The House Edge

Appearance

Chapter 2 — Outline (Phase 2) ​

Status: OUTLINE (Phase 2 — written by agent, awaiting Nick's review) Date: 2026-05-13 Working chapter: 02 — Where Liquidity Lives Carries forward from: RESEARCH.md

Working assumptions (carried from Phase 1) ​

These are the seven open-question answers from RESEARCH.md §6, treated as defaults. Override before Phase 3 is cheap; after, it isn't.

  1. The Solana convergence story is sharper than the SPEC's. The chapter develops the prop-AMM thread fully here (Temporal-operated HumidiFi, Ellipsis-operated SolFi, Tessera) rather than reserving it for Chapter 8. The reader needs the prop-AMM concept to understand 2026 Solana liquidity at all.
  2. The 2021–2022 LP-loss data is treated as "the most recent comprehensive measurement," with LVR as the live academic lens. No 2026 update is published. The chapter is explicit about that vintage.
  3. Temporal is named as HumidiFi's reported operator, with "reportedly" hedge per DL News' on-the-record attribution.
  4. Phoenix's decline opens the chapter — the cold open is the venue's $3.7M-to-$11K quarterly-revenue collapse. It is the chapter's most vivid single image.
  5. Hyperliquid appears in cameo here, not full architectural treatment. JELLY exploit is mentioned only as a one-sentence footnoted aside; full mechanical treatment of HLP and JELLY lives in Chapter 9.
  6. Worked example: Bob's $50,000 Uniswap V3 ETH/USDC LP position, tracked over a month in early 2026, used to develop AMM mechanics and the LP role.
  7. Hyperliquid spread / lead-lag data is omitted — the only published number cited is the 4.5 bps base taker fee on perps (corrected from the 3.2 bps figure originally used), footnoted with a note that the rest is not publicly measured.

A targeted secondary-research pass is running in parallel for items 2 and 7; findings will be incorporated into the Phase 3 draft if useful, noted in REVIEW_NOTES.md either way.

Title and subtitle ​

Chapter 2 — Where Liquidity LivesTwo architectures, four actors, and the quiet displacement of the passive liquidity provider.

Cold open ​

In its peak quarter — Q2 2024 — Phoenix earned $3.7 million in trading fees on Solana. Phoenix was the most technically credible attempt anyone had made at an on-chain central limit order book; the team that built it, Ellipsis Labs, was a small engineering shop whose credential list included former employees of Jane Street and Jump. They had argued, in white papers and on Twitter, that the AMM was a temporary solution — that on-chain spot markets would eventually look more like NASDAQ than a vending machine. Twenty-one months later, in April 2026, the same team announced that Phoenix Spot would be renamed "Phoenix Legacy" and they would build perpetuals instead. Q1 2026 fee revenue across the entire venue was $11,190. The team did not quit; the code did not break. The architecture lost the market.

(Why this open: a named team, a named venue, a specific number, and a punchline ($11,190 in a quarter). The chapter's job is then to explain which architectures won and why, and to follow the passive LP — the role most readers think of when they hear "DeFi liquidity" — into the role's quiet 2026 displacement.)

What this chapter answers ​

  • What is liquidity, in plain terms, and why doesn't a market just work without it?
  • How do AMMs and CLOBs differ economically — not just mechanically?
  • What does it actually look like to be a passive LP, and why did the role's outcomes turn out to be worse than the early literature promised?
  • Why have aggregators and solver markets become the dominant retail interface, and what do they capture for themselves versus pass through to the trader?
  • Why did Solana converge on prop-AMMs-routed-through-Jupiter, Hyperliquid on a native CLOB, and Ethereum on both?

Section list (one sentence each) ​

  1. Cold open — Phoenix Q2 2024 to Q1 2026: $3.7M → $11K.
  2. What this chapter answers — the five questions, as above.
  3. The setup (≈500 words). Defines liquidity in plain terms. The two architectures named: AMM, CLOB. Aggregators and RFQ as overlays. Why the reader should care: every trade in this book passes through one of these structures. Brief plant of Bob.
  4. The worked example (≈300 words). Bob, a business reader, deposits $50,000 into the Uniswap V3 ETH/USDC 5-bps pool in February 2026. The pool's mechanics are not yet explained — the reader is told Bob clicks "Deposit," signs, and is told he is now earning fees. The chapter then asks: what is he earning, and from whom?
  5. The mechanics, in detail (≈2,600 words; four H3 subsections, the chapter spec template's hard limit):
    • 5a. The AMM curve and the loss the LP doesn't see (≈700 words). The constant-product (xy = k) curve. The price-impact effect of a swap. Concentrated liquidity (V3) as an attempt to make capital more efficient. Bob's position re-introduced — what fees he earns, and what he loses to the arbitrageurs who keep his pool's price aligned with the broader market. Impermanent loss is named and then immediately reframed as Loss-Versus-Rebalancing (Milionis–Moallemi–Roughgarden–Zhang). The Topaze/Bancor finding ($60.8M aggregate LP underperformance, 49.5% net negative) and the Uniswap Labs 5-bps finding (passive V3 LPs underperformed V2 by 68% in this tier) anchor the empirical picture.
    • 5b. The CLOB and why every on-chain attempt at spot has failed except Hyperliquid (≈700 words). The order book. Bid, ask, spread, depth. The market maker as a participant — first appearance of the book's Meet the Market Maker sidebar (Wintermute on Hyperliquid: 76 markets, ~$199M resting notional). Why CLOBs on Ethereum-style architectures lose to AMMs: latency, gas, the cost of cancelling stale orders. Hyperliquid as the case where the architecture (no public mempool, ~200ms median end-to-end / 900ms p99 latency, HyperBFT consensus, perps-first) made the CLOB viable. Phoenix's decline as the counter-example (callback to cold open).
    • 5c. Aggregators, solvers, and the retail interface that replaced the pool (≈600 words). The routing problem: Bob's neighbour Alice wants to swap $10,000 of USDC for SOL, and the best price is fragmented across five pools. Jupiter solves this on Solana — 93.6% of aggregator-routed flow in early Q1 2026, with >70% of all Solana DEX volume now routed through aggregators. On Ethereum, the equivalent layer is fragmented: CoW Swap, UniswapX, 1inch Fusion. The new mechanism the Ethereum side introduced — solvers, who fill trader intents from inventory — is mechanically different from a pool but produces the same end result: the AMM pool becomes a back-stop, not the front of the market. SCP + Wintermute = >90% of UniswapX volume; Barter targeting >50% of CoW solver share.
    • 5d. The prop-AMM displacement (≈600 words). The Solana liquidity story in 2026 is not "AMMs and aggregators" — it is prop-AMMs routed through aggregators. HumidiFi (reportedly run by Temporal), SolFi (Ellipsis), Tessera. >50% of Solana spot DEX volume combined. HumidiFi alone holds ~65% of the prop-AMM segment and gets >95% of its flow from aggregators. The economic argument: passive LPs absorbed adverse selection at a discount; prop shops handle the same adverse selection directly, price it correctly, and capture the spread for themselves. The structural consequence: the entity providing liquidity on Solana in 2026 is not Bob and not a public DAO-governed pool — it is a private market-making firm whose code is closed and whose inventory is not on a public dashboard.
  6. How this plays out on each chain (≈350 words, three paragraphs). The required chain-comparison.
    • Solana: prop-AMMs (>50% of spot DEX volume) routed through Jupiter (93.6% of aggregator flow; >70% aggregator dominance). The public AMM business is real (Meteora, Raydium, Orca) but increasingly supplies the long tail. Public CLOB attempts have wound down — Phoenix Legacy, OpenBook small.
    • Hyperliquid: native on-chain CLOB ($177.66B 30-day perp volume; ~70% of on-chain perp share globally). Adverse selection is absorbed by HLP, the protocol's own LP vault ($391M TVL). Wintermute, Flowdesk, and other professional MMs supply the rest. Spot is ~1.75% of perps — the architecture works for perps specifically.
    • Ethereum and L2s: the both-and case. Uniswap V3 + V4 remain the AMM standard ($2.1B + $805M TVL); solver-mediated routing (UniswapX, CoW, 1inch Fusion) has displaced direct AMM trading for a meaningful share of retail flow. On L2s the AMM business is fragmented and dominated by venues like Aerodrome on Base.
  7. Who wins, who loses, why (≈350 words). The verdict — lighter in tone than Chapter 4's because this chapter is structural.
    • Winners: prop-AMM operators (HumidiFi, SolFi, Tessera) and the firms reportedly behind them (Temporal); aggregators (Jupiter and its 93.6% share); solvers on Ethereum (SCP and Wintermute, who together fill >90% of UniswapX); Hyperliquid validators (paid via HLP's accrual to the protocol).
    • Losers: passive LPs on the dominant V3 ETH/USDC tier (the role that 2021–2022 empirical work showed loses to HODL; no 2026 re-measurement exists but the structural argument has tightened with LVR); on-chain CLOB designers who chose spot (Phoenix, OpenBook); retail traders who route directly to pools without an aggregator.
    • Is this bad?: not particularly. The chapter is explicit that the architectures that won are the ones that handle adverse selection cheapest, and adverse selection is a feature of every market that has ever existed. What changed is who the providers are. The role Bob volunteered for in 2021 is being done in 2026 by a different kind of entity. That is what the rest of the book is about.
  8. What changes when… (one paragraph). The transition: what changes when a trader's transaction is visible before it reaches the pool, and the prop AMM can see it coming? That is Chapter 3.
  9. Footnotes and sources — numbered, URL'd, with access dates. Approximately 15 footnotes.

The worked example and where it threads through ​

Bob's $50,000 LP position on the Uniswap V3 ETH/USDC 5-bps pool, February 2026 → mid-March 2026.

SectionBeat
§3 (Setup)Plant: "imagine a LP position on Uniswap. We'll come back to it."
§4 (Worked example)Bob deposits. He sees the wallet confirmation. The chapter asks: he is now earning fees — from whom?
§5a (AMM curve)Payoff: the mechanics. Bob's monthly P&L sketched in dollars: fees ≈ $X, IL/LVR loss ≈ $Y, net ≈ $Y − $X. The 5-bps tier finding (passive V3 LPs underperform V2 by 68% in fees) lands here.
§5c (Aggregators)Continuity: Bob's pool gets some flow from Alice via Jupiter routing on a different chain — the chapter shows the connection.
§6 (Chain comparison)Bob's specific outcome contextualised: same role would look different on Hyperliquid (he can't be a passive LP on a CLOB; he could deposit in HLP and earn ~22% trailing-12-month, though that number is somewhat stale).
§7 (Verdict)Bob is the named loser. Not catastrophically. But the role he signed up for has changed underneath him.

Diagrams needed ​

Three. (Bible budget 2–5; SPEC suggested 3–4.)

  1. D1 — The AMM curve (Mermaid + static SVG). A constant-product xy = k bonding curve, annotated with: (a) the resting state, (b) what a $50K swap does to it, (c) where Bob's concentrated-liquidity range sits relative to the current price. Mermaid + an SVG fallback for the print pipeline (Mermaid pie/sequence are well-supported by mermaid-cli but bonding curves aren't a standard Mermaid type — likely a pure SVG for this one).

  2. D2 — The CLOB order book snapshot (Mermaid or markdown table). Five levels each side of a sample Hyperliquid BTC perpetual book, with named market makers attributed where data permits (e.g., "Wintermute at $X depth at $Y level"). The reader should see the spread, the depth, and the named participants. Markdown table is probably cleaner than Mermaid for this one.

  3. D3 — Aggregator routing Sankey diagram (Mermaid). Alice's $10K swap from Chapter 4's prologue, routed across three venues: 60% HumidiFi (via Jupiter), 25% Orca, 15% Meteora. Each leg annotated with the price it filled at. Mermaid flowchart LR with weighted edges works for this.

(No fourth diagram. The MM vs LP P&L table is folded into §5b's prose as a couple of sentences plus the MM sidebar.)

Glossary terms this chapter introduces ​

To be appended to GLOSSARY.md in Phase 3.

Defined in full (first appearance):

  • Liquidity — the supply of resting orders or pooled inventory ready to fill a trade at a known price.
  • AMM (automated market maker) — a pool of paired assets whose price is set by an algorithm rather than by matched bids and asks.
  • CLOB (central limit order book) — a venue where bids and asks are matched explicitly, in the order they arrive.
  • Liquidity provider (LP) — a participant who supplies inventory to a market in exchange for fees.
  • Passive LP — a participant who deposits into a pool and accepts the pool's algorithm to manage their position, versus an active LP who manages a range or quote.
  • Market maker (MM) — a participant who supplies two-sided quotes to a venue, earning the spread between bids and asks.
  • Spread — the difference between the best bid and the best ask, expressed in price units or basis points.
  • Slippage — the difference between the price a trader expected and the price they got.
  • Impermanent loss (IL) — the difference between what an LP would have held by simply holding the underlying assets and what they actually hold in the pool, as the assets' relative prices move.
  • Loss-Versus-Rebalancing (LVR) — a sharper formulation of LP loss: the amount an LP loses to arbitrageurs who keep the pool's price aligned with the broader market, before fees.
  • Aggregator — a protocol that takes a trader's intent and routes it across multiple venues to minimise effective slippage.
  • RFQ (request for quote) — a model in which a trader requests a fillable price from one or more market makers, who respond with quotes; the trader picks the best.
  • Prop-AMM — a private, market-maker-operated AMM whose code is closed and whose inventory is not publicly tracked, settling on-chain but quoted off-chain.
  • Solver — the Ethereum-side analogue of a prop-AMM: a firm that fills a user's transaction intent from inventory sourced wherever it chooses.
  • HLP (Hyperliquid liquidity provider vault) — Hyperliquid's protocol-owned LP vault that quotes against incoming flow and absorbs liquidation surplus.

Cameo (inline definition; full glossary entry but no full prose treatment):

  • Constant-product curve, concentrated liquidity, V3 fee tier, resting order, marketable order, taker, maker, taker fee.

Backward:

  • Prologue (not yet drafted): Alice's $10K USDC→SOL swap is recalled in §5c's aggregator routing diagram.
  • Chapter 1 (not yet drafted): the chapter must briefly establish "what a trade is" in §3, since Ch 1 isn't ready. One sentence carries the freight.

Forward:

  • Chapter 3 (Mempool, not yet drafted): §8 transitions to it directly. The "what does it mean that prop AMMs see your transaction before it reaches the pool" question is Chapter 3.
  • Chapter 4 (The Searcher, drafted): §5a explicitly references the passive LP as the role Chapter 4 already implicated in CEX-DEX arbitrage. §7 names the LP as a chronic loser, paying off the half-introduction Ch 4 made.
  • Chapter 6 (Infrastructure): aggregator and solver-market mechanics get fuller treatment.
  • Chapter 7 (Exclusive Order Flow): solver-as-private-fill becomes the central thread.
  • Chapter 8 (Solana): the prop-AMM story expands.
  • Chapter 9 (Hyperliquid): HLP gets its full mechanical treatment; the JELLY exploit gets a full case study.
  • Chapter 11 (Who's at a Disadvantage): the passive LP is the first of the four chronic-loser archetypes.

Phase 2 is complete. Per the user's instruction to "proceed with putting together a draft," Phase 3 begins immediately rather than waiting for an explicit checkpoint.