The House Edge

Appearance

Chapter 8 — Outline (Phase 2) ​

Status: OUTLINE (Phase 2 — written by agent, accompanying RESEARCH.md) Date: 2026-05-14 Working title: Solana: AMMs, Auctions, and the Speed GameWorking subtitle: The chain where the validator IS the slot leader IS the block constructor — and what that produces when you wire eight cities and three software stacks into one market.

Cold open (1 paragraph) ​

On 19 January 2025 — a Sunday — Solana's network captured $56.8 million in real economic value (REV) in a single day, of which approximately two-thirds was Jito tips paid by searchers for inclusion priority and bundle execution. The chain's January 2025 monthly REV was $551.7 million; the 2025 annual total was approximately $1.4 billion. Sixteen months later, in Q1 2026, Solana's REV had collapsed 68% year-over-year to ~$89.5 million, putting the chain at #2 by REV behind Hyperliquid for the first time in its history. The compression is not the story; the compression is what made the structure visible. The chain that produces $56.8 million of value-extraction in a single day, then loses two-thirds of that flow in a year as memecoin trading cools, is a chain whose underlying architecture — who captures the value when it is there, who keeps capturing it when it shrinks — is now legible in a way it was not when the volume was higher. This chapter is about that architecture: about why aggregators won spot, why no order book has, what the Jito auction actually auctions, what the validator-client diversity does and does not change, and why a Frankendancer Harmonic Performance validator co-located in Equinix FR5 in Frankfurt earns approximately double the priority fees per block of a stock Agave validator in the median rack. The Solana the trader sees is Phantom and Jupiter. The Solana that captures her dollar is eight cities, three client stacks, and six firms.

Section list (one-sentence summaries) ​

1. What this chapter answers ​

Four questions:

  • What is the Solana market structure that produces the +101% Harmonic Performance / +3% operational ratio Chapter 7 documented — and why is the gap specifically a Solana phenomenon rather than a generic on-chain one?
  • Why did the spot CLOB experiment end on Solana, and what replaced it?
  • What does the Jito Block Engine actually auction — and what changes under BAM and JIP-24?
  • Where does retail's dollar go on Solana in 2026, and which firms have the most concentrated claim on it?

2. The setup ​

The chain's architectural premise stated cleanly: Solana's validator is simultaneously the slot leader, the block constructor, and (under BAM) the executor of TEE-attested ordering decisions made by separate node operators. No PBS layer separates the proposer from the builder; no consensus-internal matching engine subsumes both. The integration is what makes client choice, scheduler mode, and infrastructure relationships dominate Solana validator economics in a way they cannot dominate the others. The chapter previews three stratified-by-architecture findings: (a) retail loses less in 2026 than in 2024 because Beam + BAM + the 8 April 2026 patch + Foundation enforcement closed the formal-sandwich surface — but the cumulative take at every stack layer remains the cost of routing through five firms the trader never sees; (b) market makers internalised the spot venue by displacing the public AMM with prop-AMMs; (c) validators with infrastructure relationships have the largest dollar magnitude per actor and the lowest visibility.

3. The worked example ​

Alice (Chs 1, 3, 4) makes her $10,000 USDC → SOL swap one more time. The chapter follows the dollar through every layer of the chain's 2026 stack. Phantom wallet → Helius RPC (50/50 rebate split) → Jupiter aggregator (~93.6% share) → Jupiter Beam (the default protected routing) → Jito block engine in Frankfurt (~1–2ms LAN-local) → Frankendancer Harmonic Performance validator co-located in Equinix FR5 → Solana settlement. The per-layer dollar trace lands at approximately $8–$15 cumulative take versus Alice's ~$73 of formal-sandwich slippage in 2024 (Ch 3) — the chain compressed the most-visible extraction surface, and the structural take still routes through a handful of firms the trader never sees. The chapter walks the trace as a labelled table early in the worked-example subsection and uses it as the spine across §5's mechanics subsections.

4. The mechanics, in detail ​

Five H4 subsections:

4a. Aggregators ate the interface — and Jupiter took the aggregator layer ​

What Ch 2 framed as "aggregators are the front of the market" the chapter develops as the structural shape of Solana spot in 2026. Jupiter ~93.6% of aggregator share; aggregator-routed flow ~74% of Solana DEX volume (up from ~40% six months prior); 2025 cumulative ~$716B routed; Ultra (Jupiter's protected-routing product) ~95% of aggregator volume and ~50% of total DEX volume. Beam — Jupiter's transaction-landing engine routing signed transactions to Jito as private bundles — is the default behind Ultra and opened as a public Transaction Submission API in April 2026 (any signed Solana tx, not just Jupiter-routed). Published Ultra V3 slippage: ~+0.006% on Beam-protected paths vs −0.14% on unprotected (Ch 3). The structural fact: the interface user sees (Phantom + Jupiter) is the interface that handles roughly half of all DEX volume — and routes it into one private bundle market.

4b. The CLOB experiment ended; prop-AMMs replaced it ​

Phoenix peak Q2 2024 $3.7M → Q1 2026 $68,604 quarterly revenue (~54× decline); renamed Phoenix Legacy; pivoted to perpetuals (Ch 2 anchor). Manifest at ~$3.9B 30-day volume is the only above-noise spot CLOB and is <2% of Solana's ~$284.5B Q1 2026 DEX spot volume. The architectural diagnosis: mempool surface (Ch 3) + prop-AMM displacement (Ch 2) — retail routes through Jupiter, Jupiter routes to prop-AMMs, the order-book design produced too little volume to compensate the market makers who would host it. Those market makers (Ellipsis Labs, Wintermute, Temporal) now operate prop-AMMs instead. Update Ch 2's mid-2025 HumidiFi-65% framing: March 2026 prop-AMM share ~55% of DEX volume (down from 68% in February); BisonFi (Forward Industries, Dec 2025 launch) overtook HumidiFi in late January 2026; both then declined >55% in March. Frame the segment as competitively contested. Operator map: HumidiFi → Temporal (reported); SolFi → Ellipsis Labs (publicly claimed); Tessera V → Wintermute (Wintermute-confirmed to DL News Aug 2025); BisonFi → Forward Industries; ZeroFi / Goonfi → undisclosed. Forward Industries / BisonFi is the chapter's one new-to-the-book institutional cameo — a US-listed company operating a Solana dark AMM compresses the regulatory-disclosure gap the segment previously enjoyed.

4c. What the Jito auction actually auctions ​

Three things in one product, not "MEV from a public mempool" (Solana's Gulf Stream produces none, Ch 3): (a) inclusion priority for searcher bundles into the next slot's block, against SOL tips; (b) atomic bundle execution with revert protection (whole-or-nothing); (c) routing access to validators on Jito-derivative clients. Stake distribution behind the auction: ~32% Agave Jito + ~28% JitoBAM ≈ ~94% Jito-Solana-variant stake by Q2 2025 (Chs 5, 6). Jito tip data: late-2024 ~$2.5M/day average; single-day high $14.7M (17 Nov 2024). Post-JIP-24: the entire 6% Block Engine + BAM cut routes to the Jito DAO treasury (Ch 6) — projected $15–50M annual run-rate. BAM as the architectural shift the chapter develops in full: launched on Solana mainnet 21 July 2025; four launch BAM Node operators (Helius, Triton One, SOL Strategies, Figment); ~25% of stake connected by late February 2026; new sites Dublin + Dallas; Lithuania next; open-source target mid-Q2 2026. BAM Nodes run inside AMD SEV-SNP TEE enclaves; sequencing delegated to BAM Nodes; execution stays with the validator (Ch 3 / Ch 5 anchor).

4d. The client + geography matrix — Chapter 8's signature new material ​

Two subordinate threads in one subsection:

Client diversity: March 2026 stake-weighted distribution — Agave Jito 32%, JitoBAM 28%, Agave Harmonic 17%, Frankendancer 12%, Rakurai 6%, Firedancer 2%. ~86% on Agave-based clients. Maintainers: Anza (Agave); Jito Labs (Jito-Solana variants + JitoBAM); Jump Crypto (Frankendancer + Firedancer); Rakurai (closed-source fork). Structural point: none of these clients is sandwich-default in 2026. The combination of the June 2024 Foundation enforcement + Harmonic's SFDP-compliant strategies + BAM TEE + Beam protected routing + the 8 April 2026 protocol patch (Ch 3 anchor) collectively closed the formal-sandwich surface. Harmonic's role at the orderflow-builder-aggregation layer: four scheduling strategies (FBA + Priority Fee Ordering, FIFO, MREV, Custom) with FBA + Priority Fee as the SFDP-compliant default; Ben Coverston is CEO of both Temporal and Harmonic (Ch 6 anchor — concentration fact). Rakurai's "no timing games" positioning: explicit commitment in the Figment migration report (Ch 5); the 2 March 2026 Figment migration as the cleanest published institutional-adoption case of 2026 (SRR 6.85% → 7.17%, +60% priority fees, ~5× MEV).

Geography: Solana stake by city — Frankfurt 19% / Amsterdam 16% / London 12% / Vilnius 6% / Tokyo 4% / Ashburn 4% (top 6 ≈ 61%). West Europe + N. America ≈ 70% of all stake. Jito mainnet block engines run in eight cities: Amsterdam, Dublin, Frankfurt, London, NY, Salt Lake City, Singapore, Tokyo — five of which match the top-six stake cities. Optimal latency block-engine-to-validator <50ms; <100ms operates well within bundle throughput. Standard colocation: bare-metal AMD EPYC 9354P + 512GB RAM + 3×NVMe (one-time $8–15K), rack+power $200–500/mo, premium $400–800/mo. Equinix FR5 (Frankfurt) and NY5 (New York) are the named anchor data centres. The "next to the TPU port" pattern from Ch 3 (non-voting validator alongside paid RPC in the same rack, forwarding to current leader via QUIC) is the load-bearing latency play; RPC Fast reports 5–10× latency reduction from LAN-local vs remote cloud. The structural argument: the +101% Harmonic Performance validator is, in practice, the validator co-located in Equinix FR5 or AM3 on a Jito-derivative client with a BAM-node relationship. The validator-quintile gap Chapter 7 documented is geographic, software, and relationship-based — not operational. Block-engine footprint maps onto stake footprint because both follow data-centre infrastructure that predates either; equilibrium-shape, not one-way causation.

4e. SFDP and Alpenglow — the two governance levers, present and future ​

The Solana Foundation Delegation Program is the chain's largest single-source stake-delegation pool and de facto soft-governance lever. Starting 1 May 2026: participants must operate on an ASN and hosting provider holding <25% of network stake; data-centre concentration must be <15% across all staked validators at any single data-centre provider. "Onboard 1, offboard 3" policy retires validators with <1,000 SOL external stake after 18 months. ~150 validators projected to lose Foundation stake under the May 2026 rule. Two structural levers: anti-concentration (the data-centre rule directly counteracts the Frankfurt/Amsterdam colocation gravity §4d documented) and anti-extraction (censorship-resistance + "no sandwich-enabling private mempools"; June 2024's ~32-validator removal, Ch 3 / Ch 7 anchor). DeezNode/Vpe forward-reference only — Chs 3, 4, 7 already developed; no new material. Alpenglow as a ~300-word forward-looking close to the mechanics section: SIMD-0326 activated on testnet 11 May 2026; mainnet target late Q3 / early Q4 2026; three Solana-specific shifts — (a) vote transactions removed from blockspace (~70% of all Solana transactions at peak), freeing block capacity; (b) VAT ~1.6 SOL/epoch entirely burned (first explicit per-epoch fixed validator cost); (c) profitable-validator stake threshold drops from ~4,850 SOL to ~450 SOL (~90% reduction). Full architectural treatment is Ch 12; the chapter notes the forward-looking shifts and the structural question they pose: does VAT + threshold drop reverse the geographic-concentration equilibrium, or does the access gap survive the cheaper-validator economics intact?

5. How this plays out on each chain — and what Solana's structure isn't ​

Comparison paragraph, three sentences each: Solana (the integrated-validator architecture; the access-vs-operational gap as a software-and-geography function; the chain whose REV compresses fastest as memecoin volume cools); Hyperliquid (the no-mempool / in-consensus-matching architecture; the structural surface Solana's vertical integration could not fully eliminate, which Hyperliquid's design does; #1 by Q1 2026 REV — full treatment Ch 9); Ethereum (the explicit PBS separation; the chain where exclusive flow reached its most mature form, Ch 7; cross-chain comparison reserved for Ch 10).

6. Who wins, who loses, why ​

Three-actor stratification — the verdict frame: (a) retail traders lose via slippage, residual long-tail sandwiches, and the cumulative take at every stack layer — but lose less than in 2024; (b) market makers win spot trading by internalising adverse selection via prop-AMMs (Temporal, Ellipsis Labs, Wintermute, Forward Industries) and solvers; (c) validators with infrastructure relationships win the most, with the largest dollar magnitude per actor and the lowest visibility. The "top validator" — Frankendancer Harmonic Performance in Frankfurt FR5 with a Helius-backed RPC and a BAM-node relationship — earns ~+101% priority fees vs a stock Agave-default median validator with no relationships and no SFDP delegation. The difference is overwhelmingly software + infrastructure access, not operational excellence. Closing structural observation: the chain that compressed its formal-sandwich surface ($73 of 2024 slippage → ~$8 of 2026 cumulative take in Alice's worked example) did so by replacing one visible extraction layer with five less-visible ones. Whether that is an improvement is a judgement call the chapter offers but does not impose.

7. What changes when… ​

Transition to Chapter 9 (Hyperliquid: The CLOB That Worked). Solana is the chain whose vertical integration produces the access gap Chapter 7 documented; Hyperliquid is the chain whose architecture eliminates the surface that gap operates inside. The natural follow-up question — what does a chain look like when the matching engine sits inside consensus, the validator set is permissioned-top-21, and the HLP is the chain's house? — is Chapter 9.

8. Footnotes and sources ​

20–22 numbered citations. URLs + access dates. Cross-references to Chs 2, 3, 5, 6, 7 marked "Already cited in Chapter X" for any source previously in-book.

Worked example chosen — and where it threads ​

Candidate A (full-stack dollar trace through the Solana 2026 stack) — selected per RESEARCH.md recommendation. Alice's $10,000 USDC → SOL swap, threaded:

  • Cold open: previews the 19 Jan 2025 single-day-REV anchor and the structural-compression-as-legibility framing
  • §3 Worked example: full table walked layer-by-layer (Phantom → Helius → Jupiter → Beam → Jito Frankfurt → Frankendancer Harmonic Performance validator in Equinix FR5)
  • §4a–c: each subsection references the relevant row of the table
  • §4d geography subsection: the FR5 anchor lands here; the geographic decomposition is what makes the +101% number concrete
  • §6 verdict: the $8–$15 cumulative take vs $73 of 2024 slippage is the closing dollar comparison

Candidate B (Figment Rakurai migration) deployed as a callout / footnote in §4d. Candidate C (geographic case study) integrated into §4d rather than standing alone.

Diagrams needed ​

Two diagrams. Both are Mermaid (renders inline in VitePress).

  1. D1 — Layer-by-layer flow (Mermaid flowchart LR): Alice → Phantom → Helius RPC → Jupiter → Beam → Jito Frankfurt → Frankendancer Harmonic Performance validator → Solana ledger. Annotated with per-layer take and the named firm at each step. Placed at the top of §3 (worked example).

  2. D2 — Block-engine ↔ stake-city map (Mermaid flowchart TB with two columns — block-engine cities on the left, top-stake cities on the right, edges where they overlap). Five-of-six overlap visible. Placed in §4d geography.

The validator-quintile decomposition matrix (Ch 7's table) is not re-rendered — it's referenced via Ch 7 footnote.

Glossary terms this chapter introduces ​

Three to four new entries:

  • Jupiter Beam — Jupiter's transaction-landing engine; default protected routing inside Ultra; opened as a public Transaction Submission API in April 2026.
  • BAM Node — the TEE-attested-sequencing operator role under Jito's Block Assembly Marketplace; four launch operators (Helius, Triton One, SOL Strategies, Figment).
  • SFDP (Solana Foundation Delegation Program) — the chain's largest single-source stake-delegation pool and de facto soft-governance lever; the May 2026 ASN / data-centre concentration rules.
  • Colocation (Solana validator context) — bare-metal AMD EPYC + 512GB RAM in Equinix-tier data centres; $200–800/mo; the "next to the TPU port" latency play.

Cross-references: most relevant terms (Agave, Jito-Solana, JitoBAM, Frankendancer, Firedancer, Rakurai, Harmonic, prop-AMM, validator client, Alpenglow, Validator Admission Ticket, Votor, Rotor, etc.) are already defined in the glossary; this chapter links to them rather than redefining.

Backward (chapters this builds on):

  • Ch 1 — the trader's $10,000 USDC → SOL swap; the dollar trace device
  • Ch 2 — aggregator dominance, the CLOB experiment, prop-AMM displacement, the Phoenix decline
  • Ch 3 — Gulf Stream / no-mempool architecture; June 2024 Foundation enforcement; the 8 April 2026 patch; DeezNode/Vpe
  • Ch 4 — searcher economics behind the Jito bundle market
  • Ch 5 — validator client diversity; scheduler modes; timing games; the Figment Rakurai migration
  • Ch 6 — Helius, Jito Labs, Temporal, Harmonic profiles; JIP-24
  • Ch 7 — the +33–101% / +3% access-vs-operational decomposition

Forward (chapters this sets up):

  • Ch 9 — Hyperliquid's no-mempool / in-consensus-matching architecture as the structural alternative; the "Solana #2 by Q1 2026 REV behind Hyperliquid" anchor
  • Ch 10 — Ethereum + L2s and PBS as the third structural option
  • Ch 11 — chronic losers: bottom-quintile Solana validators, retail-AMM passive LPs, slow market makers
  • Ch 12 — Alpenglow's full architectural treatment; SVM appchains; HIP-3-style markets

Chapter-level voice and structure notes ​

  • Synthesis chapter, not concept chapter. No new economic concepts beyond what Chs 2–7 introduced; the chapter's value is in assembling them into a single integrated picture of Solana 2026.
  • No "Meet the actor" sidebar. All actors are returnees. Per Ch 7 precedent, the chapter develops existing characters' Solana-specific roles in one frame.
  • Word budget: 5,500–6,500 words inc. footnotes (lighter than Chs 5–7 because synthesis).
  • Footnote count target: 20–22. Most are cross-references to earlier chapters marked "Already cited in Chapter X."
  • Voice anchors: the verdict's three-actor stratification (retail / MM / validator-with-relationships) is the chapter's organising frame, previewed in §2 and landed in §6.
  • Banned-move audit at draft time: no "Solana is fast" framing without follow-up; no tribal endorsement; no Web 1.0 / Web 2.0; specific firms named everywhere; show-the-dollar every subsection.

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