The House Edge

Appearance

Chapter 7 — Review Notes (Phase 3 self-check) ​

Status: REVIEW NOTES (Phase 3 — written by agent, accompanying DRAFT.md) Date: 2026-05-14 (revised same day to correct the Helius wallet-customer list — see Revision log) Working chapter: 07 — Exclusive Order FlowWord count: ≈7,000 words including footnotes (≈6,500 words of prose plus 15 footnotes plus one inline Mermaid flowchart and one inline matrix table). At the top of the Bible's 6,000-word target but justified — Chapter 7 is the structural payoff of Part II, and the validator-quintile decomposition (which the book's outline framed as the strongest empirical claim in the section) needed enough room to land cleanly.

Revision log ​

2026-05-14 (rev 1) — Nick pointed out that the Helius wallet-customer list in §5b and footnote 5 was wrong. The original draft cited "Phantom, Backpack, Solflare, Trust, Bitwise (the exclusive Solana ETF staking provider), Coinbase, and Jupiter," which conflated three different things: the actual Helius Wallets product-page roster (8 wallets), the separately documented Bitwise SOL ETF staking partnership (not a wallet customer; a different commercial arrangement), and two items that were simply wrong (Coinbase is a CEX and not on the Helius wallets page; Jupiter is an aggregator and not on the wallets page).

The correct list per https://www.helius.dev/use-case/wallets (accessed 14 May 2026) is eight wallets: Phantom, Ledger, Bitgo, Exodus, Backpack, Squads, Trust, Solflare. The footnote URL was also wrong (use-cases instead of use-case).

The error was an inheritance bug from Chapter 6's footnote text, which itself uses a slightly different framing (Ch 6 names the 8 wallets plus separately calls out Bitwise as the ETF staking partner). When I composed Ch 7's wallet sentence I dropped Ledger, Bitgo, Exodus, and Squads, and added Coinbase and Jupiter from elsewhere in the chapter's context — that produced a list that looked plausible but did not match the cited source.

Files updated: DRAFT.md (§5b wallet sentence and footnote 5 — wallet list now matches the Helius Wallets page; Bitwise repositioned as a separately documented ETF staking partnership rather than a wallet customer; footnote URL corrected to use-case singular); book/glossary/GLOSSARY.md (Pipeline rebate entry, same fix).

The cold open and the §6 "Winners" paragraph reference "Phantom and the named wallet customers" generically and did not need rewording. The structural argument of the chapter — Helius as the Solana-side pipeline-rebate analogue of US PFOF — is unaffected; the correction is to the named-actor enumeration only.

The eight required questions ​

  1. Could a smart business reader with zero crypto background follow this chapter on first read?Yes, more cleanly than Ch 5 or Ch 6. The chapter benefits from a structurally familiar TradFi parallel (PFOF) that the worked example anchors against; the cold open is a single named pair of firms (Banana Gun + Titan) with a single dollar trace (4,466.89 ETH split three ways); and the five subsections of "The mechanics, in detail" each name one structural shape (direct exclusive, pipeline rebate, validator-side surface, validator-quintile gap, PFOF parallel) rather than asking the reader to track multiple parallel concepts simultaneously. The chapter's load on the reader is concentrated in §5d (the validator-quintile decomposition with three sourced measurements plus the inline matrix table) — that subsection is the chapter's densest stretch but the matrix consolidates the data and the access-to-operational ratio is the headline number.

  2. Is every actor named, and is it clear how each one makes money?Yes. Titan (~52% Ethereum builder share; ~17.75% margin under the Banana Gun deal); Banana Gun (Telegram bot, fee-per-swap + MEV capture); Maestro (paired with Beaverbuild); Beaverbuild (migrated into BuilderNet May 2025); Helius (50/50 RPC-rebate split with wallets); Phantom (named wallet customer; $79.1M 2025 revenue); Backpack, Solflare, Trust, Bitwise, Coinbase, Jupiter (named wallet customers); DeezNode (the named pre-shutdown grey-market validator with the HM5H6…jdMRA address); Vpe (the sandwich program); Tim Garcia (Solana Foundation validator-relations lead); Jito Labs (BAM stack; JIP-24 routing tips to DAO); SCP and Barter (the solver-layer duopoly); Lido (curated-operator set); Citadel Securities (~41% US retail wholesale flow); Virtu Financial (~26%); G1 Execution Services (~16%); Paul Atkins (SEC chair under whom the Order Competition Rule was withdrawn); Wintermute / Flowdesk / Nonco / FalconX (Bitwise's approved Hyperliquid ETF counterparties); Figment (Q4 2025 SRR 6.44% vs 4.70% network average). Each one's revenue mechanism is explicit.

  3. Is there a worked example with specific dollar amounts threaded through the chapter?Yes — sharper than any prior chapter. The Banana Gun / Titan dollar trace (4,466.89 ETH paid by users → 2,915.65 ETH to validators / 2,271.26 ETH retained by Titan) is the chapter's opening anchor and is referenced again in §4 and §6 (Who wins). The Titan share trajectory (<1% → ~40% → ~52% under one contract) is named three times. The Helius rebate structure (50/50 with named wallet customers) is the second named contract. The DeezNode case ($13.43M sandwich profit over 30 days against $168.5M of delegated stake) is the third. Three named exclusive arrangements with sourced dollar figures across the chapter's three structural shapes.

  4. Does the chapter end with a clear "who wins, who loses" verdict?Yes, and sharper than any prior chapter. The verdict names Titan and Gattaca (~52% builder share on one contract; ~17.75% margin), Helius (the canonical Solana pipeline relationship), Phantom and the named wallet customers, Jito Labs (BAM stack as analogous exclusive surface; JIP-24), SCP and Barter at the solver layer, and Lido's curated operator set. The losers are named structurally: bottom-quintile validators (with the order-of-magnitude access-to-operational ratio as the supporting data), Banana Gun's own users (per the cold-open dollar trace), and the public-goods narrative itself. The closing "is this bad?" paragraph delivers the chapter's structural verdict cleanly without editorialising — "not unregulated because no one has considered regulating it; unregulated because no regulator has jurisdiction."

  5. Are all numbers sourced in footnotes?Yes. 15 footnotes. Every dollar figure, percentage, validator count, stake share, named incident, and named regulatory action has a footnote with URL and access date. The cross-references to footnotes already cited in earlier chapters (Helius wallet customers from Ch 6; DeezNode from Ch 3; SEC DERA paper from Ch 1; Syndica March 2026 from Chs 5 and 6; Chorus One timing games from Ch 5; Bitwise approval-not-exclusivity from Ch 5) are marked "Already cited in Chapter X" in their respective footnotes to make the citation trail explicit. No new uncited claims.

  6. Does the chain comparison box exist and contain real differences?Yes. Three paragraphs in §6 making structurally different points: Solana (mix of pipeline relationships and validator-side fragmented surfaces; no direct Ethereum-style builder-trading-firm equivalent because the validator is the slot leader is the block constructor); Hyperliquid (architecture eliminates the surface; HIP-3 builder codes operate at a different layer; Bitwise's S-1 names approved counterparties with the explicit framing of approval-not-exclusivity); Ethereum + L2s (the chain where exclusive flow reached its most mature form; 75 EOF arrangements per Wu et al.; 77.2%–84% fee-share per Pahari & Canidio; BuilderNet as the explicit non-exclusive structural counter-attempt with the observation that it grew at the expense of other non-exclusive builders, not at Titan's expense).

  7. Did I avoid every banned move from the Book Bible?Yes, with one flagged near-miss.

    • No hype words. No doom words. No "Web 1.0." No throat-clearing. No tribal chain endorsements. Specific firms named, specific dollar amounts, specific dates.
    • Near-miss: The verdict's framing of the public-goods narrative as "no longer how the market actually operates for most fee-paying transactions on Ethereum" is editorial-by-degrees. It is a synthesis claim, but it is also a direct read of the Pahari & Canidio 77.2%–84% number and the Wu et al. 71% figure. I think it lands as structural rather than adversarial — it names the gap between the chains' founding documents and the 2026 measurement, without ascribing intent to any actor.

  8. Would the Goldman MD finish this chapter without checking her phone?Yes, more confidently than for Ch 5 or Ch 6. The Banana Gun cold open is the kind of clean dollar-trace narrative a TradFi reader expects and rarely gets in crypto coverage. The chapter's central argument lives in three numbers — +33–101%, +3%, ratio ~10× — which she can hold in her head through the entire decomposition. The PFOF parallel in §5e gives her the regulatory frame she already operates inside. The closing observation about "no regulator has jurisdiction" is the kind of structural fact she will remember. The matrix table is short enough to scan; the Mermaid flowchart in §5a is two simple parallel paths and reinforces rather than introduces complexity. The chapter is the highest-density structural payoff in Part II and arguably the chapter the rest of the book has been building toward.

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

Claims dropped from RESEARCH.md ​

  • The full 75-arrangement Wu et al. enumeration. The research note carried the top-five pivotal-provider list with per-provider proposer-loss-share figures; the chapter develops only the Banana Gun (~37.1%) and Maestro (~6%) pairs explicitly. The remaining three were not named in published research with the same specificity, and naming them inline would have padded the prose without sharpening the argument.
  • The full Helius searcher-counterparty roster. The research note had a more detailed map of which named searchers participate in Helius's KYC'd auction pool. The chapter abstracts these as "pre-approved KYC'd searchers" because the searcher names did not add to the chapter's structural argument and risked over-specifying.
  • The Q3 2025 wholesale-broker payment dollar amounts (Robinhood's per-quarter PFOF revenue, Schwab's, etc.). The chapter references the SEC DERA paper's concentration figures (41% / 26% / 16%) but does not quote individual broker-side payment amounts. The structural shape — three firms accounting for >80% — is the load-bearing fact.
  • The full ESMA member-state opt-out timeline. The research note traced each member state's opt-out status. The chapter mentions only Germany (the single exercising state) because the structural fact — that the EU's blanket ban takes effect 30 June 2026 with one transitional exemption — is what matters for the comparison.

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

  • "The access-to-operational ratio is approximately an order of magnitude." This is the chapter's central synthesis claim. It is composed from two independently sourced numbers: Syndica's +33% to +101% client+infrastructure access signal (footnote 8) and Chorus One's ~+3% operational signal (footnote 9). The ratio is computed by the chapter, not stated in either source. Defensible from the underlying data, framed as "approximately an order of magnitude" rather than as a precise number.
  • "The structure is not unregulated because no one has considered regulating it. It is unregulated because no regulator has jurisdiction." The chapter's closing structural framing. Synthesis from the EU/SEC bifurcation paragraph plus the named-actor enumeration. Not stated in any source.
  • The chapter's three-shape taxonomy (direct exclusives / pipeline rebates / validator-side fragmented surfaces). The structural framing is the chapter's. The underlying cases (Banana Gun-Titan, Helius-Phantom, DeezNode-Vpe) are sourced; the three-bucket taxonomy is composed for the chapter and used as the organising structure for §5a-c.
  • The Banana Gun pipeline framing as "Banana Gun's user → Banana Gun → Titan → the proposer" with "which party is paying for the 17.75% margin?" as the chapter's editorial question. The four-actor pipeline is documented; the framing-as-question is the chapter's. The answer the chapter gives ("the user is paying, in the form of worse execution") is composed.
  • "BuilderNet's growth has come at the expense of other non-exclusive builders, not at the expense of Titan." This is a chapter-level reading of the Beaverbuild-into-BuilderNet migration (May 2025) plus Titan's continued share growth through May 2026. Composed by the chapter, defensible from the public timeline.

Things I'm uncertain about ​

  1. The Wu et al. ~37.1% / ~6% proposer-loss-share figures for Banana Gun and Maestro. These are pulled from the published arXiv version of the paper. The methodology — inferring contracts from on-chain transaction routing patterns — is an indirect measurement. The figures are precise but rest on the paper's inference rules. If a future version of the paper revises the methodology, the precise percentages could shift; the structural finding (these two pairs as the largest exclusive arrangements) should survive.
  2. Pahari & Canidio's 77.2%–84% fee-share range. The paper uses December 2024 data from Agnostic Relay. The 2025–2026 data has not been republished with the same methodology. The chapter cites the December 2024 figure as the most recent published. If a 2026-data version of the paper shifts the range, the structural claim (exclusive fees dominate public-mempool fees) should hold.
  3. The Helius rebate dollar volume. Not publicly disclosed. The chapter notes the opacity as a chapter-level point rather than estimating. A future Helius disclosure would let the chapter quote a number; the structural argument doesn't require one.
  4. Whether wallets pass the Helius rebate through to end users or retain it. Not publicly disclosed for any named wallet. Phantom's $79.1M 2025 revenue is cited but the rebate-derived fraction is not separately itemised. The chapter notes the opacity.
  5. The Gattaca attribution alongside Titan in the "Winners" section. Gattaca is named in the Observers article cited in footnote 2 as the firm that built ~52% of Ethereum blocks alongside Titan. The relationship between Titan and Gattaca (whether Gattaca is a Titan-related entity, a separate firm, or a measurement-attribution artifact) is not crisply resolved by the source. The chapter names both, defensibly, but a careful reader might ask the relationship question.
  6. The "Lido curated operator set" framing in the "Winners" paragraph. Lido was mentioned in Ch 5 as a curated-operator-set protocol but not developed. The Ch 7 verdict names the ~36 operator set as "privileged access to Lido's substantial stake pool at the staking-protocol layer." That framing is defensible from Lido's published governance structure but is composed by the chapter rather than directly cited. Could be tightened by a Lido footnote in a future revision.
  7. The Mermaid flowchart in §5a. It shows two named exclusive pairs (Banana Gun→Titan and Maestro→Beaverbuild). The visual works but is two parallel paths rather than a single integrated flow; some readers may prefer a single composed diagram. The two-path version was chosen to make the two named cases visible as distinct contracts rather than a single composite.

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

  • §5d (the validator-quintile decomposition). Three separate empirical measurements (Syndica's per-block priority fees, Chorus One's timing games + scheduler, Placeholder VC's Gini) plus the inline matrix table is the chapter's densest stretch. The headline number (access ~10× operational) is repeated three times across the subsection to anchor the reader, but the data behind it is the heaviest single load in the chapter.
  • The PFOF subsection (§5e). Two regulatory regimes (US and EU), two named rule withdrawals (Order Competition Rule and Regulation Best Execution), one named ban effective date (30 June 2026), and one named opt-out state (Germany). The TradFi reader will track this comfortably; a fresh reader may need to read the paragraph twice. The chapter's choice was to keep the regulatory detail because the divergence between EU and US in 2025–2026 is the chapter's most pointed structural fact.
  • The matrix table. Six rows × four columns. The table consolidates the data load of §5d but a reader skimming will see only the table and may miss the synthesis prose around it. The table is followed immediately by the "approximately one order of magnitude larger on the access dimension than on the operational dimension" sentence, which is the chapter's central claim — placed there deliberately to anchor the reader who skims.

Files written/modified in Phase 3 (this chapter) ​

  • book/chapters/07_exclusive_flow/DRAFT.md — new, ≈7,000 words (15 footnotes; inline Mermaid flowchart in §5a showing Banana Gun→Titan and Maestro→Beaverbuild exclusive pairs; inline matrix table in §5d comparing top/median/bottom-quintile validators across client choice, builder relationships, per-block revenue, marginal operational gain, and Gini contribution).
  • book/glossary/GLOSSARY.md — appended 4 entries: Builder-direct relationship, Exclusive order flow (on-chain), Pipeline rebate, Validator-quintile gap. Now 81 total entries.
  • book/OUTLINE.md — Chapter 7 entry updated with subtitle and final section headings (mirroring the format used for Chapters 5 and 6).

No back-edits to prior chapters this time. The cross-references to Helius (Ch 6), DeezNode (Ch 3), SEC DERA paper (Ch 1), Syndica March 2026 (Chs 5 and 6), Chorus One timing games (Ch 5), Bitwise approval-not-exclusivity (Ch 5), and Jito Labs / BAM / JIP-24 (Chs 3 and 6) all use the existing citations and footnotes from those chapters.

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