zerokn0wledge

Paradex architecture: RPI, zero-fee PFOF and unified portfolio margin

Quote512 words

Breaks down Paradex's design: flow segmentation via RPI, PFOF funding zero retail fees, STARK-proven settlement, universal portfolio margin via CairoVM, perpetual options, state-diff data posting and multi-stream revenue. Distribution against incumbents is the open test.

Key claims
Market makers pay 0.5-1 bps for PFOF access
Universal portfolio margin can cut margin 80-90% on basis trades
Roughly 10x gas cost reduction vs L1 via state diffs and EIP-4844 blobs
The Financial Supercenter of the Internet Economy. Built on Ethereum. Amid the CLOB wars, one key player I'm watching is @paradex. Instead of solely maximizing for speed and throughput, Paradex is on an ambitious mission to solve the fundamental market structure problem that plagues most CLOB DEXes. Let me break down their architecture and explain the USPs ↓ ➤ The "cancel-priority" mechanism that lets HFT market makers withdraw liquidity in the final milliseconds before adverse selection doesn't exist on Paradex. Instead, they've architected away the problem entirely through pre-trade flow segmentation on a ZK rollup built with Starknet's CairoVM. ➤ The hybrid architecture combines offchain matching engine for sub-100ms execution, with STARK-proven settlement (through @Starknet's SHARP) on Ethereum, achieving 7,000+ TPS while maintaining mathematical finality guarantees. ➤ RPI (Retail Price Improvement) creates parallel order books, one API-accessible for algos, another UI-only with tighter spreads from market makers who know they're protected from toxic flow, resulting in visibly "crossed" books where retail gets better prices than bots. ➤ Zero fees for retail funded through PFOF model where market makers pay 0.5-1 bps for access to segmented, non-toxic retail flow. This unique approach directly counters CEX economics which charge 6 bps to users to then kickback 80-90% to KOLs that control user flow. ➤ Universal portfolio margin across spot, perps, and options in a single account with real-time cross-margining calculations enabled by CairoVM's computational power, which can reduce margin requirements by 80-90% for basis trades. ➤ This is accompanied by serious product-level innovation, with perpetual options eliminating expiry management headaches by using continuous funding to represent theta decay, behaving like 5-day options but without rollovers (a.k.a. crypto's answer to the 0DTE option explosion in TradFi). ➤ Paradex only posts state diff instead of posting full calldata (possible thanks to @StarkWareLtd's STARK proofs) and uses EIP-4844 blobs for DA, achieving roughly 10x gas cost reduction compared to L1, a fundamental data efficiency advantage over other Ethereum rollups. ➤ Privacy is a fundamental feature and enabled through deliberate design choices. The stack includes execution privacy (no public mempool exposure) and position privacy through RPC masking, with cryptographic ZK privacy on the roadmap, truly critical for institutional adoption. ➤ Innovative tech is cool but destined to fail without a functioning business model on top. That's why Paradex's multi-pronged revenue model beyond just trading fees is another crucial USP: PFOF (~$56M projected FCF), vault management fees, stablecoin yield share, and platform fees from third-party apps building on their "SuperChain" a.k.a. DimeVM. ➤ Deep institutional DNA via incubation by @tradeparadigm (the liquidity network handling $1.2B daily options volume, not the VC) provides day-one sophisticated market maker relationships. The real test will be distribution against entrenched incumbents (CEXes but also player like Hyperliquid). However, eliminating the structural "execution tax" that retail pays everywhere else seems like a very compelling wedge to me if they can achieve escape velocity. Highly recommend to DYOR, and for more info, make sure to check out the full @a1research__ article linked below! Looking to try Paradex's game-changing product yourself? This is the way ↓ app.paradex.trade/r/a1res
QUOTING @a1research__
t.co/WhOnKTYb4m

2 images on X

Open on X

What came next · since 03 Oct 2025

zerokn0wledge avatar
Working together

Research, advisory and builds for deep-tech teams.