The House Edge

Appearance

Chapter 12 — Outline (Phase 2) ​

Status: OUTLINE (Phase 2 — written by agent, accompanying RESEARCH.md) Date: 2026-05-14 Working title: Where This Is GoingWorking subtitle: Three forward-looking trends that reshape where the extraction happens — and what they do not change.

Cold open (1 paragraph) ​

In January 2026, the open interest of all HIP-3-deployed perpetual markets on Hyperliquid sat at approximately $280 million. Four months later, in early May 2026, it stood at approximately $2.50 billion — a roughly nine-fold increase in a quarter, with one operator (TradeXYZ) holding more than 90% of the open interest and approximately $12.7 billion of cumulative volume.[^cold-1] In the same week, Ethereum's twelfth devnet for the Glamsterdam upgrade — the fork that is supposed to bring in-protocol proposer-builder separation (ePBS, EIP-7732) into Ethereum L1 — entered yet another iteration cycle, with the Ethereum Foundation's April 2026 Checkpoint #9 stating that the implementation has proved "trickier than anticipated" and that activation in Q2 2026 was now considered unlikely. In the same month, Astria sunset its mainnet shared-sequencer network at block 15,360,577 after a year of operation; Espresso, the surviving named operator in the shared-sequencer category, continued to expand toward its 25 MB/s throughput target. Solana's Alpenglow consensus upgrade entered live validator testing on 11 May 2026, with mainnet activation targeted alongside Agave 4.1 in late 2026 — bringing a tenfold drop in the profitable-validator stake threshold (from approximately 4,850 SOL to approximately 450 SOL). The book has documented three chain architectures and the categorical concentration each produces. This chapter is about the three forward-looking trends — intent-based architectures, appchains and sequencers, and cross-chain execution layers — that are supposed to reshape that picture over the next three to five years. The chapter's claim is that they will indeed reshape where the extraction happens. They will not change whether it concentrates.

Section list (one-sentence summaries) ​

1. What this chapter answers ​

Four questions:

  • What are the three forward-looking trends that will reshape on-chain trading over the next three to five years?
  • What does each trend change about the extraction surface the book has documented?
  • What does each trend not change about the operator concentration the book has documented?
  • Do any of the trends materially address the consent gap Chapter 11 raised, and if not, what does the book leave the reader to consider?

2. The setup ​

The chapter's organising frame: the book has documented architecture as the variable that produces value-capture shape, and operator concentration as the variable that does not vary across architectures. Chapter 8 documented Solana's vertical integration as compressed-surface / multi-firm-stack; Chapter 9 documented Hyperliquid's in-consensus matching as eliminated-surface / four-locus re-concentration; Chapter 10 documented Ethereum's PBS + L2 + restaking trifecta as dispersed-along-architectural-lines / concentrated-along-operator-lines; Chapter 11 documented the four chronic-loser categories and the consent gap. The forward-looking trends the chapter develops — intent-based architectures, appchains and sequencers, and cross-chain execution layers — are three more architectural choices that will reshape the surface but, the chapter argues, will not address the structural concentration. The chapter also engages with two protocol-level events on the immediate roadmap: ePBS (Ethereum's in-protocol PBS, slipping past Glamsterdam's H1 2026 target with realistic Q3/Q4 2026 timing) and Alpenglow (Solana's consensus upgrade, in live validator testing as of 11 May 2026 with mainnet expected with Agave 4.1 in late 2026). The chapter previews three structural findings. First: intent-based architectures deepen the consent gap by abstracting the execution path entirely from the user, while the solver market that wins the intent concentrates (SCP + Wintermute >90% of UniswapX; Barter targeting >50% of CoW Swap via the Copium acquisition). Second: appchains and sequencers relocate the moat from a relationship-based gate (Ethereum builder-direct contracts) to a capital-based gate (HIP-3's 500K HYPE deployment stake; HIP-4's 1M HYPE Phase 2 stake); the operator concentration follows the moat. Third: cross-chain execution layers compress the cross-chain MEV surface (the ~$10M/year Maire et al. surface from Chapter 10) but concentrate operator participation in the relayer / bridge-protocol / resolver layers, with LayerZero handling approximately 75% of cross-chain bridge volume and Across's top-5 relayers handling approximately 70% of its protocol's volume.

3. The worked example ​

Per RESEARCH.md recommendation, the chapter uses three short hypothetical traces — one per trend — rather than a single dollar-traced anchor (the forward-looking nature of the chapter makes a specific dollar trace illustrative rather than empirical):

Trace A — Alice's intent. Alice signs an intent: $10,000 USDC → SOL at minimum acceptable rate of 0.04 SOL/USDC over 30 seconds. The intent is broadcast to the UniswapX solver auction. SCP and Wintermute compete; SCP wins with a quoted price that beats the AMM-routed alternative by approximately 5 basis points. Alice's all-in cost: approximately ~$3–5 on the $10K notional. What Alice did not sign: the routing path through SCP's inventory, the cross-chain leg (if the trade goes through ETH bridge inventory), the per-layer trace of who captured what. The chapter develops this as the cleanest case of the deepened consent gap.

Trace B — Carlos's HIP-3 deployment. Carlos (the trader returning from Chapters 3 and 11; now hypothetically a small VC-backed protocol team) wants to launch a niche tokenized-asset perpetual market — say, a perpetual on a specific tokenized commodity index. He cannot afford the 500K HYPE deployment stake (~$25M at $50 HYPE). His options: (a) accumulate the stake (years of operation if not VC-funded); (b) wait for protocol-level changes that lower the threshold (no roadmap commitment exists); (c) partner with an existing HIP-3 deployer (TradeXYZ or one of the named operators) for sub-market deployment — but the structural economics flow to TradeXYZ's stake-moat. The chapter develops this as the appchain-moat structural finding: appchains are permissionless-by-stake-not-by-permission, but the stake is the moat.

Trace C — A cross-chain searcher's reallocation. A cross-chain MEV searcher who was capturing approximately 1% of the ~$10M/year cross-chain arbitrage surface (Maire et al., Chapter 10) sees ERC-7683 adoption rise to approximately 88% of Across's volume and Optimism's Superchain native interop ship to devnet. The searcher's surface compresses. The searcher reallocates capital toward becoming an Across relayer or a 1inch Fusion+ resolver — the same skill set, the same capital, the same firm, operating in a different architecture. The chapter develops this as the structural reading: the cross-chain MEV surface compresses, but the operators capturing it move to the new architecture rather than exit the market.

4. The mechanics, in detail ​

Three H4 subsections, one per trend, plus a fourth H4 for the two protocol-level events.

4a. Intent-based architectures — what they change and don't change ​

Intent-based architectures separate what the user signs (a desired outcome) from how the trade is executed (chosen by a solver competing against other solvers). The mechanism is well-defined: the user specifies the trade's destination (e.g., "$10,000 USDC → SOL at a minimum acceptable rate of 0.04 SOL/USDC"), signs the intent, and broadcasts it; solvers compete to fulfil it within a time window; the winning solver settles the trade and pays the protocol fee. UniswapX, CoW Swap, 1inch Fusion, and 1inch Fusion+ (cross-chain) are the four canonical 2026 single-chain-and-cross-chain cases. Anoma's mainnet (Phase 1 launched 29 September 2025 on Ethereum) is the most explicitly intent-first architecture in production, with the XAN token live, governance active, and protocol adapters planned for Arbitrum, Optimism, Base, then Solana and Bitcoin.

The structural argument the chapter develops: the solver market that wins the intent is itself concentrating, in the same shape as Chapter 7's exclusive-flow finding one architectural level down. On UniswapX, SCP and Wintermute (operating as Rizzolver) make up more than 90% of volume; recent dynamics show Barter overtook Wintermute as the No. 1 UniswapX solver during the second half of 2025, but the duopoly structure is intact. On CoW Swap, Barter is consolidating around a target of >50% market share via the Copium Capital solver-codebase acquisition (mid-2025; Barter held ~28% pre-acquisition, executed more than $18B total volume, ~$900M weekly average). The structural reading: the firms that won the public-mempool MEV market (Chapters 4–6) and the exclusive-flow market (Chapter 7) are also winning the intent-solver market — the same capital, the same technology, the same access requirements.

The structural finding the subsection lands: intent-based architectures deepen the consent gap relative to Chapter 11's framing. A retail trader routing through Phantom + Jupiter + Beam does not see the per-layer trace (Helius rebate, Jito tip, Harmonic Performance validator capture); the same trader signing a UniswapX intent does not see any execution path at all. The compression at the trade level holds; the informational asymmetry widens. The chapter develops this without moralising: intents change what the user agrees to, not who captures the residual; the solver-market-as-concentrating-supplier is the same concentration pattern documented at every other layer of the book; the consent gap is now structural because the user cannot see what they signed away because they did not sign anything specific. 1inch Fusion+ extends this pattern to cross-chain swaps; the resolver concentration is comparable.

4b. Appchains and sequencers — what they change and don't change ​

The canonical 2026 case of the appchain pattern's relocation of value capture into the appchain operator is HIP-3 on Hyperliquid (Chapter 9 anchor). Mainnet launched 13 October 2025 with a 500,000 HYPE (~$25M) deployer stake; 50/50 deployer/protocol fee split; 2× user fee on HIP-3 markets vs validator-operated markets. Updated as of May 2026: aggregate 30-day HIP-3 trading volume approximately $2.82–$2.86 billion across ~17,630 unique traders and ~1.25 million trades; HIP-3 markets represent more than 35% of all Hyperliquid trading volume; HIP-3 network-wide open interest rose from approximately $280 million on 1 January 2026 to approximately $2.50 billion in May 2026, peaking at $2.47 billion and reaching $1.43 billion in March 2026. TradeXYZ retains more than 90% of HIP-3 open interest (~$12.7 billion cumulative volume per Chapter 9). HIP-4 outcome markets launched on 2 May 2026 and extends the pattern with a 1 million HYPE Phase 2 deployer stake (2× the HIP-3 threshold).

The structural argument the chapter develops: HIP-3 is permissionless-by-stake-not-by-permission; the 500K HYPE deployment stake is itself the moat; the moat is the same shape as the Ethereum builder-direct relationships (Wu et al.'s 75 EOF arrangements with ~71% of trading-related builder revenue concentrated per Chapter 7). The appchain pattern doesn't eliminate the concentration; it relocates the moat from a relationship-based gate (Ethereum) to a capital-based gate (HIP-3). HIP-4's 2× threshold makes the moat steeper. The structural reading: appchains relocate the contest into the appchain operator's economics, and the appchain operator's economics are stake-moated to institutional or VC-funded teams.

The L2 sequencer / shared-sequencer category completes the appchain trend story. The Optimism Superchain is shipping native interop on the devnet path; production rollout is expected in 2026 alongside a shared sequencer (likely Espresso). Espresso launched mainnet on 12 February 2026; HotShot consensus achieves sub-6-second finality with a planned throughput upgrade from 5+ MB/s to 25 MB/s; integrations are in progress with Arbitrum, Polygon, and Optimism. Astria intentionally halted at block 15,360,577 after one year of mainnet operation and rolled back its Flame integration; the shared-sequencer category consolidated to one named operator within 2026. The structural reading the chapter develops asymmetrically: the shared-sequencer concept is the architectural answer to L2 fragmentation; the firms shipping it are concentrating (one survivor in the category Chapter 10 anchored). Castro et al.'s September 2025 measurement of Arbitrum Timeboost (two firms winning >90% of priority slots) is what the shared-sequencer model would generalise across multiple L2s.

The SVM appchain landscape gets a one-paragraph treatment. Eclipse retrenched in late 2025 (95% TVL decline from peak; ~65% staff reduction; pivot from neutral infrastructure to in-house studio); Anza is modularising the SVM into standalone components over 2026, with the modular SVM allowing developers to run independent SVM implementations; DoubleZero Edge launched on 16 April 2026 delivering high-speed data infrastructure to Solana validators. The chapter's reading: SVM appchain expansion is happening but is not a 2026 mass-market story; the more visible 2026 appchain case is HIP-3, not an SVM appchain.

4c. Cross-chain execution layers — what they change and don't change ​

The cross-chain bridge / execution layer has consolidated to a 2–3-firm market in 2026, with the same concentration pattern documented at every other layer. LayerZero handles approximately 75% of cross-chain bridge volume: $293M average daily transfers; 1.2M messages/day; $225.4B lifetime across 168 chains; LayerZero V2 OFT standard processed more than $166.9B in cross-chain transfers across 733+ omnichain tokens; 160+ endpoints with OFT-based assets totalling $90B+. LayerZero acquired Stargate (August 2025 announcement; closed early 2026); Stargate revenue now flows to ZRO buybacks. Chainlink CCIP grew 1,972% in 2025 to $7.77B and connects 60+ chains while securing $33.6B; Kraken adopted CCIP as its cross-chain standard on 14 May 2026 (replacing LayerZero — one day before this chapter is written). Wormhole holds $60B lifetime volume and was named the only unconditionally approved cross-chain protocol by Uniswap DAO. deBridge processed approximately $12.5B over 63 weeks (5.6M transactions, ~1M new users as of December 2025). Across filled approximately $14B+ across 12.4M cross-chain transfers as of April 2026 (median 8-second fill, 99.7% success, 15+ chains).

The structural argument the chapter develops: the volume is concentrating into 2–3 named firms across all major routes. The intent-based cross-chain pattern (Across, 1inch Fusion+, UniswapX cross-chain, ERC-7683) is the cleanest 2026 shape of cross-chain execution. ERC-7683 (Uniswap Labs + Across; ratified early 2025) is now implemented by Across, UniswapX, CoW Protocol, and Eco; wallet support includes Safe, Argent, Rabby, MetaMask; major L2 support across Arbitrum, Optimism, Polygon, Base; the Open Intents Framework launched February 2025 with 30+ teams. Across migrated its production solver network to ERC-7683 in Q3 2025; approximately 88% of Across volume now uses ERC-7683 orders. Across has 40+ active relayers but the top 5 handle approximately 70% of volume — because optimistic settlement ties up relayer capital for hours per order, concentrating participation among well-capitalised actors. The structural finding: ERC-7683 adoption is the cleanest single piece of evidence that the cross-chain intent pattern is winning the architectural argument; the relayer-concentration data is the cleanest single piece of evidence that the firms operating that pattern are concentrating.

The cross-chain MEV surface (Maire et al., Chapter 10 anchor) is a measured ~$10M/year (Sep 2023 – Aug 2024; 242,535 arbitrages; 5.5× activity growth; top-5 addresses execute >50% of trades; one address ~40% of daily post-Dencun volume). The shared-sequencer / atomic-composability solutions compress this surface; the cross-chain intent protocols compete with searchers for the surface; the relayer-concentration pattern shows the surface compression does not produce operator competition. The chapter's structural reading: the cross-chain extraction surface is real but small relative to intra-chain MEV; the cross-chain execution layer compresses the surface and concentrates the operators capturing it.

4d. ePBS and Alpenglow — two protocol-level events on the immediate roadmap ​

ePBS (EIP-7732, the in-protocol replacement for MEV-Boost's out-of-protocol PBS) is slipping past Glamsterdam's nominal H1 2026 target. Per the Ethereum Foundation's April 2026 Checkpoint #9: ePBS implementation is "trickier than anticipated"; the first generalized Glamsterdam devnet is pending stabilisation of the ePBS-only devnet; multi-devnet iteration is required before client releases, security reviews, and testnets. Justin Drake's "Strawmap" (January 2026) outlines seven forks through 2029 on six-month cadence: Glamsterdam (Q2/Q3 2026); Hegotá (Q1/Q2 2027, with FOCIL as headliner via EIP-7805 and account abstraction via EIP-8141 in the minor feature set); then five more through 2029. Vitalik Buterin has stated EIP-8141 ships within a year of March 2026. The structural argument: ePBS replaces MEV-Boost's de facto out-of-protocol PBS with in-protocol PBS, addressing the relay-as-trusted-intermediary risk (Chapters 5 and 6); it does not address the builder-side concentration (Titan ~52% / BuilderNet ~25% / Quasar ~15% per Chapter 10) or the exclusive-flow contracts (Wu et al.'s 75 EOF arrangements per Chapter 7). ePBS dents one category of trust assumption; it does not dent the concentration the book has documented.

Alpenglow is in live validator testing as of 11 May 2026; mainnet activation is expected with Agave 4.1 in late 2026. Per Chapter 8's anchor updated: SIMD-0326 passed with 98.27% validator approval (2025); Votor moves vote consensus off-chain via direct messaging and signature aggregation (vote transactions previously consumed >50% of block throughput); the Validator Admission Ticket is a 1.6 SOL per-epoch fee; targeted block confirmation ~150ms (potentially ~100ms under favourable conditions); the profitable-validator stake threshold drops approximately 10× (from ~4,850 SOL to ~450 SOL per Chapter 8). Anatoly Yakovenko has framed the upgrade as evidence that "Solana's design is working." The structural argument: Alpenglow widens the economic floor (the SFDP-removed-validator category Chapter 11 documented becomes economically viable again at ~450 SOL); it does not address the access-vs-operational gap (+33–101% access signal vs ~+3% operational signal per Chapters 5, 7, 8). A small-stake operator who passes the new economic floor still operates without the BAM Node access, the Harmonic strategy enablement, or the Frankfurt colocation. Alpenglow restores the bottom; it does not narrow the top.

5. The closing structural argument ​

The chapter's load-bearing claim: the three forward-looking trends reshape where the extraction happens, not whether it concentrates.

  • Intent-based architectures move the contest from the trade-execution layer to the solver-market layer (UniswapX SCP/Wintermute >90%; CoW Swap Barter targeting >50%; same firms across all venues).
  • Appchains and sequencers move it from the general-purpose chain to the single-application chain (HIP-3 TradeXYZ >90% of OI; the deployer-stake-as-moat shape; HIP-4 raising the moat by 2×).
  • Cross-chain execution layers move it from intra-chain MEV to cross-chain execution (LayerZero ~75% of bridge volume; Across top-5 relayers ~70% of volume; CCIP growing 1,972%).

For each trend, the chapter develops the symmetrical answer: what changes (the surface), what doesn't change (the concentration). The consent gap from Chapter 11 is deepened by intents (the user signs an outcome, not a path), relocated by appchains (the user sees the appchain operator but not its stake-moat economics), and partially compressed by cross-chain atomic execution (the surface shrinks) — but in every case, the operator concentration the book has documented persists in the new architecture. The book's structural finding holds: architecture choices change the actor shape; they do not change the concentration shape.

The two protocol-level events on the immediate roadmap reinforce the finding: ePBS addresses one category of trust assumption without addressing the builder oligopoly; Alpenglow widens the economic floor without narrowing the access-vs-operational top. Neither materially addresses the consent gap.

The chapter's tentative answer to Chapter 11's central question — do any of these trends materially address the consent gap? — is no, in retail's case; partially, in the LP / market-maker / validator cases where informed consent already exists. The chapter does not predict whether this will produce policy change, technology change, or behavioural change. The chapter offers the structural reading and the data; the verdict belongs to the reader.

6. What changes when… (closing) ​

The book ends here. The Epilogue — written separately, not by this agent — closes with a note on who commissioned the book and why. The chapter's closing observation: a reader who has come through twelve chapters now has the structural map. What she does with it — whether she participates differently, advises differently, regulates differently, builds differently — is a function of her position and her reading of the data, not of any injunction the book has issued. The chapter offers the comparison and the structural finding; the verdict belongs to the reader.

7. Footnotes and sources ​

22–24 numbered citations. URLs + access dates. Cross-references to Chs 2, 5, 7, 8, 9, 10, 11 marked "Already cited in Chapter X" where source previously in-book.

Worked example chosen — and where it threads ​

Three short hypothetical traces (per RESEARCH.md recommendation), one per trend. The forward-looking nature of the chapter makes hypothetical traces appropriate. The three traces are deployed in §3 as quick illustrations and referenced again in §4a (Alice's intent), §4b (Carlos's HIP-3 deployment thought experiment), and §4c (the cross-chain searcher's reallocation).

Diagrams needed ​

One Mermaid diagram and one summary table.

  1. D1 — The three trends and what they change vs don't change (Mermaid flowchart TB or summary table): three rows (intents / appchains / cross-chain) × three columns (architectural change; consent-gap effect; operator concentration). Placed in §5 (closing argument) as the chapter's central illustration.

  2. D2 — The ePBS + Alpenglow roadmap timing table (inline Markdown table): Ethereum forks per Strawmap (Glamsterdam, Hegotá, then five more); Solana Alpenglow + Agave 4.1 timing. Placed in §4d.

Glossary terms this chapter introduces ​

Two to three new entries (the chapter is largely synthesis using already-defined terms):

  • Intent (deepened framing) — the chapter updates the existing "Intent" glossary entry to include the deepened-consent-gap framing.
  • HIP-4 (Hyperliquid outcome markets) — extends the HIP-3 pattern; 1M HYPE Phase 2 deployer stake; launched 2 May 2026.
  • ERC-7683 (cross-chain intent standard) — Uniswap Labs + Across; ratified early 2025; implemented by Across, UniswapX, CoW, Eco; ~88% of Across volume by April 2026.

Cross-references: solver, intent, HIP-3, HLP, EigenLayer, ePBS, Alpenglow, VAT, L2 sequencer, cross-chain MEV, Timeboost, Espresso, builder-direct relationship, exclusive order flow — all already defined.

Backward (chapters this builds on):

  • Chs 4–6 — solver economics, infrastructure firms; the public-mempool MEV market the intent pattern competes with
  • Ch 7 — exclusive-flow concentration (Wu et al. 75 EOF); the solver-layer concentration framing
  • Ch 8 — Alpenglow as Solana's consensus upgrade; validator economics
  • Ch 9 — HIP-3 mainnet launch; HLP; the Hyperliquid validator set
  • Ch 10 — L2 sequencer economy; cross-chain MEV (Maire et al.); ePBS roadmap; EigenLayer
  • Ch 11 — consent gap; chronic-loser categories; the structural argument the chapter must engage with

Forward (epilogue):

  • The Figment reveal; the implicit pitch; the call to action

Chapter-level voice and structure notes ​

  • Forward-looking chapter as book-closer. The chapter's primary job is to land the "architecture choices change the actor shape; they do not change the concentration shape" closing argument and to engage with the consent-gap question Chapter 11 raised.
  • No new "Meet the actor" sidebar. Three short hypothetical worked examples replace the named-trader pattern of Chs 1–10.
  • Word budget: 5,500–6,500 words inc. footnotes. Substantial enough to develop the three trends and the two protocol-level events with empirical anchors; lighter than Chs 5–7 because the chapter is synthesis-plus-projection rather than concept-development.
  • Footnote count target: 22–24. Most are primary-source citations to research papers, protocol docs, and 2025–2026 news coverage; cross-references where appropriate.
  • Voice anchors: the "where vs whether" framing is the through-line. Each trend subsection ends with the symmetrical answer (what changes; what doesn't change).
  • Banned-move audit at draft time: no predictions beyond 3–5 years with anything but hedged language; no "future of finance" / "next big thing"; no specific price predictions or yield forecasts; clinical not predictive; specific firms named everywhere.
  • The chapter's editorial contribution: the closing structural argument that architecture choices change the actor shape without changing the concentration shape. This is the book's final structural finding before the Epilogue.

Phase 2 is complete. Phase 3 (DRAFT.md) follows immediately per the user's compressed-review pattern.