The House Edge

Appearance

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

Status: REVIEW NOTES (Phase 3 — written by agent, accompanying DRAFT.md) Date: 2026-05-14 (revised same day for timeline correction + diagram replacement — see Revision log below) Working chapter: 03 — The Mempool and What Replaces It Word count: ≈5,500 words including footnotes (≈5,000 words of prose plus 24 footnotes plus the inline Mermaid timeline and chain-comparison table). At the top of the Book Bible's 4,000–6,000 target — the user's content directive (full Jito mempool history + BAM + Harmonic + ETH private routing + L2 sequencers + Hyperliquid + censorship trade-off) needed the room.

Revision log ​

2026-05-25 (rev 3) — Fixed reader-order references. Chapter 3 now introduces Vpe directly instead of relying on prior Chapter 4 knowledge, and the verdict no longer depends on a Chapter 4 loss figure as prior context. Chapter 2 also had one reader-facing out-of-order Alice reference removed. Forward references to Chapter 4 are now phrased as previews, not back-references.

2026-05-14 (rev 2) — Nick flagged a substantive structural correction: the post-shutdown Solana surface didn't disappear, it fragmented. The chapter previously presented Jito Labs' shutdown plus the formal Beam / BAM / Harmonic stack as the full story. The corrected framing acknowledges that, underneath the formal stack, the visibility-as-product market re-emerged at three layers:

  • RPC level: Kolibrio (VC-backed Order Flow Auction startup, $2M Jump-led seed Feb 2023; operates "Meow RPC" — integrators route through Kolibrio, who auctions visibility to searchers and rebates back to integrators)
  • Wallet infrastructure level: Helius (network's largest single validator at 15M+ SOL stake; runs RPC for Phantom, Backpack, Solflare et al.; openly markets "SOL rebates from trades that create arbitrages")
  • Validator co-location level: "non-voting validator node in the same rack as a paid RPC, forwarding to leader TPU via QUIC ahead of public gossip" — the "right next to the TPU port" pattern

Additionally, the chapter now identifies Vpe — the canonical sandwich operator introduced here and later revisited in Chapter 4 — as operated by DeezNode, a specific Solana validator (address HM5H6…jdMRA) whose stake nearly tripled (308K SOL → 803K SOL across November–December 2024) on the back of a public 50% revenue-share offer to opting-in validators. This is the cleanest documented case of "the shutdown sent the market underground": an anonymous sandwich program turns out to be linked to a specific validator that grew its stake by selling visibility access.

The chapter's Harmonic framing was also refined to acknowledge that its "configurable preferences" architecture creates a structural hook for partner-specific exclusivity, even though Harmonic's own marketing frames the marketplace as open. This is an inference from the preferences-as-primitive design rather than a documented Harmonic practice — the footnote 29 hedge is explicit.

The verdict was rewritten to reflect the fragmented-not-eliminated reality. The previous verdict framed the architectural responses as having "compressed the surface" — true but incomplete. The new verdict acknowledges that the formal surface compression is real and the dollar amounts per sandwich attack have materially fallen, but that the market for visibility — the willingness of intermediaries to sell pre-leader transaction sight and of searchers to pay for it — is alive in a way Jito's 2024 statement implied it would not be.

Files updated: DRAFT.md (new "fragmented surface" paragraph in §5a Solana; refined Harmonic paragraph; rewritten verdict; five new footnotes 25–29); RESEARCH.md (new claims 4a, 4b; refined claim 11); book/glossary/GLOSSARY.md (new Kolibrio entry, alphabetised before Liquidation).

Open Ch 4 question not addressed in this revision: Chapter 4's cold open says of Vpe "No one knows who runs it" — at the human-identity level this is still true (DeezNode itself is an anonymous validator address HM5H6…jdMRA), but at the on-chain-entity level the Vpe program and DeezNode are publicly linked. Worth a one-sentence update to Ch 4 footnote 1 to acknowledge the linkage; not done in that revision because Nick's directive was Ch 3-focused.

2026-05-14 (rev 1) — Nick flagged two issues after first read:

  1. The Jito mempool timeline was wrong. The draft said "nineteen months" of operation and "August 2022" launch — both inferred from the Helius source by the original research agent. Per Nick's correction, sourced to Jito Labs and Solana Compass coverage:
    • Late 2022: Jito-Solana entered mainnet and testnet in validator-only form.
    • 25 May 2023: Block Engine opened broad searcher access to the mempool, after the validator-only cohort had processed over a million bundles.
    • 8 March 2024: Jito Labs shut the mempool down.
    • Total: roughly ten months of broad searcher access, with about six months of prior validator-only operation. Not nineteen months.
    • Also corrected: the "exceeded ten thousand SOL on the day before shutdown" framing replaced with the more accurate run-rate framing (~10K SOL/day in tips, ~$500M+ annualised at the prevailing price).
    • The chapter's "the reason that did not need to be stated" closing of the cold open also got sharper attribution: sandwich attacks during the memecoin frenzy made the negative externalities untenable for the firm to host.
  2. The three-paths Mermaid flowchart at the top of the Solana subsection was weird in context. The chart preceded the prose that introduced the three layers it was supposed to illustrate, and the labels were truncating in Mermaid's default box sizing. Replaced with a Mermaid timeline diagram showing Solana's mempool arc from late 2022 to early 2026 (Jito-Solana mainnet → 25 May 2023 broad access → 8 March 2024 shutdown → June 2024 Foundation enforcement → 21 July 2025 BAM launch → early 2026 Beam-as-default → 8 April 2026 protocol patch). The timeline lands later in the subsection — after the historical prose has established what each milestone meant — so it serves as a synthesis visual rather than a confusing preview.

Files updated: DRAFT.md (cold open, Solana subsection paragraph 2, footnote 4, diagram replacement); RESEARCH.md (claim 1, N1 row in the numbers-to-verify table); book/OUTLINE.md (Ch 3 cold open description).

No load-bearing chapter argument changed. The Jito-shut-the-mempool-down beat, the Foundation enforcement, the BAM / Beam / Harmonic stack, and the verdict all stand. The corrections tighten the timeline and replace a confusing diagram with a clearer one.

The eight required questions ​

  1. Could a smart business reader with zero crypto background follow this chapter on first read?Mostly yes, but the Solana subsection is the chapter's densest stretch. Three architectural layers introduced in sequence (Beam, BAM, Harmonic) plus the historical arc (Jito mempool launch August 2022 → shutdown March 2024 → Foundation enforcement June 2024 → Beam → BAM 2025 → April 8 2026 patch → Constellation forward-looking) is a lot of named institutions and dates. The Mermaid timeline is meant to carry some of that load. A reader who got through Chapters 1 and 2 will have the trade and liquidity vocabulary, but Vpe and the searcher role must be introduced here cleanly before Chapter 4 develops them.

  2. Is every actor named, and is it clear how each one makes money?Yes. Jito Labs (6% of BAM tips, Block Engine fees); Jupiter (routing margin via Ultra/Beam); the Solana Foundation (no direct revenue; enforcement role); Vpe (introduced here as the canonical operator); Flashbots (the BuilderNet operator now, not centralised builder); CoW DAO and Consensys SMG (MEV-Blocker); Titan, BuilderNet, Quasar (Ethereum block builders); Coinbase (Base sequencer revenue ~$42.4M H1 2025); Offchain Labs (Arbitrum sequencer + Timeboost auction); the Optimism Foundation / Unichain (Flashblocks; MEV-back-to-user). Plus the named individuals (Tim Garcia, Hasu, Vitalik) who appear in cameo.

  3. Is there a worked example with specific dollar amounts threaded through the chapter?Yes, but lighter than Chs 2 and 4. Alice's $10,000 USDC→SOL swap is the anchor, with the +0.006% (protected) vs −0.14% (unprotected) Ultra V3 numbers as the dollar payoff. Carlos (new character) appears in the Ethereum subsection but his transaction's dollar outcome is not separately quantified — the section's load-bearing numbers are aggregate (Flashbots Protect's 2.1M+ accounts, $43B+ shielded volume; Wahrstätter's 54.6% of block value private; Titan's 50.6% of blocks built). The chapter's dollar-density is intentionally lighter because the structural-concentration story is the chapter's primary argument; individual-trade dollars matter less here than in Ch 4.

  4. Does the chapter end with a clear "who wins, who loses" verdict?Yes — the chapter's verdict is the sharpest in the book so far. Per the OUTLINE, this is the hinge chapter; the verdict matches Chapter 4's tone (clinical but explicit about who captured what). The closing observation ("the censorship-resistance properties of 'anyone can submit a visible transaction to a public network' have been substantively reduced") is the strongest single claim the chapter makes. It is sourced and defensible.

  5. Are all numbers sourced in footnotes?Yes. 24 footnotes. Every dollar figure, percentage, validator count, share number, and named incident has a footnote with URL and access date. The one exception is the April 8 2026 Solana protocol-level patch — the chapter cites the effect (simple validator-leader sandwiches largely suppressed) from secondary coverage in footnote 12, with explicit acknowledgement that the technical mechanism (SIMD, release tag, scope) is not publicly documented in any source the chapter could verify. This is flagged for a final verification pass before publication.

  6. Does the chain comparison box exist and contain real differences (not "Solana is faster")?Yes — and the chapter's chain-comparison table (D2 in OUTLINE) is unusually load-bearing because the architectural responses are the chapter's argument. Three paragraphs plus the table: Solana (mempool eliminated, three-layer stack replaced it); Hyperliquid (architectural elimination, not management); Ethereum + L2s (public mempool now holds the value-leftovers; L2 sequencers are one-firm private mempools by construction). FOCIL and Flashnet named as the protocol-level responses in development; explicit "not live on mainnet as of mid-2026" hedge.

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

    • No hype words. No doom words. No "Web 1.0." No throat-clearing. No tribal chain endorsements.
    • Near-miss 1: The verdict's closing observation ("the trade has not been a single decision by a single actor; it has been the cumulative effect of dozens of decisions by different actors over several years") is editorial-by-degrees — a synthesis claim, not a directly sourced one. I think it's defensible because the chapter has just spent 5,000 words documenting those decisions, but it pushes the structural-not-adversarial framing.
    • Near-miss 2: The "the reason that did not need to be stated" framing in the cold open is interpretive — Jito Labs' actual reasoning is in their March 2024 statement, and "the sandwich industry the firm could no longer comfortably host" is the chapter's reading rather than Jito's stated motive. This is fair editorial framing given the surrounding evidence (the firm gave up substantial revenue; the Foundation enforcement action against private-mempool participants followed three months later) but it's worth flagging as the chapter's most adversarial single sentence.

  8. Would the Goldman MD finish this chapter without checking her phone?I think so, but with reservations on the Solana subsection. The cold open is dated, named, and load-bearing — the right hook for this reader. The setup is short. The worked example is concrete. The Solana subsection (~1,500 words) is the chapter's densest stretch and is the one I'd most worry about. The Ethereum subsection lands cleanly (Wahrstätter's 54.6% number is the kind of single-stat anchor the Bible favours). The L2 subsection is structurally clear because the "one-firm sequencer" framing carries it. The verdict is sharp and earns its position. The 5,500-word total is at the top of the Bible's range — the chapter is denser than Chapter 1 or 2 because it has more to cover.

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

Claims dropped from RESEARCH.md ​

  • The full BAM revenue economics breakdown (RESEARCH.md claim 8): the Q2 2025 DAO share of ~22,391 SOL ($4M) is in the chapter, but the full 6%-tip-split distribution diagram is not. The chapter notes "no direct user fees" and the 6% split, and moves on. Full BAM business mechanics belong in Chapter 6 (Infrastructure Layer).
  • Constellation's specific committee structure (RESEARCH.md claim 13): the 16-proposer / 256-attester / priority-fee-per-CU ordering details are not in the chapter. The chapter mentions Constellation as a forward-looking proposal with the explicit hedge "a proposal, not a commitment." Full architectural treatment belongs in Chapter 12.
  • The Kairos resale-market detail for Arbitrum Timeboost (RESEARCH.md claim 20): named in the L2 subsection with a one-line cameo. The fuller arXiv paper finding (searchers routing around the primary auction) is not developed because that's a Chapter 7 (Exclusive Order Flow) question.
  • The Helius "Constellation" blog post specifics: cited in passing; the chapter does not develop the post-patch residual-vector argument because Chapter 4 already mentioned it and Chapter 12 will develop it.

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

  • The chapter's structural argument that "when 91% of Ethereum blocks are built by three firms, the question of what is visible in the public mempool matters less than the question of which firms have the relationships to put a transaction in front of those builders" — synthesis, supported by the cited Titan/BuilderNet/Quasar share numbers but not stated directly in any source. This is the chapter's most novel single claim and the bridge to Chapter 7.
  • "The visibility surface on Ethereum L2s, in 2026, is a one-firm private mempool" — framing, supported by the cited sequencer-operator facts but not stated this directly in any L2 source. The closest is L2BEAT's reporting on sequencer centralisation, which I did not cite directly.
  • The verdict's "the public-goods narrative of 'anyone can run a node and see the network' has been quietly traded for the institutional reality of 'submit a transaction visible to one firm, who chooses who else to share it with'" — synthesis claim. Defensible from the cited data; this is the chapter's strongest editorial sentence and the most likely candidate for Nick to soften or sharpen.

Things I'm uncertain about ​

  1. The April 8 2026 Solana protocol-level patch mechanism. Chapter 4 cites the patch; Chapter 3 must as well. The exact technical mechanism (SIMD, release tag, consensus-layer-vs-scheduler-layer change) is not documented in any source I could verify. The chapter cites the patch's effect with explicit hedging in footnote 12. This is the same flag carried forward from Chapter 4 — would benefit from a primary-source verification pass before publication.
  2. The relayscan.io May 14 2026 24-hour snapshot. Live numbers will drift. Footnote 17 cites them with the date stamp; the chapter notes the structural conclusion (top three builders ~91%) which is the more durable claim. If the numbers shift materially by publication time, the snapshot footnote should be re-pulled.
  3. The Wahrstätter et al. 54.6% figure for Ethereum private-flow value share. This is a peer-reviewed measurement updated through 2025, but a 2026 update would be ideal. If a fresher number surfaces in research before publication, the chapter's most important Ethereum statistic should be updated.
  4. Carlos as a new character. The book has used Alice across multiple chapters; introducing Carlos for one subsection breaks the convention slightly. I judged that re-using Alice for a different chain at the same trade size was awkward (Alice's identity is tied to Solana via the prologue and Chapter 4); Carlos's institutional framing was a cleaner pedagogical choice. If you'd rather have Alice carry both worked examples, the change is straightforward — Alice becomes the institutional Ethereum trader for the Carlos beat and the chapter loses the "the book has used Alice enough" framing.
  5. The Solana Foundation enforcement framing. The June 2024 removal of ~32 validators is cited as a clean accountability beat. Whether to develop this further (it is currently a paragraph) or leave it as it stands is a judgment call. My instinct: leave it. The book has a structural argument to make and shouldn't dwell on a single enforcement event.
  6. The verdict's two-sided framing on the trade-off ("Whether that trade is good depends on which risk the reader weighs more heavily"). This is the most explicitly readers-decide framing the book has used. If Nick wants a sharper verdict, the chapter has the evidence to make one.

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

  • The Solana subsection's three-architectural-layer development (Beam → BAM → Harmonic). The Mermaid flowchart helps but the prose density is real. If a reader stops here, the chapter is hard to recover from. Worth reading aloud to confirm the transitions land.
  • The BAM TEE explanation. "AMD SEV-SNP hardware enclaves" lands two acronyms in one paragraph. The chapter's TradFi analogue (a clearinghouse running inside an HSM) is in the glossary but not in the main prose. Could be pulled in if the section reads as too technical.
  • The MEV-Boost relay/builder/proposer separation. The chapter develops the architecture in roughly two paragraphs. For a reader who hasn't met MEV-Boost before, this is a lot. If the Ethereum subsection feels dense, the fix is to push the relay-share statistics to footnotes and lead with the structural concentration (top three builders = 91%) as the load-bearing observation.
  • The L2 sequencer subsection has three case studies (Base, Arbitrum, Optimism Unichain) in 700 words. Tight but manageable. If anything tightens further, the Optimism Unichain Flashblocks paragraph is the most cuttable — the architectural point (MEV-back-to-user via revenue share) lands without the specific 200ms-block-time figure.

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

  • book/chapters/03_mempool/DRAFT.md — new, ≈5,500 words (24 footnotes; inline Mermaid three-paths flowchart + inline markdown chain-comparison table)
  • book/glossary/GLOSSARY.md — appended 12 entries (BAM, Flashbots Protect, Flashnet, Harmonic, Inclusion list / FOCIL, MEV-Blocker, MEV-Boost, MEV-Share, Private order flow, Public order flow, Relay (Ethereum), TEE / SEV-SNP); now 50 total entries
  • book/OUTLINE.md — Chapter 3 entry updated with subtitle and final section headings

No back-edits to prior chapters this time — Chapter 2's Hyperliquid corrections were already made in the previous turn, and Chapter 4's April 8 patch citation stands (the same hedge applies in both chapters).

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

Per production order (book/OUTLINE.md), all four "setup and voice-calibration" chapters (Chs 1, 2, 3, 4) are now drafted. The next chapter to tackle is Chapter 5 — The Validator (and the Builder) — the first of the actor chapters, building on the validator/builder cameos this chapter set up.