The House Edge

Appearance

1. What Is a Trade, On-Chain ​

Intent, settlement, finality — and where the money goes. ​

Cold open ​

The New York Stock Exchange moved from T+2 settlement to T+1 on 28 May 2024. It took ten years of negotiation and coordinated upgrades across broker-dealers, custodians, and the DTCC clearinghouse. The payoff was simple: a stock trade placed on Tuesday now becomes final on Wednesday instead of Thursday.[1]

That is a big institutional achievement. It is also a useful contrast. On the same Tuesday afternoon, Solana and Ethereum moved hundreds of thousands of trades from click to final record in seconds. Not because they are magic, and not because the comparison is one-for-one. They just do the work differently.

That work is the point here. A trade is not one moment. It is three: the trader states what they want, the venue executes it, and the result becomes final. On-chain trading changes who can see each step, who gets paid at each step, and how long the trader has to wait before the trade cannot be unwound.

What this chapter answers ​

Four questions matter here:

  • What is a trade, in plain English?
  • What is different when that trade happens on-chain instead of inside Coinbase?
  • When does the trader's order become visible to other people?
  • Who gets paid before the trader ends up with the asset they wanted?

The setup ​

Every trade has the same rough shape, whether it happens on Coinbase, on the floor of a traditional exchange, or inside a block.

First comes intent. The trader has decided what they want to do and has expressed it in a form the venue understands. On Coinbase, that is the moment she enters a quantity and clicks Buy. On Solana, it is the moment her wallet signs a transaction and sends it out. Intent is free. No money has moved yet.

Then comes settlement. This is when the trade actually executes. On Coinbase, the matching engine finds the other side of the order and updates the exchange's ledger. On Solana, a validator includes the transaction in a block and the chain records the swap. This is where the trader starts paying.

Then comes finality. This is the point at which the trade is treated as irreversible. Coinbase gives the user that feeling immediately: the account balance changes, and from the user's point of view the trade is done. The legal and cash movement behind Coinbase can still take longer. On-chain, finality depends on the chain. Solana finalises blocks in about 12.8 seconds. Ethereum L1 takes about 12.8 minutes. Hyperliquid is around 200 milliseconds for a co-located client. Ethereum L2s can feel instant while still waiting days for full settlement back to L1.

Same three steps. Different machinery. Different audience. Different bill.

Alice is the trader. She wants to swap $10,000 of USDC for SOL. She will do it twice on the same Tuesday afternoon: once on Coinbase Advanced Trade, and once on Solana through Jupiter. The question is not which screen looks cleaner. The question is what happens after she clicks.

The worked example ​

Alice opens Coinbase Advanced Trade and buys SOL with $10,000 of USDC. The order goes into Coinbase's central limit order book. Within milliseconds it matches against the best available ask, or against several asks if her order has to walk up the book. Her balance updates. She now holds about 65 SOL, assuming $153 per SOL, and Coinbase has deducted about $25 in fees. The screen says the trade succeeded. She closes the tab.

Then she opens her wallet, goes to Jupiter, and enters $10,000 USDC → SOL. Jupiter finds a route that splits the swap across three pools. Alice signs. The transaction goes out to Solana. About 0.6 seconds later, it appears in a block. About 12.8 seconds later, that block is final. Alice now holds about 65 SOL in her own wallet, minus roughly $50 of pool slippage and a small priority fee. This screen also says the trade succeeded.

Both screens told the truth. They also hid most of the machinery.

Now slow the film down.

What happens underneath ​

Intent ​

On Coinbase, Alice's intent travels one way: from her phone to Coinbase's matching engine. It is encrypted in transit. Coinbase sees it. Outsiders do not. By the time Coinbase publishes the trade to a data feed, the order has already filled.

On Solana, Alice's intent is a signed transaction. Her wallet sends it to an RPC endpoint, which is the server that helps ordinary users talk to the chain. Solana does not have one global public mempool like Ethereum. Instead, its pre-leader path, Gulf Stream, forwards the transaction toward the current and next slot leaders over the public internet.[2] That makes the window short, not private. Anyone running the right RPC setup, or paying someone who does, may be able to see Alice's transaction before it executes. The transaction should land in a block within about a second, and its blockhash expires after roughly a minute. Still, for that second, Alice's trade is visible.

On Ethereum, the same intent would usually land in the public mempool: the waiting room where pending transactions sit before a block builder includes them. Every Ethereum node keeps its own local pool and shares those pending transactions with peers. Anyone running a node can watch. Traders who do not want that exposure can use private routes such as Flashbots Protect, MEV-Blocker, or MEV-Share, which send the transaction straight to a builder. Chapter 3 handles those trade-offs. For now, the point is simple: Ethereum has a public waiting room, Solana has a shorter pre-leader visibility window, and Hyperliquid has no public pending state at all.[3]

Intent costs Alice nothing. Settlement is where the meter starts.

Settlement ​

On Coinbase, Alice's $10,000 order crosses the spread at the best available ask. It fills against another user's resting order, or against a market maker quoting on Coinbase. Her balance updates within milliseconds. Coinbase deducts a taker fee. For a customer of Alice's size, the base Advanced Trade taker fee is about twenty-five basis points, or roughly $25 on a $10,000 trade.[4]

That fee is the visible bill. Alice may still cross a spread, but she does not separately pay gas, a validator, or pool slippage. Coinbase matched the order inside its own book without showing the order to the outside world first.

One clarification matters for readers coming from US equities. Coinbase Advanced Trade does not, for spot retail orders, sell Alice's order to a wholesaler for payment for order flow. The order goes to Coinbase's exchange book. Coinbase makes money from explicit maker and taker fees, not from auctioning Alice's order to someone else.[5]

An aside on payment for order flow

In US equities and options, retail brokers such as Robinhood can be paid by wholesalers — Citadel Securities, Virtu Financial, Susquehanna's market-making arm, and a few others — to send customer orders to the wholesaler instead of to a public exchange. The wholesaler fills the order from its own inventory and pays for that privilege. The model has been legal in the US since the 1980s. Australia, Canada, Singapore, and the UK ban it; the European Union agreed in 2023 to phase it out by mid-2026.[6] In January 2025, the SEC's Division of Economic and Risk Analysis published a working paper finding that payment-for-order-flow rates per dollar traded in crypto were about forty-five times higher than in US equities, and about four and a half times higher than in US options.[7] The authors' explanation is blunt: crypto retail flow is less informed, so it is more valuable to internalise. The on-chain cousin of this model — exclusive order flow agreements between builders, validators, and searcher firms — returns in Chapter 7.

On Solana, settlement is also fast, but the bill is split among more people. Alice's signed transaction reaches the slot leader. The leader includes it about 0.6 seconds after she signed. Jupiter's route trades against three liquidity pools, and Alice's wallet ends up with about 65 SOL.

Her bill has four parts. There is a base network fee: 5,000 lamports per signature, or about $0.0008 at a $150 SOL price. There is a priority fee, paid to the validator for better scheduling inside the slot; call it about $0.30 for a clean major-pair swap in a quiet moment. There may be a Jito tip, if Alice or her router includes one. And there is pool slippage: the worse price Alice accepts because she is trading against automated liquidity pools rather than a tight order book. On a $10,000 USDC→SOL swap split across deep pools, that slippage is usually the biggest number, around $50 for this trade size and pair.[8]

So Alice paid Coinbase about $25 in explicit fees. On Solana, she paid about $50 in pool slippage plus a small amount in fees and tips. The totals are not worlds apart. The recipients are.

PhaseCoinbase Advanced TradeSolana via Jupiter
Intent$0$0
Settlement — taker fee / explicit~$25 (25 bps)~$0
Settlement — base network feenone~$0.001
Settlement — priority fee + tipnone~$0.30
Settlement — slippage$0 (filled at on-book best ask)~$50 (pool depth)
Finality — when Alice can treat it as doneInstant in-app; T+1 if external clearing is involved~0.6s confirmed; ~12.8s finalized
Total bill~$25~$50

Coinbase kept the Coinbase fee. On Solana, some of Alice's money went to the liquidity providers behind the pools she traded against — increasingly, in 2026, the prop-AMM operators discussed in Chapter 2 — and some went to the validator that included her transaction. Chapter 5 follows that validator money.

Finality ​

Finality is where the venues feel most different.

On Coinbase, Alice gets finality immediately as a user. The matching engine fills her order, her balance changes, and the trade is done from her point of view. Behind the screen, Coinbase still has books to reconcile. It aggregates trades, nets positions internally, and handles any external movement with the relevant counterparty. If the other side is another Coinbase user, that can all be internal. If an external bank or clearing process is involved, the legal and cash settlement may take until T+1. Alice does not see that. She sees SOL in her account.

On Solana, finality comes in steps. Processed means the slot leader has tentatively included Alice's transaction in a block, about 0.4 seconds after she signed. Confirmed means at least 66% of the network's staked SOL has voted on that block, about 0.6 seconds after signing. The block is unlikely to disappear, but it still could. Finalized means thirty-two slots have passed and the network treats the block as irreversible, about 12.8 seconds later.[9] Most wallets show Alice success at confirmed, not finalized. For her $10,000 swap, that is usually fine. For a large treasury transfer, the distinction matters.

Solana is trying to shorten that gap. Alpenglow, an upgrade live on a community test cluster as of mid-2026 with mainnet activation expected later in the year, targets about 150-millisecond block finality.[10] If it ships as planned, the distance between "your wallet says success" and "the network treats this as final" shrinks dramatically.

Ethereum L1 takes longer. Twelve-second slots and thirty-two-slot epochs put economic finality via Casper FFG at roughly 12.8 minutes after a transaction is included. Hyperliquid is much faster: its HyperBFT consensus reports about 200-millisecond median end-to-end latency for a co-located client, with the 99th percentile around 900 milliseconds.[11] Ethereum L2s sit in the awkward middle. A sequencer can tell the user "yes" quickly — roughly 250 milliseconds on Arbitrum, two seconds on Base — but optimistic rollups still have a seven-day fraud-proof window before withdrawals back to L1 are final.[12] A Base trade can feel done at once while the chain's full settlement process is still open.

Finality is not one number. It is the distance between "the app says done" and "no one can roll this back." Traders often ignore that distance. Venues do not.

How this plays out on each chain ​

Solana. The production chain has roughly 400-millisecond slots, confirmation around 0.6 seconds, and full finality around 12.8 seconds. Alpenglow aims to bring finality closer to 150 milliseconds.[10:1] Average transaction cost in early 2026 was about $0.017, up from roughly $0.004 in 2024 as priority fees became normal. Roughly 67% of user transactions now pay one.[13] Failed transactions peaked around 70% of non-vote transactions during the 2024 memecoin congestion, but most of that was not normal user activity. Helius attributed roughly 92% of failures to bots, with 95% of failures coming from 0.1% of addresses.[14] There is no global mempool. The visibility risk sits in the pre-leader RPC path.

Hyperliquid. HyperBFT produces about 200-millisecond median end-to-end latency for a co-located client. There is no public mempool. The order book lives inside consensus, and transactions are sequenced atomically. The base taker fee on perpetuals is 4.5 basis points; the top-tier taker fee is 2.4 basis points. Spot taker fees are 7 basis points.[15] This is closer to an exchange order book than to an AMM chain; Chapter 9 picks that apart.

Ethereum and its L2s. Ethereum L1 finality is about 12.8 minutes via Casper FFG. A typical Uniswap V3 swap on L1 cost roughly $0.25–$0.39 in May 2026 at a 0.4-gwei gas price. The May 2025 Pectra upgrade raised the gas limit to 60 million and kept block utilisation between 25% and 35%.[16] Major L2 swaps are cheaper: Arbitrum around $0.27, Optimism around $0.18, Base around $0.01–$0.05.[17] Soft confirmation is quick. Full optimistic-rollup settlement back to L1 still waits for the seven-day fraud-proof window.[12:1]

Who wins, who loses, why ​

The verdict here is deliberately plain.

Winners: traders who choose the right venue for the job. A retail buyer who wants a familiar interface and accepts a 25-basis-point explicit fee can be perfectly well served by Coinbase. A trader who wants self-custody and accepts pool slippage can be well served by Solana. An institutional trader who needs fast cancels and tight perpetuals markets may be better served by Hyperliquid. The venues win too. They collect their fees either way.

Losers: traders who choose badly. A small trade on Ethereum L1 during a high-gas moment can spend more on fees than the trade justifies. A large trade routed through a thin Solana pool can lose more to slippage than it would have paid on a deep order book. A trader who waits for full Solana finality when confirmed would have been enough loses time for no benefit.

Is this bad? Not by itself. Different routes have different prices, speeds, and failure modes. The trader's job is to pick the route that fits the trade.

The harder questions start when one actor gets a better view of Alice's trade than everyone else. Who gets that view, how they use it, and what it costs Alice is the subject of the next several chapters.

What changes when… ​

What changes when Alice's intent is visible before settlement — when anyone running an Ethereum node, or paying for the right Solana RPC feed, can see what she is about to do?

They can do quite a lot. Chapter 3 is about what they do first.

Footnotes and sources ​

  1. DTCC, Shortening the US Equities Settlement Cycle, https://www.dtcc.com/ust1; SIFMA, T+1 After Action Report (mid-2024), https://www.sifma.org/resources/guides-playbooks/t1-after-action-report. The transition affected DTCC-eligible securities — equities, corporate and municipal bonds, unit investment trusts — and replaced T+2, which had itself replaced T+3 in 2017. Accessed 2026-05-13. ↩︎

  2. Helius Research, Solana's Gulf Stream: Mo Mempool, Mo Problems, https://www.helius.dev/blog/solana-gulf-stream. Accessed 2026-05-13. Solana's leader schedule is published per epoch (approximately every two days); blockhashes expire after approximately 150 slots, or roughly sixty seconds at the current slot time. ↩︎

  3. Hyperliquid Documentation, HyperCore Overview, https://hyperliquid.gitbook.io/hyperliquid-docs/hypercore/overview. The chain matches orders inside HyperBFT consensus and exposes no public pending state. Accessed 2026-05-13. ↩︎

  4. Coinbase Help, Coinbase Advanced Trade — Pricing and Fees, https://help.coinbase.com/en/coinbase/trading-and-funding/pricing-and-fees/fees. Taker fees on Advanced Trade vary by trader volume; the 25 basis-points figure approximates a retail trader's base-tier taker fee. Accessed 2026-05-13. ↩︎

  5. Coinbase Help, Order Matching, https://help.coinbase.com/en/derivatives/features/order-matching. The matching engine is described as anonymous FIFO with price-time priority on a central limit order book. Coinbase's revenue model is explicit: taker and maker fees on Advanced Trade, plus a spread-and-flat-fee model on the consumer-facing simple app. Accessed 2026-05-13. ↩︎

  6. Thomas J. Boulton, Thomas D. Shohfi, and Michael Walz, How Does Payment for Order Flow Influence Markets? Evidence from Robinhood Crypto Token Introductions, U.S. Securities and Exchange Commission Division of Economic and Risk Analysis Working Paper (January 2025), https://www.sec.gov/files/dera_wp_payment-order-flow-2501.pdf. Section 2.2 documents the regulatory history; jurisdictions naming a ban include Australia, Canada, Singapore, the UK, and the EU (mid-2026 effective date). Accessed 2026-05-13. ↩︎

  7. Boulton, Shohfi, and Walz (2025), op. cit., Abstract and Section 5. The 45×-equities and 4.5×-options figures are per dollar of trading value. The authors estimate the daily trading-cost impact of crypto PFOF on non-Robinhood venues at approximately $4.8 million across affected tokens. Section 2.2 of the paper documents the concentration of US wholesalers: three firms account for 70–82% of equity PFOF and 73–90% of options PFOF; Citadel Securities and Virtu Financial alone account for 60–70% of equity wholesale flow between 2017 and 2021, citing Hu and Murphy (2024) and Bryzgalova, Pavlova and Sikorskaya (2023). Accessed 2026-05-13. ↩︎

  8. Solana Compass, Fees + Burn Tracker, https://solanacompass.com/statistics/fees; Helius Research, Priority Fees: Understanding Solana's Transaction Fee Mechanics, https://www.helius.dev/blog/priority-fees-understanding-solanas-transaction-fee-mechanics. Pool-slippage figures for a $10K major-pair swap are illustrative, derived from typical Jupiter routing across deep Solana liquidity venues. Accessed 2026-05-13. ↩︎

  9. Helius Research, What are Solana Commitment Levels?, https://www.helius.dev/blog/solana-commitment-levels. Three levels: processed (~0.4s, leader has included tentatively), confirmed (~0.6s, ≥66% stake-weighted vote), finalized (~12.8s, 32 slots elapsed). Accessed 2026-05-13. ↩︎

  10. CoinDesk, The Protocol: Solana's 'Alpenglow' upgrade is live for testing, 13 May 2026, https://www.coindesk.com/tech/2026/05/13/the-protocol-solana-s-alpenglow-upgrade-is-live-for-testing. Mainnet activation is expected late Q3 / early Q4 2026 with Agave 4.1; target finality is approximately 150 milliseconds (potentially as low as 100ms). The chapter cites the current production figure (12.8 seconds) as the authoritative number while flagging Alpenglow as shipping. Accessed 2026-05-13. ↩︎ ↩︎

  11. Hyperliquid Documentation, HyperCore Overview, op. cit. End-to-end latency for a geographically co-located client is 200 milliseconds at the median and 900 milliseconds at the 99th percentile. Accessed 2026-05-13. ↩︎

  12. Jump Crypto, Bridging and Finality: Optimism and Arbitrum, https://jumpcrypto.com/resources/bridging-and-finality-optimism-and-arbitrum; Arbitrum Documentation, The Sequencer and Censorship Resistance, https://docs.arbitrum.io/how-arbitrum-works/deep-dives/sequencer. Native USDC via Circle's CCTP completes in 15–25 minutes; third-party liquidity bridges (Across, Hop) complete in under 5 minutes for 0.05–0.15% in fees, but canonical L1 settlement for optimistic rollups requires the full seven-day fraud-proof window. Accessed 2026-05-13. ↩︎ ↩︎

  13. Solana Compass, op. cit.; Helius Research, Priority Fees, op. cit. The $0.017 average is for early 2026; the $0.004 figure is the 2024 baseline before priority-fee usage normalised. Accessed 2026-05-13. ↩︎

  14. Helius Research, Solana Ecosystem Report H1 2025, https://www.helius.dev/blog/solana-ecosystem-report-h1-2025; Why Does My Transaction Fail? (arXiv:2504.18055), https://arxiv.org/html/2504.18055. The 70% peak is a 2024 figure during memecoin congestion; the 92%-bot and 95%-from-0.1%-of-addresses framings are Helius's. Accessed 2026-05-13. ↩︎

  15. Hyperliquid Documentation, Fees, https://hyperliquid.gitbook.io/hyperliquid-docs/trading/fees. Base-tier perpetuals: maker 1.5 bps, taker 4.5 bps. Top-tier (highest-volume) perpetuals: maker 0 bps, taker 2.4 bps. Spot: maker 4.0 bps, taker 7.0 bps. Accessed 2026-05-13. ↩︎

  16. Etherscan Gas Tracker, https://etherscan.io/gastracker; SQ Magazine, Ethereum Gas Fees Statistics 2026, https://sqmagazine.co.uk/ethereum-gas-fees-statistics/. Pectra (May 2025) raised the gas limit to 60 million; block utilisation in early 2026 was 25–35%, materially reduced from the 2024 peak. Accessed 2026-05-13. ↩︎

  17. L2Fees.info, https://l2fees.info/, retrieved 2026-05-13. ↩︎