The Cat-and-Mouse game of Market Making w/ Rahul Jain

Rahul Jain is head of trading at Ellipsis Labs, where he moved after running a crypto options desk at a Chicago prop firm that held roughly 20 to 25 percent market share across all crypto options exchanges on Deribit. At Ellipsis he progressed through Phoenix spot order book, prop AMM, and now perpetuals trading, giving him a ground-level view of how each model handles the core problem of market maker exposure to stale quotes.

The structural weakness of order book market making on Solana is that validators can be tipped to land transactions ahead of cancellations. If a market maker has a resting bid at 100 and price drops to 90, they either pay to cancel or get picked off. Sophisticated firms arrived roughly a year after Phoenix spot launched and began exploiting this to the tune of tens of thousands of dollars per day in losses for market makers. The prop AMM model addresses this by placing the market maker at the application layer, giving them control over matching logic within the transaction itself. This enables transaction introspection, meaning the market maker can inspect an incoming transaction and apply differentiated pricing based on whether the counterparty appears toxic or retail. Jain identifies this as the key innovation Ellipsis added over Lefinity, an earlier prop AMM that used oracle-based quoting but could not distinguish counterparty types.

The cat-and-mouse dynamic between market makers and toxic flow actors intensified sharply over 2024 and into 2025. In 2024, toxic actors used a single public key, making identification straightforward. Jain's response was to block that key roughly 83 percent of the time rather than outright, deliberately confusing the counterparty to slow their adaptation. Toxic actors eventually rotated public keys, escalating the arms race. By 2025 Jain was spending approximately six hours a day scrolling Solscan to identify adversarial patterns across combinations of five transaction properties. Toxic actors eventually learned to replicate the exact on-chain appearance of retail UI transactions by invoking smart contracts directly without using the front end. Routers responded by issuing signed retail flags on transactions, allowing prop AMMs to identify confirmed non-toxic flow and price it very tightly, making the prop AMM function similarly to an RFQ system. The key trade-off is that prop AMMs guarantee execution but may deliver negative slippage, while RFQ guarantees the quoted price but only fills around 90 to 95 percent of the time.

Jain argues that the better effective price available on-chain for SOL-USDC compared to Binance exists primarily because Binance charges an exchange fee on top of the market maker spread, whereas on Solana no such fee exists. The underlying market maker spread is probably similar on both venues.

Ellipsis moved into perpetuals with Phoenix Perps after being reluctant to do so two to three years earlier, with the core hesitation being the difficulty of guaranteeing that oracle and market maker transactions land in the correct order, a problem that becomes especially dangerous with leverage. Jain identifies oracle attacks as the primary MEV threat to perpetual exchanges, where a bad actor controlling enough block slots can manipulate the mark price and force liquidations into their own orders. The Jelly incident on Hyperliquid illustrates the related risk of listing perps on assets whose spot price can be manipulated due to low liquidity. Defenses include small open interest caps, EMA-based pricing, isolated markets, and aggressive auto-deleveraging, which caps exchange losses and puts users in a PvP environment against each other.

Jain views Multiple Concurrent Proposers as by far the most important architectural improvement needed on Solana, stating nothing else comes close. Without MCP, a single block leader can toxically order an entire block and extract maximum value from market makers every slot. MCP merges proposals from multiple concurrent proposers, making it much harder to censor specific updates or control full block ordering. MCP also enables Application Controlled Execution, which would allow a program like Phoenix Perps to dictate that oracle updates always land first and taker trades land last, directly reducing exchange risk. He believes MCP could enable scaling TVL and volume by an order of magnitude but acknowledges it faces serious incentive incompatibility blockers, with existing projects that built businesses under current rules having little reason to support the change, and a decentralized blockchain having no mechanism to force consensus.

Jain sees Solana at a fork where it either becomes a serious trading venue or gets relegated to being a meme coin chain, and argues the worst position for any chain is being in the middle and not the best at anything. On market structure, he notes that perps in crypto is probably winner-takes-most but less concentrated than traditional finance, where splits run 80-20 or 90-10, while in crypto the equivalent might be closer to 55-45 or 60-40 due to community loyalty and the absence of any legal obligation to route to the venue with best execution.