The House Edge

Appearance

Chapter 4 — Review Notes (Phase 3 self-check) ​

Status: REVIEW NOTES (Phase 3 — written by agent, accompanying DRAFT.md) Date: 2026-05-13 (revised same day for 2026-data refresh — see "Revision log" below) Working chapter: 04 — The SearcherWord count: ≈4,400 words including footnotes (≈3,950 words of prose, within the Book Bible's 4,000–6,000 target).

Revision log ​

2026-05-13 (rev 1) — Nick flagged two issues on the first draft:

  1. The HFT/NASDAQ analogy sentence in §3 was unclear ("in mechanism if not in regulatory regime…"). Rewritten as: "A high-frequency arbitrage desk at Citadel Securities or Jane Street earns its living the same way on NASDAQ: it reads the order flow, prices the spread, and pays to be first. The mechanism is the same on-chain; the rails and the regulators are not."
  2. The Winners paragraph in §7 leaned on 2024–early-2025 data without acknowledging it. I ran a focused 2026-data hunt and found four load-bearing new sources that did not exist when the first research pass ran:
    • The DeFi Report / PANews, 20 April 2026: Q1 2026 Solana network REV $89.9M, down 68% YoY; Jito tips down 72% YoY. → §5d, §6 Solana paragraph, §7 Winners.
    • arXiv 2604.00234 (Wang, Saraf, Heimbach, Babel, Zhang), 31 March 2026: direct academic follow-up to Flashbots' June 2025 "Limits of Scaling" with Q1 2026 Base/Arbitrum spam data; finds that gas-price floors produce a one-time step-down rather than a durable reduction (Arbitrum's Jan 2026 fee increase saw spam surge above pre-change levels by mid-Feb 2026). → §6 Ethereum/L2 paragraph.
    • relayscan.io Q1 2026 builder snapshots: Titan ~52% of all Ethereum blocks. → §6 Ethereum/L2, §7 Winners.
    • Helius "Constellation" blog post, April 2026: multi-slot sandwich attacks remain the residual vector after the April 8 patch; no rigorous post-patch measurement exists yet. → §6 Solana paragraph.
    • arXiv 2601.00738 (subslots paper), 2 January 2026: theoretical work on Ethereum sub-second slot timing; cited in §5a as the most recent academic work on CEX-DEX, with the explicit note that no 2026 source has updated the 23→11 firm count from arXiv 2507.13023.

Sections changed: §3 (HFT analogy), §5a (CEX-DEX framing — adds "most recent comprehensive measurement" hedge plus subslots cite), §5d (adds Q1 2026 contraction sentence after Vpe table), §6 Solana paragraph (leads with Q1 2026 REV decline, adds post-patch measurement gap), §6 Ethereum/L2 paragraph (adds Titan builder share, replaces Flashbots-only framing with arXiv 2604.00234 as the freshest measurement), §7 Winners (rewritten to open with the Q1 2026 picture before bringing in the older firm-count and Vpe-peak numbers).

New footnotes: 16 (subslots), 17 (DeFi Report Q1 2026 Solana REV), 18 (Helius Constellation post-patch), 19 (relayscan.io Titan share), 20 (Blockspace Under Pressure).

What I did not change:

  • The cold open (Vpe Dec 2024 – Jan 2025) — properly dated, still the cleanest named-actor anchor.
  • The worked example (Alice's Feb 2025 sandwich) — properly framed as retrospective.
  • The Losers and "Is this bad?" paragraphs in §7 — already use current (Q1 2026) framings via the Figment validator-dispersion citation; no refresh needed.
  • The closing "What changes when…" transition — fine.

What I considered but did not incorporate:

  • The 2026 Helius monthly Ethereum sandwich update — looked for it, doesn't exist past Oct 2025.
  • An updated CEX-DEX firm count for 2026 — looked for it, doesn't exist.
  • Blockworks Advisory Solana Q1 2026 DEX volume ($284.5B, 41% of on-chain trading) — relevant to the chapter's framing but not load-bearing for the Winners/Losers verdict; would belong in Chapter 8 (Solana) when that gets drafted.

This is the eight-question self-check the chapter spec template requires, plus an explicit ledger of what changed between RESEARCH/OUTLINE and DRAFT and what I'm uncertain about. Nick reviews this alongside the draft.

The eight required questions ​

  1. Could a smart business reader with zero crypto background follow this chapter on first read?Yes — with one caveat. Every term that appears for the first time is defined inline before it carries any load. "Mempool" gets one sentence in the setup and a glossary entry; "Jito tip" gets a sentence at first use and a glossary entry; "frontrun"/"backrun" are both defined in the sandwich section before the diagram uses them. The TradFi anchor (HFT desk circa 2010) is planted in the setup and returned to in the verdict. The one caveat: §5a leans briefly on "passive liquidity provider" with only two-sentence inline context, because the full LP treatment lives in Chapter 2 (not yet drafted). If Chapter 2 is produced first, this caveat goes away.

  2. Is every actor named, and is it clear how each one makes money?Yes. The chapter names: the searcher (full sidebar with revenue and cost breakdown), the validator and the block builder (cameo — money source named as Jito tips and priority fees, full treatment in Ch 5), Jupiter (revenue named as routing-margin / slippage-protection sales), the passive LP (income from pool fees, loss source named), the borrower (no revenue, named only as the liquidatable party), and the specific named programs (Vpe, 2Fast, jaredfromsubway.eth — though the last is only implicit in the cumulative-extraction language).

  3. Is there a worked example with specific dollar amounts threaded through the chapter?Yes. Alice's $10,000 USDC→SOL Jupiter swap in Feb 2025, $73 sandwich loss. Six explicit beats: setup in §3, planting in §4 ("the chapter resumes Alice's trade in a moment"), payoff in §5c with the D1 sequence and D2 dollar-split, gross-vs-net pivot in §5d ("the searcher's $45 of gross is not net profit"), counterfactual in §6 ("the same Alice trade in May 2026, routed through Jupiter, would lose closer to zero"), and named-loser anchor in §7 ("Alice's $73 on a $10,000 swap is a defensible illustrative number").

  4. Does the chapter end with a clear "who wins, who loses" verdict?Yes. §7 is the required half-page, structured as winners / losers / the honest answer to "is this bad?" — exactly the three subsections the chapter spec template prescribes. The closing line ("the dollar of slippage Alice paid in February 2025 is not lost to corruption; it is lost to a market structure that, until recently, paid more attention to throughput than to who pays for it") is the line I think Nick will most want to either keep verbatim or rewrite.

  5. Are all numbers sourced in footnotes?Yes, with two explicit caveats. Every dollar figure, percentage, throughput number, and named incident has a footnote with a URL and an access date. The two caveats, called out in-line:

    • Footnote 7 acknowledges that the $50–$150 sandwich-loss range for a $10K swap is derived from Vpe data scaled to the trade size, not directly published. The $73 is illustrative.
    • Footnote 9 acknowledges that the "hundreds of millions" cumulative Solana sandwich figure is order-of-magnitude.

  6. Does the chain comparison box exist and contain real differences (not "Solana is faster")?Yes. The three paragraphs make three structurally different points: Solana's surface has shrunk because of a specific protocol change on a specific date (and the patch is named); Hyperliquid has no surface at all because of a specific architectural choice (HyperBFT match-inside-consensus, no mempool); Ethereum's L1 sandwich surface has compressed by 4× while its L2 surface looks worse-by-different-shape (spam consuming half of Base's gas while paying a tenth of its fees). No throughput-comparison sentences.

  7. Did I avoid every banned move from the Book Bible?Yes, but I want to flag two near-misses for your review.

    • I did not use "revolutionary," "game-changing," "paradigm-shifting," "the future of," "disrupting," "rugged," "broken," "dead," or "scam" anywhere. I did not compare anything to "Web 1.0" or "the early internet." I did not use "It's important to note that…" or similar throat-clearing.
    • Near-miss 1: The sentence in §5d "The competitive moat is no longer raw algorithmic edge. The moat is whether a builder will give your bundle priority…" could read as slightly editorial. I think it's load-bearing analysis the chapter has earned by §5d, but it's the sentence most vulnerable to the "sounds like a podcast" test. Open to softening.
    • Near-miss 2: The verdict's "is this bad?" paragraph uses the phrase "pure transfer with no useful function" about sandwiches. The Book Bible encourages clinical neutrality, but this is the chapter's strongest moral claim and I think it's defensible — the architectural response by both Solana and Jupiter is the evidence that the system agrees with the assessment. Flagging anyway because it's the chapter's hottest sentence.

  8. Would the Goldman MD finish this chapter without checking her phone?I read it aloud — twice. The Vpe cold open earns the first page. The Alice setup in §4 sets up an expectation that the §5c reveal pays off. §5d's gross-vs-net pivot is the right kind of reveal for a numerate reader (the searcher's $45 is not the searcher's net profit). The chain comparison is genuinely informative because the three paragraphs say different things. The verdict's closing line is the kind of sentence the Bible's Sample C suggests aiming for. There are two moments where I think she might slow down: the JIT footnote (10) and the back half of §5d. Both are optional rereads, not dead air.

What changed between phases — and what's load-bearing ​

Claims dropped from RESEARCH.md ​

  • The $370M–$500M Solana sandwich aggregate (RESEARCH.md §1, claim N4). Dropped per the working assumption set in OUTLINE.md. Replaced with a hedged "in the hundreds of millions of dollars" in footnote 9 only, never as a load-bearing chapter claim. If you'd prefer to include it explicitly even as a hedged estimate, easy fix.
  • The 80–90% gas-as-revenue ratio (RESEARCH.md §3). Dropped because no clean primary source exists for it. Replaced functionally by the Paradigm theoretical ceiling + the Flashbots Base examples in §5d. The chapter's "real business with real costs" point now rests on those two sources, not on an unsourced ratio.
  • The Jared 2.0 / jaredfromsubway.eth $50M+ cumulative figure (RESEARCH.md N12). Dropped — never named in the draft. The named-program slot is taken by Vpe (Solana, fresher data, cleaner source) and 2Fast (the one-shot arbitrage cameo). Bringing Jared back would require either replacing 2Fast in §5a or expanding §5c with an Ethereum case — either is doable but feels like padding.
  • The Helius Solana Ecosystem Report H1 2025 peak-tip-day numbers (RESEARCH.md N11, $25.16M Jito tips on TRUMP launch day Jan 20 2025). Dropped — the one-day peak is dramatic but the chapter already has the Vpe 30-day window doing similar work. Holding this for a possible Chapter 5 callback.

New claims added in the draft (and where they came from) ​

  • The "moat is the relationship with the builder" claim in §5d. Inferred from the arXiv 2507.13023 paper's consolidation finding (23→11 firms, top three 90%) plus Flashbots' two-firm-80% finding on Base. The paper does not say "the moat is the relationship" in those words; the chapter says it as my synthesis. Flagging because it's the chapter's central analytical claim and is editorial-by-degrees.
  • The phrase "the polite word for this difference is MEV" in §3. Editorializing; the underlying definition is the Book Bible-approved one. If the framing reads as too cute, the sentence is easy to cut.
  • Jupiter monetizing "slippage protection to traders who would otherwise have lost it" in §7. The framing is mine; Jupiter does monetize, but I have not cited their revenue model directly. The sentence is defensible — they capture routing fees and the value of better execution shows up in their take-rate — but a careful reader might want a source. Easy to soften or footnote.
  • The "0.07-second finality" Hyperliquid figure in §6. From the Hyperliquid docs, but I rounded "roughly 70ms" to "roughly seventy-millisecond." Worth checking against the spec page directly if Nick cares about precise figures.

Things I'm uncertain about ​

  1. Should Alice be named or kept second-person? I chose to name her (the prologue, per the Bible's Sample B, used "you"). My instinct: giving her a name makes the example narratively cleaner because the chapter then has a specific named loser to point to in the verdict. But it does break with the prologue's second-person framing. Reversible if you'd rather she stay "you" or "a retail trader."
  2. The illustrative $73 split. I anchored the tip share to the Vpe ratio (34% of gross), which gives $23. The remaining $45 / $5 split between searcher gross and gas is partly a judgment call. The split is illustrative and footnote 7 says so. If you want the numbers re-derived from a different anchor (e.g., a specific MEV-Inspect dashboard for an Ethereum sandwich), I can redo D2.
  3. JIT in a footnote vs. in §5c body. I put JIT in footnote 10 to avoid breaking the sandwich narrative's momentum. The Book Bible's voice rule 9 says "diagrams replace prose, not supplement it" — by similar logic, footnotes can carry the asterisks the body doesn't need. But a careful reader who skips footnotes will miss JIT entirely. Easy to promote into body text if you'd prefer.
  4. The "Meet the Searcher" sidebar formatting. I used VitePress's ::: info custom container, which renders as a styled callout in the browser view. In a printed manuscript it'll render as either an <aside> block (HTML/EPUB) or a custom LaTeX environment (PDF). The format is fine for the dev view but the cross-format render needs a Pandoc filter at MANUSCRIPT.md assembly time. Not a Phase 3 blocker — flagging for the publishing-layer task.
  5. The "is this bad?" verdict tone. I wrote it more declarative than RESEARCH.md's hedged language suggested. If you find it too declarative on a re-read, the fix is to soften "pure transfer with no useful function" to "a transfer with no obvious useful function." But I'd push back on softening further — the architectural responses on both chains are the evidence that lets the chapter state the conclusion plainly.

Places where the prose got technical and might lose the reader ​

  • §5c, the paragraph defining frontrun and backrun. Technical definitions, immediately after a diagram. I tried to keep them short and to define each one in a single sentence. If your reader stops reading here, the fix is to delete the explicit definitions and rely on the diagram + the inline use of "the frontrun" / "the backrun." I'd test it on a real reader before cutting.
  • §5d, the Paradigm-ceiling paragraph. Reads as more theoretical than the rest. I kept it to two sentences and put the empirical Flashbots data immediately after as the load-bearing version of the same point. If the abstract-then-concrete pattern doesn't land, the Paradigm cite can drop into a footnote and §5d becomes purely empirical.
  • The CEX–DEX paragraph in §5a. "$233.8 million extracted over nineteen months across 7.2 million transactions on $241.7 billion of routed volume" is a lot of numbers in one sentence. I broke it across two clauses; could break further if it skims off the reader's attention.
  • The Hyperliquid paragraph in §6. The HyperBFT / HLP / on-chain-CLOB stack lands four new proper nouns in one paragraph. I leaned hard on "no public mempool, no pending state to observe" as the load-bearing simplification, and trusted Chapter 9 to do the full architectural treatment. If the paragraph reads as dense, the fix is to cut one of HyperBFT or HLP and lean harder on the architectural verdict ("the surface does not exist").

Files written/modified in Phase 3 ​

The dev server picks all of this up on restart and renders the chapter at /book/chapters/04_searcher/DRAFT.

Phase 3 is complete. The chapter is now in Nick's review queue.