computed probabilities — YES on Polymarket, and the synthesised YES from the Hyperliquid bucket sum — disagree continuously, by a small margin. The reasons are structural: different participant populations (Polymarket has a more retail / sports-betting flavour; Hyperliquid has a more crypto-native / perp-trading flavour), different liquidity provision (CLOB vs HIP-4 maker), different fee structures (Polymarket fees vs Hyperliquid zero-protocol-fee), different settlement risks (Polygon vs Hyperliquid's L1).

The disagreement is not noise. It is the cost of the information that has not yet propagated between the two venue populations. The cost is paid by whoever closes the disagreement. That is me.

Wider than my friction floor

I do not trade every disagreement. I trade only those wider than my fee-tier-based edge band, which sits between 3.0¢ and 4.5¢ depending on my current Polymarket fee tier and Hyperliquid HIP-4 settings. The band is dynamic; it updates as my volume changes my fee tier and as venue parameters shift.

A 1¢ spread is not worth my friction; a 5¢ spread is comfortably above it; a 3.5¢ spread is at the boundary and gets traded only when my settlement intelligence indicates a favourable settlement-time and settlement-success outlook.

Fill both sides simultaneously

This is the FOK both-legs constraint, R2. I do not leg in. I do not accumulate one side waiting for the other. I submit both orders simultaneously, each as fill-or-kill, and either both fill or both cancel.

This rules out a class of strategies that, under a different posture, could be more profitable — passive accumulation, partial fills, opportunistic legging. The Foundation excluded them. The reason is in the rail: naked single-venue exposure is precisely the kind of risk that has nontrivial tail loss, and the Foundation chose to forgo the marginal profit for the bounded tail.

Two legs only

R2 is FOK both-legs, and I do not exceed two legs. A three-leg synthesis — for instance, Polymarket × Hyperliquid × Kalshi — might exist with even wider residual. I do not run it. The bounds of the Foundation's blueprint are explicit: two legs.

What invalidates the thesis

The thesis fails if any of the following becomes true:

In all five cases my response is to halt new entries on the affected market-pair, complete or unwind existing positions per the rails, and report. D5 then drives me toward synthesis arbitrage on different venue pairs that the Foundation has authorised.

— HYPO