The House Edge

Appearance

Chapter 11 — Outline (Phase 2) ​

Status: OUTLINE (Phase 2 — written by agent, accompanying RESEARCH.md) Date: 2026-05-14 Working title: Who's at a Disadvantage, and Why They Don't MoveWorking subtitle: Four chronic-loser categories across three chain architectures — and the consent gap that keeps the table populated.

Cold open (1 paragraph) ​

In May 2025, the Pump.fun launchpad — Solana's dominant memecoin-issuance venue — reported that approximately 30.1% of all trading wallets ended a measured period profitable. Twelve months later, in April 2026, that figure stood at approximately 73.3%. The platform appeared, by the most-cited measure of retail outcomes, to have produced a roughly 2.4× improvement in retail profitability over the year. The number is true. The framing is misleading. Over the same period, Pump.fun's monthly active wallet count collapsed from approximately 5.2 million to approximately 1.8 million — a 65% exodus.[^cold-1] The platform's "improvement" in retail-trader outcomes was not produced by losers becoming winners. It was produced by losers leaving the table. The wallets that did not leave were, on average, the more skilled ones; the remaining 73.3% profitability rate reflects survivorship, not learning. This chapter is about the trader, the LP, the market maker, and the validator who didn't leave the table — and the structural reasons they didn't. Part III documented three chain architectures (Solana's vertical integration that compressed the extraction surface; Hyperliquid's in-consensus matching that eliminated it; Ethereum's PBS + L2 + restaking trifecta that dispersed it). This chapter documents the four chronic-loser categories that exist across all three: retail traders unaware of toxic flow, passive LPs in informed-flow pools, slow market makers, and validators without infrastructure relationships. The question is not whether each category loses — that is the empirical anchor of the prior ten chapters. The question is why each category continues to participate anyway, and why "just leave the table" is harder than it sounds.

Section list (one-sentence summaries) ​

1. What this chapter answers ​

Four questions:

  • Who are the four chronic-loser categories on-chain, and how do their losses appear in the data Chapters 1–10 documented?
  • Why does each category continue to participate, given that the structural disadvantages are documented?
  • Where is the consent gap — the difference between informed participation in a known disadvantageous structure and uninformed participation in a structure the participant does not understand?
  • What does "just leave the table" actually look like for each category, and why is that exit path harder than the framing suggests?

2. The setup ​

The chapter's organising frame: participation is not the same as consent. A passive LP who deposits into a Uniswap V3 ETH/USDC pool knowing that the LVR framework predicts a roughly 5–7% annual loss to informed flow has consented to a known structural disadvantage in exchange for accepting volatility-as-payment. A retail trader on a Solana memecoin venue who believes they are competing against other retail traders on equal terms, when they are in fact trading against prop-AMM operators with privileged routing visibility, has not consented to that structural disadvantage because they do not know it exists. The four loser categories the chapter develops differ along two axes: (a) the magnitude of the structural loss they bear; (b) the informational gap between the loss as experienced and the loss as understood. The chapter previews three structural findings. First: the loser categories are not symmetric. Retail traders bear small per-trade losses across high transaction counts; passive LPs bear continuous LVR drag on slow-moving capital; slow market makers face displacement (loss of the competitive position itself, not loss-per-trade); validators without infrastructure relationships bear opportunity costs in the form of unrealised top-quintile revenue rather than direct losses. Second: each category's exit path is structurally constrained. Retail's alternative is centralised exchanges in their jurisdiction (often unavailable, particularly in the US); LPs' alternative is depositing nowhere (the LVR-aware LP is rare in 2026 retail behaviour); market makers' alternative is pivoting to prop-AMMs (an option only available to firms with prop-AMM-grade capital and technology); validators' alternative is exiting the network (which forfeits the staking yield itself). Third: the consent gap is the chapter's clinical core. The chapter does not argue that the structures are illegitimate — three of the four loser categories include participants who knew the structure when they entered it. The chapter argues that the consent gap is largest for retail, and that the structural mechanisms that produce this gap (UX abstraction, information asymmetry between platform and user, the absence of mandatory disclosure of toxic-flow routing) are the chapter's load-bearing finding.

3. The worked example ​

Three short worked examples, one per category-with-named-actors. The slow market maker category is developed at the category level rather than via a single named firm (per RESEARCH.md recommendation).

Worked Example A — Carlos, the Pump.fun chronic loser. Carlos (introduced in Ch 3 as the Ethereum case-study trader, but reused here as a generic Solana retail name) is one of the approximately 1.8 million wallets still trading on Pump.fun in April 2026. Per the platform's survivorship data, Carlos is statistically more likely to be profitable than the average 2025 user — but only because the unprofitable 65% have left. Carlos's structural position: trading against prop-AMM operators routing through Helius and Jupiter Beam (Ch 8); pays the cumulative ~$8–15 take per $10K notional (Ch 8 Alice trace); does not know that approximately ~$0.50–$1 of his per-swap cost is captured at the Helius/Phantom rebate layer and split 50/50 (Ch 7) without his explicit consent. Carlos's continued participation is explained by: (a) UX advantage (Phantom is the easiest crypto wallet experience available); (b) absence of equivalent CEX access (he is US-domiciled and most major CEXes that allow Solana memecoin trading are no longer fully available to US retail post-2024); (c) the BIS Bulletin 69's "73–81% of retail crypto investors lose money" finding has not produced a behavioural change in Carlos's category because the framing of the loss is per-period rather than per-trade.[^a]

Worked Example B — Bob's Uniswap V3 LP revisited. Bob (introduced in Ch 2 as the LP case) deposited $50,000 into the ETH/USDC 5-bps pool on Uniswap V3 in late 2024. By May 2026, Bob has earned approximately $4,500 in fees and incurred approximately $7,200 of LVR-driven impermanent loss against the equivalent buy-and-hold benchmark — a roughly $2,700 net loss on the position. Bob's behaviour is informed: he read the Milionis et al. LVR paper after his second quarter of net-negative performance, and he understands the structural mechanism. Bob's continued participation is explained by: (a) the absence of a better LP venue in the size of capital he wants to deploy passively; (b) the "fees-earned" framing on Uniswap's UI which displays gross fees prominently and LVR not at all; (c) the structural fact that withdrawing the position crystallises the LVR loss as realised, where leaving the position open allows Bob to mentally treat the loss as unrealised. The HLP on Hyperliquid (Ch 9) is the structurally different passive-LP-equivalent — depositors who consented to a clearinghouse-default-fund-style role and were JELLY-tested on 26 March 2025 with a 65% TVL drawdown. The HLP depositors are informed in a way Bob is not; the structural exposure is comparable.[^b]

Worked Example C — Erika, the SFDP-removed Solana validator. Erika operates a small Solana validator with approximately 8,000 SOL of external stake plus 25,000 SOL of SFDP delegation. In April 2026, the Foundation announced that approximately 150 validators would lose SFDP delegation under the May 2026 ASN-concentration and data-centre-concentration rules (Ch 8). Erika's data centre — a smaller European facility outside the Frankfurt-Amsterdam axis — exceeds the new 15% concentration threshold; her SFDP stake will be revoked. Her structural position: stock Agave-Jito client; no BAM Node relationship; no Harmonic Performance strategy; no Frankfurt or NY5 colocation; per Syndica's March 2026 telemetry, her per-block priority-fee capture sits approximately at or below network median (Ch 8). Post-SFDP revocation, her per-block revenue falls to the ~6,000 SOL minimum economically viable threshold; she has 30 days to either acquire enough external stake to remain in the Foundation-stake program or wind down. Erika's continued participation is explained by: (a) the sunk-cost / hardware-asset framing (her one-time $12,000 colocation CapEx); (b) Alpenglow's expected late-Q3/early-Q4 2026 mainnet activation with VAT dropping the profitable-validator threshold from ~4,850 SOL to ~450 SOL — if she can survive the SFDP transition, the post-Alpenglow economics may restore viability; (c) the structural fact that exiting forfeits the validator-set membership entirely, where remaining preserves the option value.[^c]

4. The mechanics, in detail ​

Four H4 subsections, one per loser category.

4a. Retail traders unaware of toxic flow ​

The empirical anchors: BIS Bulletin 69 (Aramonte / Huang / Schrimpf 2022) documented that approximately 73–81% of retail crypto investors lose money across a 95-country study covering 2015–2022, with average per-trader losses of approximately 47.89%. Pump.fun's survivorship-driven "improvement" — 30.1% → 73.3% profitable wallets coinciding with a 65% MAU collapse — is the cleanest single 2026 data point on the "losers leave, platform metrics improve, structural losses persist for remaining participants" pattern. Banana Gun's user base — approximately 1.3 million users at $635 average trade size across 25.3 million cumulative trades — provides the L1 Ethereum analogue: the bot is in active use by retail-equivalent users, and those users are paying Titan ~17.75% margin (Ch 7) without explicit disclosure. The FCA's 2025 UK retail-crypto-ownership study documented that ownership fell from 12% to 8% of UK adults over 2023–2025, with disclosure-rule changes (Financial Promotions Rules October 2023) producing the participation reduction but not the per-trade loss reduction — the wallets that remained were not better at trading, they were simply the harder-to-deter subset. Carlos's continued participation is the structural finding: UX advantage + CEX inaccessibility + per-period-rather-than-per-trade loss framing keeps the chronic-loser category populated. The chapter's clinical observation: the consent gap is largest here because the loser does not have access to the per-layer dollar trace that Chs 8 and 10 documented for Alice — Carlos sees the Phantom UI showing his SOL balance, not the $0.50–$1 Helius rebate split or the +101% Harmonic Performance validator-side capture.

4b. Passive LPs in informed-flow pools ​

The empirical anchors: Milionis, Moallemi, Roughgarden, and Zhang (2022) formalised the Loss-Versus-Rebalancing framework — the result that a passive AMM LP earns approximately 5–7% of deposited capital per year less than the equivalent constant-mix portfolio rebalanced at oracle prices, for high-volume pairs like ETH/USDC and BTC/USDC. Topaze Blue's 2022 study found that approximately 49.5% of Uniswap V3 LPs were net-negative versus their buy-and-hold alternatives across the studied period — a finding that has been confirmed and tightened by subsequent empirical updates through 2025. Uniswap Labs's own 2024 analysis found that the V3 ETH/USDC 5-bps tier underperformed the V2 equivalent by approximately 68% on a fees-vs-IL basis, surprising the protocol's own team. The structural mechanism: the LP earns the spread the AMM curve embeds, but pays adverse selection to informed flow (CEX-DEX arbitrageurs, sandwich attackers in the pre-2026 mempool era, and the prop-AMM internalisation that has displaced the spot CLOB on Solana, Ch 2). Bob's continued participation is the structural finding: the absence of an LVR-aware passive LP venue at his capital scale; the Uniswap UI's gross-fees-prominent, LVR-absent display; the unrealised-loss psychology that makes withdrawal feel like crystallising a loss rather than stopping a structural bleed. HLP on Hyperliquid (Ch 9) is the structurally-different case the chapter develops as a comparison: HLP depositors are paid for the toxic-flow-absorption role, the role is explicitly disclosed, and the JELLY incident (March 2025, 65% TVL drawdown) demonstrated the tail risk while preserving the protocol's depositor base because the depositors had consented to the role. HLP is the chapter's clearest example of an informed passive-LP class with explicit consent; Bob's position is the chapter's clearest example of the same structural role without explicit consent.

4c. Slow market makers ​

This subsection is developed at the category level rather than via a single named firm — per RESEARCH.md recommendation, named-firm events (Wintermute's mid-May 2026 BTC/ETH liquidity compression) reflect volatility within the winner set rather than displacement of the loser category. The empirical anchors: Phoenix's revenue collapse from $3.7M (Q2 2024) to $68,604 (Q1 2026) — a 54× decline — documents the structural displacement of the Solana spot CLOB venue by prop-AMM-routed flow. The market makers who attempted to host quote-driven liquidity on Phoenix and OpenBook as their primary business model lost the competition not by being out-traded but by being out-architected: the venues they quoted on lost volume to aggregator-routed prop-AMM internalisation (Ch 2, Ch 8). Wu et al.'s 75 EOF arrangements accounting for ~71% of trading-related Ethereum builder revenue (Ch 7) document the L1 analogue: the market makers without builder-direct contracts (Banana Gun + Titan; Maestro + Beaverbuild) lost share of the most fee-paying retail flow to a small set of bot-builder pairs. The +33%–+101% access signal vs ~+3% operational signal decomposition (Ch 7) describes the same dynamic from the validator-side: a market maker without a Helius-RPC / BAM-Node / Frankfurt-colocation / Harmonic-strategy stack is competing against firms with all four. The structural finding: speed-and-relationship moats compress the competitive surface to a small number of vertically-integrated firms. The slow market makers' continued participation is explained by: (a) the sunk-cost of existing infrastructure; (b) the option value of remaining ready to deploy if the structural shape changes (BAM open-sourcing mid-Q2 2026 per Ch 8 could in principle reset the relationship-based moats); (c) the structural difficulty of pivoting to prop-AMM operation (which requires the firm to originate trading flow rather than to quote on it — a different business). The slow market maker is, in the chapter's framing, the loser category most able to consent to its position because the firms understand the structural mechanism; what they lack is the capital or the operational ability to exit.

4d. Validators without infrastructure relationships ​

The empirical anchors: Solana's validator count collapse from approximately 2,560 active validators in mid-2024 to approximately 770 in March 2026 documents the structural compression of the small-stake validator class.[^d] The ~150 validators projected to lose Foundation stake under the May 2026 SFDP concentration rules (Ch 8) is the chapter's freshest single data point on the validator-side loser category. The +33%–+101% access signal vs ~+3% operational signal decomposition (Ch 7) is the load-bearing structural anchor: a stock-Agave-default validator earns at-or-below network median priority fees regardless of operational discipline. The Solana validator profit Gini ~0.93 (Placeholder VC, September 2025; Ch 7) documents the resulting extreme inequality. Hyperliquid's 525,000-HYPE slot-21 threshold vs the 10,000-HYPE self-delegation minimum (Ch 9) produces an approximately 52× capital ratio between the threshold for active-set membership and the minimum operational position — the small-stake Hyperliquid validator who cannot accumulate the 525K HYPE delegation is structurally locked out of validator-set revenue. The Ethereum solo home staker is the named comparison: per Vitalik Buterin's published commentary and ethstaker.cc analysis, solo home stakers continue to earn the base PoS yield but lose share to Lido (~24% of all staked ETH), Coinbase Cloud (~5–12%), and Kiln (~6%); the solo staker's yield survives, but the solo staker's share of the network compresses as institutional staking accumulates. Erika's continued participation is the structural finding: hardware sunk costs + Alpenglow option value + the validator-set-membership option value keep the small-stake operator in the game even as the access-vs-operational gap makes their per-block revenue uncompetitive.

5. The consent gap — the chapter's clinical core ​

The chapter's editorial contribution: the four loser categories differ in their informed-consent relationship to the structural disadvantage.

CategoryLoss magnitudeLoss visibility (Layer 1 framing)Loss visibility (Layer 2 framing)Consent type
Retail tradersSmall per-trade × high countVisible per-trade (slippage)Invisible per-layer (RPC rebate, validator capture, builder margin)Uninformed — UX abstraction obscures the per-layer trace
Passive LPsContinuous LVR dragInvisible (fees displayed, LVR not displayed)Mathematically visible to those who read Milionis et al.Partial — LVR research exists but is not surfaced at deposit time
Slow market makersDisplacement, not loss-per-tradeVisible at firm level (volume share data)Visible at category level (Phoenix, EOF research)Informed — firms understand the structural mechanism
Validators without relationshipsOpportunity cost (top-quintile vs median)Visible per-epoch (rewards data)Visible cross-validator (Gini, +33-101% telemetry)Informed — operators have access to comparison data

The structural argument: the consent gap is largest for retail, and the consent gap is what makes the retail loser category structurally different from the other three. The other three categories include participants who know the structure and have made informed decisions (often constrained by lack of alternatives) to remain in it. The retail category includes a meaningful share of participants who do not know the structure — and whose continued participation cannot be straightforwardly described as consensual in the way the other categories' can.

The chapter does not argue that the structures are illegitimate. The chapter argues that the asymmetry of informed consent across the four loser categories is the most important structural fact about who-loses-and-why on-chain in 2026, and that the regulatory frameworks (Ch 7's EU PFOF ban; SEC's Order Competition Rule withdrawal; the ETF wrappers) do not address the consent gap because they regulate the asset-exposure layer rather than the trade-execution layer.

6. Why "just leave the table" is harder than it sounds ​

A short subsection developing the structural reasons each category does not exit.

  • Retail: alternative is CEX access (often unavailable by jurisdiction), or non-participation (forfeits asset exposure). Behavioural finding (FCA 2025 UK study): disclosure rules reduce participation modestly but do not improve per-trade losses for the residual participants.
  • Passive LPs: alternative is depositing nowhere (no comparable yield product at the capital scale of typical LP positions); HLP is a structurally-different option but requires the depositor to consent to clearinghouse-style tail exposure; CEX yield products (centralised lending) carry counterparty risk the LP is explicitly trying to avoid.
  • Slow market makers: alternative is pivoting to prop-AMM operation, which requires originating-trading-flow capability (a different business); or exiting market making entirely, which forfeits the firm's competitive position. Wintermute's prop-AMM pivot (Tessera V) and Temporal's pivot (HumidiFi) are the named cases of firms that did exit successfully; the firms that did not had a different capital and product position.
  • Validators without infrastructure relationships: alternative is exiting the validator set, which forfeits the staking yield itself and the option value of post-Alpenglow improved economics; the SFDP-removed Solana validator faces a 30-day window to acquire external stake or wind down; the small-stake Hyperliquid validator never enters the active set.

The chapter's editorial reading: "just leave the table" is a behavioural injunction that ignores the structural costs of leaving for three of the four categories, and ignores the consent gap for the fourth.

7. What changes when… ​

Transition to Chapter 12. Part IV's verdict has been the chronic-loser categorical finding. Chapter 12 closes the book with the forward-looking trends — ePBS arrival; the decentralised-sequencer roadmap (Espresso, Astria, Superchain); shared-state cross-chain interop; the appchain thesis; the intent-based architecture trend (Anoma, CoW, UniswapX). The question Chapter 12 must engage with: do any of these trends materially address the consent gap Chapter 11 has documented? The chapter's tentative answer is "in some categories yes, in retail not yet" — but the development belongs in Ch 12.

8. Footnotes and sources ​

18–22 numbered citations. URLs + access dates. Cross-references to Chs 2, 3, 7, 8, 9, 10 marked "Already cited in Chapter X" where source previously in-book.

Worked example chosen — and where it threads ​

Three short worked examples (per RESEARCH.md recommendation), one per category with named actors. Slow MM developed at category level. The three examples thread through §3 (worked example) and are referenced again in §4a, §4b, §4d, §5 (consent gap table), and §6 (exit paths).

Diagrams needed ​

Two diagrams — one Mermaid, one inline table.

  1. D1 — The consent-gap matrix table (inline Markdown table; see §5). Four rows × four columns. Each row is one loser category; columns are loss magnitude, Layer-1 visibility, Layer-2 visibility, consent type. The chapter's central editorial illustration.

  2. D2 — The exit-path constraint diagram (Mermaid flowchart TB): four loser categories on top, three exit paths (CEX / non-participation / pivot) on bottom, edges marked with the structural cost of each path. Placed in §6.

The chapter does not re-render any Chs 1–10 figure; cross-references replace re-rendering.

Glossary terms this chapter introduces ​

Two new entries (the chapter is largely a synthesis chapter that uses already-defined terms):

  • Consent gap — the chapter's central conceptual contribution; the asymmetric informational position of the four chronic-loser categories.
  • Chronic loser (on-chain) — a structurally disadvantaged participant who continues to participate despite the documented disadvantage.

Cross-references: LVR, passive LP, prop-AMM, SFDP, validator-quintile gap, exclusive order flow, Pump.fun, Banana Gun — all already defined.

Backward (chapters this builds on):

  • Ch 2 — passive LPs / LVR; the public AMM displacement by prop-AMMs
  • Ch 3 — Alice's $73 of 2024 slippage; Carlos as Ethereum trader case
  • Ch 7 — Wu et al. EOF research; Banana Gun + Titan; access-vs-operational decomposition
  • Ch 8 — Solana market structure; SFDP enforcement; validator client diversity; the +33-101% access signal
  • Ch 9 — HLP as structurally-different passive-LP-class equivalent; JELLY tail risk
  • Ch 10 — Ethereum solo home staker; L2 sequencer concentration; cross-chain MEV searcher top-5

Forward (chapters this sets up):

  • Ch 12 — forward-looking; whether the trends (ePBS, decentralised sequencers, intent-based architectures, appchains) address the consent gap
  • Epilogue — the Figment-as-marketing-asset reveal; the implicit pitch that runs through Part IV's reckoning

Chapter-level voice and structure notes ​

  • Verdict chapter, not balance-sheet chapter. The chapter's primary job is to land the four-category framing and the consent-gap argument as Part IV's central structural finding.
  • No new "Meet the actor" sidebar. All actors are returnees. Three short worked examples replace a single sidebar.
  • Word budget: 4,500–5,500 words inc. footnotes. Lighter than Chs 5–10 (which carried full concept development) — Ch 11 is synthesis; the heavy lifting has been done.
  • Footnote count target: 18–22. Most are primary-source citations or cross-references to earlier chapters.
  • Voice anchors: the consent gap is the through-line. Each loser-category subsection ends with a one-line nod to where that category sits on the consent-gap axis.
  • Banned-move audit at draft time: no moralising ("rigged," "predatory"); no doom words; no DeFi-is-dead framing; clinical not outraged; specific firms named everywhere; show-the-dollar in §3.
  • The chapter's editorial contribution: the consent-gap framing. The chapter has the clearest editorial voice in Part IV; the framing should be load-bearing without being preachy.

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