The House Edge

Appearance

Chapter 1 — Research Note (Phase 1) ​

Status: RESEARCH (Phase 1 — written by agent, awaiting Nick's review) Date: 2026-05-13 Working chapter: 01 — What Is a Trade, On-Chain Carries forward from: SPEC.md (strawman drafted by agent; Nick edited or accepted before this Phase 1 ran)

How to read this note ​

This is a setup chapter, so the research is lighter than Chapters 2 or 4. Three things to know up front:

  1. Two corrections to previously-drafted chapters surfaced in this research. Chapter 2 footnote 16 cites a 3.2 bps Hyperliquid taker fee; the current docs say 4.5 bps base (0% top-tier). Chapter 2 §5b also describes Hyperliquid as having "approximately seventy-millisecond" finality; the docs themselves say 200ms median end-to-end, 900ms p99. The "70ms" number circulates widely but isn't in the canonical source. I'll flag both for a Ch 2 patch in this chapter's REVIEW_NOTES.md.
  2. Alpenglow is live on testnet as of today. A CoinDesk piece dated 2026-05-13 reports Solana's Alpenglow upgrade went live on a community test cluster. Mainnet activation is expected late Q3 / early Q4 2026; the target finality is ~150ms (vs the current production 12.8s). Whether this changes Chapter 1's framing depends on how the chapter wants to handle "things that are about to change."
  3. The chapter is structural, not adversarial. Most of the research below is foundational mechanics that has not changed in years. The 2026-vintage numbers are for finality times, gas costs, and failed-transaction rates.

1. Key claims ​

Numbered claims the chapter is allowed to state. Sources cited inline.

The three phases

  1. Every trade — on a centralised exchange, in TradFi, or on-chain — passes through three phases: intent (the trader's stated desire to trade), settlement (the moment the trade is matched and executed at the venue), and finality (the moment the trade is irrevocable). The mechanics of each phase differ between venues; the existence of all three phases does not. There is no widely-used primary source that lays this out as a unifying frame — the chapter's three-phase structure is the chapter's own scaffolding.

  2. In US equities, settlement happens on T+1 — the day after the trade — and means a cash-and-asset transfer at a clearinghouse (DTCC/NSCC), distinct from the trade match itself. The T+1 cycle was adopted on 28 May 2024, replacing T+2. (DTCC — Shortening the US Equities Settlement Cycle; SIFMA — T+1 After Action Report, 2024)

  3. In crypto, "settlement" is most often used colloquially to mean inclusion in a block, and "finality" is the chain's guarantee that the included transaction cannot be reversed. Ethereum's documentation defines finality as "the guarantee that a set of transactions cannot be changed without a huge amount of ETH being lost"; the term "settlement" has no formal definition in the Ethereum glossary. The chapter will define both terms precisely on first use and note the TradFi-vs-crypto definitional drift. (Ethereum Foundation — Glossary)

Finality by chain in 2026

  1. Solana has three commitment levels with distinct wall-clock latencies: processed (~0.4 seconds; the leader has tentatively included the transaction), confirmed (~0.6 seconds; ≥66% of stake has voted), and finalized (~12.8 seconds; 32 slots have passed, the block is irrevocable). The 12.8-second number is the "fully final" production figure as of mid-2026. (Helius — What are Solana Commitment Levels?)

  2. Alpenglow, Solana's planned consensus upgrade, went live on a community test cluster on 2026-05-13. Its target is ~150-millisecond finality (potentially as low as 100ms). Mainnet activation is expected in late Q3 or early Q4 2026 with Agave 4.1. Until it ships, 12.8 seconds remains the production number. (CoinDesk — The Protocol: Solana's 'Alpenglow' upgrade is live for testing, 2026-05-13; Solana Compass — Alpenglow: Solana's largest protocol upgrade ever)

  3. Hyperliquid's HyperBFT consensus produces ~200-millisecond median end-to-end latency (900ms p99) for an order placed by a geographically co-located client. The "~70ms" finality number circulating in secondary commentary refers to block finality specifically, not end-to-end order latency; the canonical docs cite the end-to-end figure. (Hyperliquid Documentation — HyperCore Overview)

  4. Ethereum L1 uses 12-second slots and 32-slot epochs. Economic finality via Casper FFG takes two epochs after a transaction is included — approximately 12.8 minutes. (Ethereum Foundation — Proof-of-stake)

  5. Major Ethereum L2s have a two-tier finality story. Sequencer soft confirmation is fast — approximately 250 milliseconds on Arbitrum, 2 seconds on Base. But canonical settlement back to L1 is slow: batch posting takes minutes, and for optimistic rollups the fraud-proof window before withdrawals finalize on L1 is seven days. Native bridges via CCTP complete in 15–25 minutes; third-party liquidity bridges (Across, Hop) complete in under 5 minutes for 0.05–0.15% in fees. (Jump Crypto — Bridging and Finality: Optimism and Arbitrum; Arbitrum Documentation — The Sequencer and Censorship Resistance)

Costs in 2026

  1. Solana average per-transaction cost in early 2026 was approximately $0.017, up from roughly $0.0038 in 2024 and early 2025. The base fee remains fixed at 5,000 lamports per signature (~$0.00075 at $150 SOL); the increase comes from priority fees. Approximately 67% of user transactions now pay a priority fee. For competitive swaps (memecoin pairs, sandwich-vulnerable trades), priority fees of 0.001–0.005 SOL (~$0.15–$0.75) are typical; Jito tips are paid separately for bundle inclusion. (Solana Compass — Fees + Burn Tracker; Helius — Priority Fees: Understanding Solana's Transaction Fee Mechanics)

  2. Ethereum L1 gas was approximately 0.394 gwei as of 2026-05-13, putting a typical Uniswap swap (150,000–300,000 gas) at $0.25–$0.39 in fees. Block utilisation is 25–35% post-Pectra with the gas-limit raised to 60M, which has materially reduced fees from the 2024 peak. (Etherscan Gas Tracker; SQ Magazine — Ethereum Gas Fees Statistics 2026)

  3. L2 swap costs in 2026 are far below Ethereum L1: Arbitrum ~$0.27, Optimism ~$0.18, Base ~$0.01–$0.05. (L2Fees.info, retrieved 2026-05-13)

  4. Hyperliquid taker fee is 0.045% (4.5 basis points) on perpetuals at the base tier, scaling down to 0.024% (2.4 bps) at the top volume tier. Maker fees are 0.015% (1.5 bps) at base, 0% at top. Spot fees are higher: maker 0.040%, taker 0.070%. (Hyperliquid Documentation — Fees)

Coinbase, mempools, failures

  1. Coinbase Advanced Trade (formerly Coinbase Pro) operates an anonymous FIFO matching engine with price-time priority on a central limit order book. Coinbase does not use a payment-for-order-flow model for retail spot trading — its revenue comes from explicit taker/maker fees on Advanced Trade and a combined spread-plus-flat-fee model (~0.5% spread plus a flat fee) on the consumer-facing app. (Coinbase Help — Order Matching; Coinbase Help — Pricing and Fees Disclosures)

Payment for order flow — TradFi mechanism, crypto application

14a. Payment for order flow (PFOF) is the practice in which a retail broker is paid by a wholesaler in exchange for routing customer orders to that wholesaler, who then internalizes the orders by trading against its own inventory. The model has been legal in the US since the 1980s; it has been banned in Australia, Canada, Singapore, and the UK; and the EU agreed in June 2023 to phase it out by mid-2026. (Boulton, Shohfi & Walz — How Does Payment for Order Flow Influence Markets? Evidence from Robinhood Crypto Token Introductions, SEC DERA Working Paper, January 2025)

14b. PFOF in US equities is highly concentrated. Three wholesalers 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 order flow between 2017 and 2021. (Same source, citing Bryzgalova/Pavlova/Sikorskaya 2023 and Hu/Murphy 2024.)

14c. Crypto PFOF rates per dollar of trading value are approximately 45× higher than US equities and 4.5× higher than US options. The DERA authors attribute the gap to the fact that crypto retail order flow is more uninformed than equity order flow, which makes it more valuable to internalize. The paper's empirical work — using Robinhood Crypto token-listing events as a PFOF shock — finds that after PFOF is introduced for a given crypto asset, trading on non-Robinhood venues sees seller-driven order imbalances, larger average trade sizes, wider implied spreads, and greater volatility, with estimated daily trading-cost impact of approximately $4.8 million across affected tokens. (Same source, abstract and §4.)

14d. The chapter's use of PFOF. Chapter 1 uses the DERA paper as the canonical primary source on what PFOF is and what it does to market quality. The on-chain analogue — exclusive order flow agreements between block builders, validators, and specific searchers — is the subject of Chapter 7. The chapter explicitly plants the TradFi-to-on-chain parallel without developing it.

  1. Ethereum has a public mempool: each node maintains a local pool of pending transactions and gossips them to peers across the p2p network. Anyone running a node — or paying for an RPC provider that exposes pending data — can see every public pending transaction before it is included in a block. This visibility is the substrate for sandwich attacks and the reason private-routing services (Flashbots Protect, MEV-Blocker) exist. (Blocknative — What is the Mempool?; QuickNode — How to Access Ethereum Mempool)

  2. Solana has no global mempool. Under Gulf Stream, RPC nodes forward transactions directly to the current and next slot leaders' TPU via QUIC; the leader schedule is published per epoch (~2 days in advance). Blockhashes expire after approximately 150 slots (~60 seconds), so unprocessed transactions cannot linger. The functional consequence is similar — bots still race for inclusion — but there is no shared gossiped pool that every observer reads. (Helius — Solana's Gulf Stream: Mo Mempool, Mo Problems)

  3. Hyperliquid has no public mempool at all. HyperBFT consensus sequences transactions immediately at the consensus layer, with semantically-aware ordering within a block (non-order actions first, then cancels, then GTC/IOC sends). This is the structural reason there is "nothing to frontrun" on Hyperliquid in the Ethereum sense. (Hyperliquid Documentation — HyperCore Overview)

  4. Solana failed-transaction rates peaked at approximately 70–75% of non-vote transactions during 2024 memecoin congestion, but the framing matters: Helius's analysis attributed roughly 92% of failed transactions to bot-initiated activity, with 95% of failures coming from 0.1% of addresses. The bot failure rate was approximately 58%; the typical real-user swap failed at a materially lower rate. The chapter should be careful with this number — quoting the 70% headline without the bot caveat would mislead a business reader. (Helius — Solana Ecosystem Report H1 2025; academic study — Why Does My Transaction Fail?)

2. Numbers to verify ​

#NumberSourceDateFlag
N1Solana commitment latencies: processed ~0.4s, confirmed ~0.6s, finalized ~12.8sHelius commitment levels2024–currentFoundational mechanics; pre-Alpenglow
N2Alpenglow target finality ~150ms (potentially 100ms); live on testnet 2026-05-13, mainnet expected late Q3 / early Q4 2026CoinDesk 2026-05-132026-05-13Live news; the chapter will frame as "shipping" not "shipped"
N3Hyperliquid end-to-end latency: 200ms median, 900ms p99 (geographically co-located client)Hyperliquid HyperCore docs2026Corrects "70ms" framing used in Ch 2 — flag for back-edit
N4Ethereum L1: 12-second slots, 32-slot epochs, ~12.8 minutes to Casper FFG economic finalityEthereum.org PoS docscurrentStable; pre-Pectra mechanics unchanged
N5Arbitrum sequencer soft confirmation ~250ms; Base block time ~2s; 7-day fraud-proof window before withdrawals finalize on L1Jump Crypto; Arbitrum docs2024–2026Stable
N6Solana avg tx cost early 2026: ~$0.017 (vs ~$0.0038 in 2024); base fee 5,000 lamports/signature; ~67% of user txns pay priority feeSolana Compass; Helius priority feesearly 2026Strong primary
N7Ethereum L1 gas ~0.394 gwei (2026-05-13); typical Uniswap swap $0.25–$0.39; block utilisation 25–35% post-Pectra; gas limit 60MEtherscan; SQ Magazine2026-05-13Real-time; will drift
N8L2 swap costs: Arbitrum ~$0.27, Optimism ~$0.18, Base ~$0.01–$0.05L2Fees.info2026-05-13 liveAggregator with primary on-chain data
N9Hyperliquid taker fee 0.045% (4.5 bps) base perpetuals; 0.024% top-tier; spot taker 0.070%Hyperliquid Fees docs2026Corrects "3.2 bps" used in Ch 2 footnote 16 — flag for back-edit
N10Coinbase: anonymous FIFO matching, price-time priority CLOB; no PFOF for retail spotCoinbase Order Matching; Coinbase PricingcurrentCounterweight to common assumption that all retail brokers run PFOF
N11US equities settlement: T+1 since 2024-05-28DTCC; SIFMA After Action Reportmid-2024Foundational TradFi parallel
N12Ethereum public mempool: each node has local pool, gossiped p2p; anyone running a node sees pending txnsBlocknative; QuickNodecurrentFoundational
N13Solana Gulf Stream: no global mempool; RPCs forward to slot leader TPU via QUIC; blockhash expires ~150 slots (~60s)Helius Gulf StreamcurrentFoundational
N14Solana failed-tx peak ~70-75% in 2024 congestion; ~92% bot-initiated; 95% from 0.1% of addresses; typical bot failure rate ~58%Helius H1 2025 Ecosystem; arXiv "Why Does My Transaction Fail?"2024–2025Care with framing — the headline number misleads without the bot caveat

Numbers I deliberately did not pin: a clean Ethereum-wide failed-tx rate for 2026 (no canonical aggregate found; Dune dashboards suggest ~2–5% reverts but no single primary cite). A "best published CEX-vs-on-chain comparison piece" (couldn't find one; the chapter will compose the comparison from first principles).

3. Contested or evolving claims ​

  • Alpenglow's shipping date. Mainnet activation is expected late Q3 / early Q4 2026 but is not committed. The CoinDesk piece is dated today (2026-05-13) and references testnet activation. The chapter should frame Alpenglow as "shipping, on the testnet today, expected on mainnet later this year" rather than as a fait accompli.
  • Hyperliquid "finality" framing. The "~70ms" number circulates widely in ecosystem commentary but is not in Hyperliquid's own docs. The docs cite 200ms median end-to-end / 900ms p99. The chapter should use the docs number. Ch 2 §5b currently uses "approximately seventy-millisecond" — this is a back-edit item.
  • "Settlement" as a term. TradFi (T+1, central clearinghouse) and crypto (inclusion-in-block or post-FFG finality) use the word differently. The chapter will define both meanings on first use and stick to "finality" for the crypto-side hard guarantee.
  • The 70% Solana failed-tx headline. Quoted without the bot caveat (92% bots, 95% from 0.1% of addresses), it misleads. The chapter will quote the headline once with the caveat right next to it.

4. Characters introduced ​

This chapter is light on actors. What appears:

The trader (Alice). Returns from the prologue and Chapter 4. The chapter uses her name where the prologue used "you." She is making a $10,000 USDC→SOL swap; the chapter follows her through both Coinbase and Solana to show the three phases playing out on both venues. No new "Meet the actor" sidebar.

The validator / sequencer (cameo). One paragraph. The actor who decides which transactions go into the next block and in what order. Full sidebar in Chapter 5.

The wallet (concept, not actor). One inline definition. The user-facing software that holds private keys and signs transactions.

The mempool (concept, not actor). One paragraph cameo. The public queue of pending transactions on chains that have one. Full treatment in Chapter 3.

Coinbase (named, as the CEX comparison anchor). One paragraph: how its matching engine works, why retail trades fill differently than on-chain, where its revenue actually comes from.

5. Worked example candidates ​

The SPEC offered three options. I'll formalize the recommendation here.

Candidate A — Alice's $10,000 USDC→SOL swap on Solana, broken into the three phases ​

Alice opens her wallet, clicks "Swap $10,000 USDC for SOL" on Jupiter, and signs. The chapter walks the three phases on-chain only: she has issued intent; her transaction is now in motion through Solana's TPU forward path; about 0.6 seconds later it has been included in a block (settlement in the crypto sense); about 12 seconds after that, the block is irrevocable (finality).

  • Pros: continuity with the prologue and Chapter 4; the dollar number is already in the reader's head; clean three-phase narrative.
  • Cons: forces a Solana-first frame; the chapter is supposed to be chain-agnostic and to set up the comparison with CEX trading.

Candidate B — The same trade on Coinbase, contrasted ​

Alice opens Coinbase Advanced Trade, clicks "Buy $10,000 of SOL," and watches her order fill against the on-exchange CLOB in milliseconds. The chapter walks the same three phases: intent (her click), settlement (Coinbase's matching engine fills her order against another user's resting order), finality (instant, by Coinbase's accounting; the actual cash-and-asset transfer to Coinbase's master account on the bank side is on T+1 if Coinbase chose to settle externally, but internally it's settled the moment the match happens).

  • Pros: gives the reader the TradFi-fluent control comparison; demonstrates that finality on a CEX is instant, for the user — even if the legal mechanism is different.
  • Cons: an all-CEX example would feel off-topic for a chapter titled "What Is a Trade, On-Chain."

Candidate C — Both side-by-side ​

The chapter runs both A and B in parallel, phase-by-phase. Three sections inside §5 ("The mechanics, in detail"): intent on both sides; settlement on both sides; finality on both sides. The reader sees what differs (visibility, gas, finality timeline) and what doesn't (the three phases all happen in both; the trader is still a trader; the venue still has to match and settle).

  • Pros: gives the reader the comparison the chapter's argument depends on; explicitly demonstrates that "on-chain" is not a different kind of trade, just a different implementation.
  • Cons: more complex to thread.

Agent's recommendation ​

Candidate C (side-by-side). The chapter's job is to install the three-phase vocabulary as universal and then show what changes between CEX and on-chain. Side-by-side does both in one structure. Candidate A becomes the natural anchor for the on-chain side; Candidate B is the CEX track.

6. Open questions for Nick ​

Q1 — Treatment of Alpenglow. Alpenglow is live on Solana testnet today (2026-05-13). The chapter could (a) cite the 12.8-second production number with a "shipping faster" sentence, (b) lead with the 150ms target and treat 12.8s as legacy, or (c) skip Alpenglow entirely as too newsy for a setup chapter. My recommendation: (a) — current production numbers with a clean forward-looking sentence. The Bible says "Not breaking news. If something happened last week, mention it only if it's load-bearing." Alpenglow at the time of writing is breaking news; the production number is what reads correctly a year from now.

Q2 — Ch 2 back-edits. Two corrections surfaced from this research:

  • Ch 2 footnote 16 cites 3.2 bps Hyperliquid taker fee. Actual base is 4.5 bps.
  • Ch 2 §5b describes Hyperliquid as having "approximately seventy-millisecond intervals" finality. The docs say 200ms median end-to-end / 900ms p99. Both are easy fixes. Do you want me to patch Ch 2 as part of Phase 3 here, or hold for a dedicated back-pass?

Q3 — Worked example confirmation. Candidate C (side-by-side) is the recommendation. Confirm or override.

Q4 — How much TradFi parallel. The chapter leans heavily on Coinbase / NYSE / DTCC analogues. Is the Bible's "TradFi parallels are your best friend" guidance still the right calibration here, or does the chapter need to ease off because the reader is now (post-Chapters 2 and 4) less of a beginner than the Book Bible assumed when it was written?

Q5 — Failed-tx rate framing. The 70% Solana headline number is dramatic but misleading without the "92% bot-initiated" caveat. My plan is to quote the headline once with the caveat right next to it. Alternative: drop the headline entirely and quote only the bot-vs-real-user breakdown. Preference?

Q6 — Coinbase as the CEX comparison anchor. The chapter uses Coinbase because the reader is most likely to know it. Alternatives: Binance (largest, but less familiar to a US business reader), Robinhood (familiar but stock-only — its crypto product is small), Kraken (less familiar). Confirm Coinbase or override.

Q7 — Chapter length. Ch 4 was 4,892 words; Ch 2 was 5,053 words. Ch 1 is structurally lighter and could probably land at 3,500–4,000 words. The chapter spec template says 4,000–6,000. Should I aim for the floor or come in below it deliberately?

Sources cited ​

Primary docs and research:

Academic:

Aggregators / news (cited because primary not directly accessible):

Phase 1 is complete. Per the user's pattern of compressed reviews, Phase 2 (OUTLINE.md) will follow immediately unless Nick flags an issue.