The House Edge

Appearance

Chapter 3 — Outline (Phase 2) ​

Status: OUTLINE (Phase 2 — written by agent, awaiting Nick's review) Date: 2026-05-14 Working chapter: 03 — The Mempool and What Replaces It Carries forward from: RESEARCH.md

Working assumptions (carried from Phase 1) ​

Defaults from RESEARCH.md §6, treated as binding for Phase 3 unless Nick overrides.

  1. Word budget: ~5,500 words including footnotes, near the top of the Bible's 4,000–6,000 target. The user's content directive (full Jito mempool history + BAM + Harmonic + Ethereum + L2s + Hyperliquid + censorship trade-off) genuinely needs the room.
  2. April 8 2026 Solana patch mechanism: cite the effect ("simple validator-leader sandwiches are largely suppressed") without overcommitting to the technical mechanism. Flag for later research.
  3. Worked example: Candidate A — Alice's $10,000 swap through public Solana RPC vs Jupiter Beam — as the anchor. Carlos (a new named character, an institutional Ethereum user) appears in the Ethereum subsection as the secondary worked example.
  4. Verdict tone: match Chapter 4's. The chapter has earned it.
  5. Constellation as closing beat: yes, named in the "What changes when…" transition, with a hedge ("a proposal, not a commitment").
  6. Solana Foundation enforcement (June 2024): paragraph inside the Solana subsection, not a full sidebar.
  7. Hyperliquid weight: cameo in the chain-comparison box. Full architectural treatment is Chapter 9.

Title and subtitle ​

Chapter 3 — The Mempool and What Replaces ItA surface, three responses, and the actor whose job is to read it.

Cold open ​

On Friday, 8 March 2024, Jito Labs — the firm whose software runs on most of Solana's stake — shut down its private mempool within hours of an internal decision, with no advance notice to the searchers who had built businesses on it. The mempool had been live for nineteen months; on the day before its shutdown, MEV tips on Solana exceeded ten thousand SOL, worth about $1.5 million at the prevailing price. Jito's stated reason was to "prioritize the network's long-term growth and stability, sacrificing significant revenue in the process." The reason that did not need to be stated was that the mempool had become the substrate for a sandwich-attack industry the firm could no longer comfortably host. This chapter is about the surface Jito turned off, the surfaces other chains have built or refused to build, and the actor whose job is to read whichever surface is available.

(Why this open: a dated, named, primary-sourced event with specific numbers; a punchline ("the reason that did not need to be stated"); a closing sentence that sets up the chapter's argument in the Ch 1 / Ch 2 / Ch 4 pattern.)

What this chapter answers ​

  • What is the public mempool, and why does its existence make sandwich attacks and most other extractive techniques possible?
  • What did Jito's mempool do, and why did Jito shut it down? What replaced it on Solana — Jupiter Beam, BAM, Harmonic, the architectural patches?
  • How do Ethereum and its L2s handle the same surface, and why has private order flow become the value-weighted majority of Ethereum traffic?
  • What is the trade-off between MEV-resistance and censorship-resistance, and what are FOCIL and Flashnet doing about it?

Section list ​

  1. Cold open — Jito's 8 March 2024 mempool shutdown.

  2. What this chapter answers — the four questions above.

  3. The setup (≈400 words). The public mempool defined precisely. The two questions the mempool's existence raises (who reads it, what can they do with what they read). The reader's already-installed vocabulary: Alice, the searcher (Ch 4), the prop-AMM operator (Ch 2). A short plant: this chapter is the hinge — after it, every other chapter can refer to "visibility" without explanation.

  4. The worked example (≈300 words). Alice's $10,000 USDC→SOL swap, in early 2026. She has two options for how to submit her transaction: the default public Solana RPC her wallet was configured with, or Jupiter's Ultra routing (which uses Beam, Jupiter's private-transaction-landing engine). The chapter shows what the two paths actually do at the visibility layer.

  5. The mechanics, in detail (≈3,400 words; three H4 subsections — under the chapter spec template's hard limit of four):

    • Solana: the mempool that wasn't, the mempool that was, and what replaced both (≈1,500 words). The full arc. Solana technically never ran a public mempool the way Ethereum did — Gulf Stream forwards transactions directly to slot leaders. But Jito-Solana, the dominant validator client, ran a private mempool from August 2022 that gave searchers a 200-millisecond preview window. On 8 March 2024 Jito shut it down. In June 2024 the Solana Foundation removed ~32 validators (~1.5M SOL, ~0.5% of delegation-program stake) for participating in private mempools that enabled sandwich attacks; Tim Garcia, the foundation's validator-relations lead, described the action as ongoing. The successor architecture has three layers: Jupiter's Beam at the wallet/router layer (default in 2026; +0.006% slippage on Ultra V3); BAM at the block-construction layer (live July 2025; AMD SEV-SNP TEE enclaves; ~28% of Agave stake on JitoBAM by March 2026); and Harmonic at the validator-selection layer (an open block-building marketplace where validators pick the most valuable block from multiple builders; ~17% of stake). Vpe — the sandwich operator named in Ch 4 — appears here as the concrete counterparty whose business model the architecture is trying to compress. The April 8, 2026 protocol-level patch is named with a hedge ("the simple validator-leader sandwich is largely suppressed; the technical mechanism is not publicly documented"). Constellation (the Helius MCP proposal) closes the subsection as the forward-looking architectural answer.

    • Ethereum: the public surface and its private replacement (≈1,200 words). The public mempool, properly developed: anyone running a node sees every pending transaction; gossip propagation is roughly milliseconds-to-seconds. Carlos — a new named character, an institutional Ethereum trader — submits a $10,000 swap through Flashbots Protect (changes his wallet's RPC URL to the Flashbots endpoint; transactions become bundles delivered directly to builders, bypassing the public path). The chapter develops the three private-routing services: Flashbots Protect (2.1M+ accounts; $43B+ shielded; ~7% of Ethereum monthly tx count), MEV-Blocker (4.5M+ wallets; 6,177 ETH cumulative rebates; transferred from CoW DAO to Consensys' SMG in Jan 2026), MEV-Share (Flashbots' selective-disclosure mechanism). The structural observation: 30% of Ethereum transactions by count but 54.6% of total block value now flows through private channels (Wahrstätter et al.). The MEV-Boost layer follows: the proposer/builder/relay separation, the relayscan.io May 2026 numbers (Titan 50.6% / BuilderNet 24.55% / Quasar 15.68% — top three ~91% of blocks), Flashbots' own December 2024 migration out of running a builder and into BuilderNet. The structural punchline: when 91% of blocks are built by three firms, the question "what's in the public mempool" matters less than "which firms have the relationships to put a transaction in front of those builders." That setup is Chapter 7.

    • L2s: the sequencer as a one-firm private mempool (≈700 words). The L2 surface is structurally different from L1: most L2s (Base, Arbitrum, Optimism) run a single sequencer operated by one firm, which sees every transaction before ordering it. Three case studies:

      • Base — Coinbase-operated sequencer; OFAC-filtered; H1 2025 revenue $42.4M (87.2% of Superchain sequencer revenue); ~86% of daily revenue is priority fees, which are the surface for any third-party backrunning that does happen.
      • Arbitrum — Offchain Labs operates the sequencer; Timeboost (activated April 2025) replaces first-come-first-served ordering with a sealed-bid 200ms-express-lane auction whose proceeds flow to the Arbitrum DAO; February 2026 saw searchers route around Timeboost's primary auction via a third-party resale market called Kairos.
      • Optimism Unichain — uses Flashbots-built Flashblocks (200ms block times) and returns MEV to users via revenue-share — the most aggressive MEV-back-to-user design currently live.

      The decentralised-sequencing alternative — Espresso Systems' mainnet went GA on 12 February 2026, with ~6-second finality; Astria's mainnet alpha launched in October 2024 on Celestia DA — exists but no major L2 had migrated to a decentralised sequencer as of May 2026. The visibility surface on L2s is, for now, a one-firm private mempool.

  6. How this plays out on each chain — and Hyperliquid as the alternative (≈400 words). The required chain-comparison box, expanded slightly to include Hyperliquid's architectural elimination. Three paragraphs:

    • Solana — the surface has been compressed (Jito mempool gone, Beam by default, BAM at the block-construction layer, April 8 patch). Sandwich activity persists on long-tail tokens and unprotected wallets but the major surface is now privately routed.
    • Hyperliquid — there is no public surface at all. HyperBFT consensus matches transactions and order-book actions atomically inside consensus, with semantic ordering (non-order actions, then cancels, then sends within a block). The mempool problem the chapter has spent most of its pages on does not exist on Hyperliquid; the structural choice eliminates the surface rather than managing it. Full architectural treatment in Chapter 9.
    • Ethereum and L2s — the public mempool still exists on L1 but holds the value-leftovers; the private channels carry 54.6% of block value. L2 sequencers are one-firm private mempools by construction. FOCIL (EIP-7805) is the consensus-layer answer being built; Flashnet is the submission-layer answer; both are in development, neither is live on mainnet as of mid-2026.

  7. Who wins, who loses, why (≈400 words). Adversarial verdict, matching Ch 4's tone.

    • Winners: Jito Labs (operator of the dominant private-routing infrastructure on Solana; 6% tip take on BAM); Jupiter (Beam's default routing position; ~50% of Solana DEX volume passes through it); the named Ethereum builders (Titan, BuilderNet, Quasar build 91% of blocks); Coinbase (Base sequencer; $42.4M H1 2025 revenue); Offchain Labs (Arbitrum sequencer + Timeboost auction proceeds); the Optimism Collective (returns 100% of sequencer profit). The searcher firms whose models survived the architectural shifts — they did not need a public mempool to do their job once private routing matured.
    • Losers: retail traders who continue to submit transactions through the default public RPCs without protection (the surface is smaller but not zero, especially for long-tail tokens); the searcher firms whose models depended specifically on the public mempool surface that Jito shut down (and the ~32 validators the foundation removed); the public-goods narrative of "anyone can run a node and see the network" — the value is increasingly in the private channels.
    • Is this bad? The clinical answer: the architectural responses are real, are working, and the dollar of slippage Alice paid in February 2025 would be substantially smaller today. But the new arrangement is also more concentrated — Jito, Jupiter, Titan, BuilderNet, Coinbase, Offchain Labs — and the censorship-resistance properties of "anyone can submit a transaction visible to everyone" have been traded for the MEV-resistance properties of "submit a transaction visible to one firm, who chooses who else to share it with." The trade-off is real. Whose side it cuts on depends on which risk the reader weighs more heavily.

  8. What changes when… (one paragraph). The transition. What changes when one of the firms that handles the private routing also runs the validator that includes the block? When the relationship between who submits the bundle and who chooses what goes in the block becomes a single-firm decision? That is Chapter 7's question. Constellation, FOCIL, and Flashnet are the protocol-level proposals being built against it; none has shipped on mainnet as of mid-2026.

  9. Footnotes and sources — numbered with URLs and access dates. Approximately 20 footnotes.

The worked example and where it threads through ​

Alice's $10,000 USDC→SOL swap is the chapter's anchor. Carlos, an institutional Ethereum trader making a $10,000 USDC→ETH swap, is the secondary in the Ethereum subsection.

SectionBeat
§3 (Setup)Plant: Alice and Carlos will both submit transactions; what happens at the visibility layer differs.
§4 (Worked example)Setup: Alice picks Solana via Jupiter Ultra; the chapter promises to show what Beam does at the visibility layer.
§5 SolanaPayoff for Alice: the public-RPC path (her transaction observable for the brief window before slot inclusion) vs the Beam path (private bundle delivered directly to Jito's block engine, no public exposure, +0.006% slippage). The dollar comparison: ~$50 slippage on the unprotected path vs effectively zero on the protected path.
§5 EthereumCarlos enters. He changes his wallet's RPC URL to Flashbots Protect's endpoint; his $10K USDC→ETH swap is now bundled with searcher refunds; if any value is extractable, he gets a share back. The chapter walks the mechanism concretely.
§5 L2sNeither Alice nor Carlos — the chapter describes the L2 surface in general terms.
§6 (Chain comparison)Hyperliquid: the same trade would have no pre-execution visibility at all.
§7 (Verdict)Alice and Carlos as the named losers in the "before" case; named winners in the "after" case (their trades are better-executed, but the structure of who makes that possible has concentrated).

Diagrams needed ​

Two diagrams.

  1. D1 — Three paths for Alice's $10,000 swap (Mermaid flowchart, three columns). Left: public Solana RPC → Gulf Stream gossip → slot leader's TPU (with the "private mempool window" Jito ran from 2022-2024 shown as an extinct path, faded). Middle: Jupiter Ultra → Beam → direct to Jito's block engine → slot leader (no public surface). Right: BAM-enabled flow → BAM Node TEE → encrypted bundle → validator (the AMD SEV-SNP enclave shown as the trust assumption). Each path annotated with what's visible to whom.

  2. D2 — Private order flow share by chain (markdown table). Rows: Ethereum L1, Base, Arbitrum, Solana (after BAM), Hyperliquid. Columns: public-flow share, dominant-private-routes, who-operates-them, structural-note. Many of these are imprecise (private flow on L2 is essentially 100% because the sequencer sees everything) — that's part of the chapter's point.

Glossary terms this chapter introduces ​

To be appended to GLOSSARY.md in Phase 3.

Defined in full (first appearance):

  • Public order flow — transactions broadcast through the public path (Ethereum's public mempool, Solana's public RPC), visible to anyone observing the network.
  • Private order flow — transactions delivered directly to a block builder, sequencer, or specialised TEE enclave, bypassing the public path.
  • Flashbots Protect — canonical Ethereum private-routing RPC; the user changes their wallet endpoint, transactions are bundled with refunds back from any backrun value captured.
  • MEV-Share — Flashbots' selective-disclosure mechanism that lets users share parts of a transaction with searchers in exchange for a share of the extracted value.
  • MEV-Blocker — Ethereum private-routing service originally by CoW DAO, transferred to Consensys' Special Mechanisms Group in January 2026.
  • MEV-Boost — the Ethereum proposer/builder/relay separation infrastructure; the standard way Ethereum validators outsource block construction to specialised builders.
  • Relay (Ethereum) — entity that receives bundles from builders and submits them to validators; named relays include Ultrasound, Titan Relay, BloXroute, Aestus, Flashbots Relay.
  • Block builder — already cameoed in Chs 2 and 4; full mechanical treatment lives here (visibility into private and public flow; competition for proposer attention; bundle merging).
  • Bundle (Jito-style) — already cameoed; full mechanics here.
  • BAM (Block Assembly Marketplace) — Jito Labs' TEE-based confidential block-construction layer for Solana; live July 2025.
  • TEE / SEV-SNP — Trusted Execution Environment; AMD's hardware enclave technology used by BAM Nodes to keep transaction contents encrypted until execution.
  • Harmonic — Solana validator-selection-layer protocol; an open block-building marketplace where validators continuously pick the most valuable candidate block from multiple builders.
  • Inclusion list / FOCIL (EIP-7805) — Ethereum's proposed fork-choice-enforced inclusion-list mechanism for protocol-level censorship resistance.
  • Flashnet — Flashbots' network-anonymized mempool design (Feb 2026); a submission-layer privacy primitive complementary to FOCIL.

Cameo (one inline sentence; full treatment elsewhere or already in glossary):

  • Gulf Stream (Solana's pre-leader transmission path; already cameoed in Ch 1)
  • HyperBFT (already cameoed)
  • Sequencer (cameoed here; full treatment Ch 10)
  • Timeboost (Arbitrum-specific; one inline mention)
  • Espresso, Astria (decentralised sequencing services; one inline mention each)

Backward:

  • Chapter 1 (drafted): defined intent and cameoed the mempool. Ch 3 develops the surface Ch 1 named.
  • Chapter 2 (drafted): prop-AMMs and solver markets depend on visibility into pending flow. Ch 3 explains the surface they read from.
  • Chapter 4 (drafted): the searcher is the actor whose job is to read this surface. Vpe, named in Ch 4, returns here as the canonical operator the surface enabled.

Forward:

  • Chapter 5 (Validator and Builder, not yet drafted): the actors who actually decide what goes in a block. Ch 3 names them in cameo; Ch 5 develops them.
  • Chapter 6 (Infrastructure Layer, not yet drafted): Jito, Flashbots, Coinbase as the firms operating these surfaces. Ch 3 names the institutions; Ch 6 develops the business.
  • Chapter 7 (Exclusive Order Flow, not yet drafted): the next step when private routing dominates and one firm has a monopoly on a slice of flow. Ch 3's verdict explicitly tees this up.
  • Chapter 9 (Hyperliquid): full architectural treatment of the no-mempool design.
  • Chapter 11 (Who's at a Disadvantage): the censorship-vs-MEV-resistance trade-off plays into the chronic-loser thesis.
  • Chapter 12 (Where This Is Going): Constellation, FOCIL, Flashnet are the protocol-level answers being built — full forward-looking treatment in Ch 12.

Phase 2 is complete. Per the user's compressed-review pattern, Phase 3 begins immediately.