Primary Findings

• VENUE CATEGORY.
  Venue category is the dominant predictor of execution quality. Broker-dealers and deep CEXes are interchangeable inside half a basis point for trades below $10K. Above $100K, order-book venues — regardless of whether they sit onchain or offchain — demonstrate materially superior execution, while AMMs deteriorate dramatically: Uniswap reaches 165 bps on BTC and 54 bps on ETH at the largest size bucket, two orders of magnitude wider than the order-book alternative.
• ONCHAIN VS. OFFCHAIN.
  Hyperliquid (onchain CLOB) is competitive with offchain incumbents across both BTC and ETH. The structural divide is not onchain vs. offchain — it is order book vs. AMM-style execution.
• BROKER-DEALER DISPERSION.
  Execution quality is far from uniform across broker-dealers, particularly at larger fill sizes. One broker-dealer maintains mean spread under 1 bps across the full size range on both BTC and ETH. The remaining broker-dealers are competitive at small sizes but degrade significantly with trade size — a profile consistent with route-through rather than inventory internalization.
• ASSET SENSITIVITY.
  Cross-venue ranking is not asset-agnostic. CEX spreads run roughly 2x wider on ETH than BTC at every size bucket. AMM and PropAMM execution quality is a property of the (venue, asset) pair rather than the venue alone.
• TIME OF DAY.
  Time-of-day effects are systematic. Broker-dealers tighten during the Asia session and widen through the US session — one broker-dealer shows a 3–5x intraday ratio on BTC. CEX and onchain order books show the inverse pattern, consistent with serving different flow bases.
• DAY OF WEEK.
  Weekend microstructure diverges by category. CEX and Hyperliquid tighten on weekends while broker-dealers widen — the clearest empirical signal in the dataset that the two categories serve structurally different client bases.

The practical implication of these findings is that optimal venue selection is conditional on asset, size regime, and time of day simultaneously. Any single-axis benchmark — a pooled per-venue spread, a category-level league table, or a one-size-fits-all routing heuristic — will mis-state execution cost for a non-trivial slice of flow. A defensible best execution regime in this asset class must evaluate cost conditional on (venue, asset, size, time) jointly, not marginally.

Historical and Current Context

Best Execution (BestEx) is a foundational principle of modern financial markets, representing the obligation of brokers to execute their clients' investment decisions with diligence, fairness, and transparency. FINRA's Rule 5310, formally effective in 2012 but in practice the direct successor to NASD Rule 2320, requires that brokers "shall use reasonable diligence" to execute trades such that "the resultant price to the customer is as favorable as possible under prevailing market conditions."

The National Best Bid and Offer (NBBO) consolidates quotes from exchanges into a single best bid and offer, providing a national reference price against which execution quality can be measured. Regulation NMS, promulgated by the SEC in 2005, modernized this framework for the electronic trading era. Rule 611 prohibits trading centers from executing orders at prices inferior to protected quotes displayed on other venues; Rule 605 requires market centers to publish monthly execution quality statistics; and Rule 606 requires broker-dealers to disclose order routing practices, including Payment for Order Flow (PFOF) arrangements.

In frontier markets such as digital assets, analogous best execution frameworks have not yet been fully developed. The absence of a consolidated, NBBO-equivalent feed for crypto was a recurring theme during the prolonged era of Bitcoin spot ETF rejections. The digital asset market has since matured considerably, driven by regulatory developments both domestically and abroad. The European Union's Markets in Crypto-Assets Regulation (MiCA) entered into force in 2024, establishing the first comprehensive crypto market structure framework in a major jurisdiction, including mandatory best execution requirements under Article 78.

In the United States, the passing of the GENIUS Act in July 2025 — establishing a framework for stablecoins — and the advancing CLARITY Act — which passed the House in July 2025 and cleared the Senate Banking Committee 15-9 on May 14, 2026, now awaiting a full Senate floor vote — represent progress toward regulatory normalization that has already led to deepening institutional adoption.

The advent of tokenized securities and the SEC's consideration of an "innovation exemption" for tokenized stocks has further heightened the need for tools to measure execution quality consistently and comprehensively — across offchain and onchain venues alike.

Research Overview

This study analyzed realized intra-trade price dispersion across a heterogeneous set of crypto execution venues. The motivation is to compare execution quality on a like-for-like basis across centralized exchanges, broker-dealers, onchain orderbook venues, automated market makers, and proprietary market makers — venues that do not share a single consolidated tape, and that in several cases do not publish an order book at all.

Across the February 1 to April 29, 2026 window, centralized exchanges and broker-dealers dominate the small- and mid-size regime with sub-1 bps median spreads. Onchain venues are competitive only at the small end: Hyperliquid is the only non-CEX venue to print in the same order of magnitude as the offchain incumbents, while Uniswap-style constant-product AMMs run an order of magnitude wider in BTC and a half order of magnitude wider in ETH. The most striking single finding is that the Solana proprietary AMM in the dataset is roughly 5x tighter on ETH than on BTC — evidence that PropAMM execution quality is anchored to which asset the pool operator inventories, rather than to the PropAMM venue category itself.

‍

1.

Venue Category and Size Sensitivity

‍

1.1

Overview

Pooling all taker-side candles at the one-second frequency, the five venue categories order themselves consistently across BTC and ETH. At small trade sizes, broker-dealers and centralized exchanges are functionally indistinguishable on spread. The divergence emerges with scale — and at institutional sizes ($100K+), the gap between venue categories is not marginal. It is structural.

• Broker-dealers (BD-1, BD-2, BD-3). Tightest at the small-ticket end, with BD-2 at 0.17 bps on BTC and 0.27 bps on ETH. The internalization model lets retail-sized flow trade against operator inventory at the desk's mid-mark; what shows up as spread is microstructure noise on the post-fill quote refresh rather than a real venue cost.
• Centralized exchanges (CEX-1, CEX-2). CEX-1 (0.56 bps BTC, 1.13 bps ETH multi-fill mean) anchors the cross-venue benchmark. CEX-2 prints tight headline numbers (0.07 bps BTC, 0.13 bps ETH) but on a fraction of the volume; the tightness reflects long stretches of inactivity rather than a real execution-cost advantage.
• Onchain order book (Hyperliquid). Hyperliquid (0.82 bps BTC, 1.71 bps ETH) is competitive with mid-tier CEXes on BTC and roughly 1.5x wide on ETH. It is the only onchain venue in the dataset whose execution cost is in the same order of magnitude as the offchain incumbents.
• Proprietary AMM (SolFi on Solana). SolFi prints 14.36 bps on BTC but only 2.57 bps on ETH. The five-fold asymmetry is not a category property; it reflects what the operator is willing to inventory.
• Constant-product AMM (Uniswap on Ethereum). Uniswap is the widest venue in every regime and the only venue where >$100K candles cross into double-digit-percent territory: 165 bps on BTC, 54 bps on ETH at the largest size bucket.
  ‍

1.2

AMM Execution: Constant-Product and Proprietary

The AMM category encompasses two structurally distinct execution models, each with different implications for institutional flow.

Constant-product AMMs (Uniswap). The curve-priced model compounds both a high starting spread and a steep size slope — the only venue category in this dataset that is simultaneously wide at small sizes and significantly wider at large sizes. At $100K+ notional, Uniswap reaches 165.54 bps on BTC: two orders of magnitude wider than the order-book alternative at the same size.

Proprietary AMMs (SolFi). A proprietary AMM presents as a venue but operates as a single operator's inventory position. SolFi charges materially more on BTC than ETH at small sizes — an asymmetry explained entirely by the pool operator's inventory choice, not venue architecture. At institutional sizes, this asymmetry converges, but the structural risk remains: approving a PropAMM as a routing destination is a position contingent on one operator's undisclosed balance sheet, not a venue decision.

UNISWAP MEAN SPREAD ON >$100K BTC: 165 bps vs. 2.96 bps on the best onchain order book. Same size. Same asset. Same moment in time.

PropAMM execution quality is a property of the (venue, asset) pair — not the venue alone. Approving a PropAMM as a routing destination is not a venue decision. It is a position contingent on one operator's undisclosed balance sheet.

‍

1.3

Broker-Dealer Execution: The Size-Slope Divergence

Evaluated on pooled headline spreads, broker-dealers appear nearly identical. The pooled headline is dominated by retail-sized candles where every broker-dealer internalizes efficiently. The divergence becomes visible only in size-bucketed analysis.

The 206x slope on BD-2 reflects classical limit-orderbook behavior: tight at the top of book, expensive past local depth. BD-1's 7x slope is consistent with genuine inventory internalization — a model that absorbs trade size without proportional cost escalation.

The same leaderboard ordering holds on ETH, with tighter absolute spreads but equivalent relative structure. On ETH, BD-1 holds 0.75 bps at >$100K while BD-2 and BD-3 reach 8.39 bps and 11.15 bps respectively. The gap is proportionally similar; the absolute levels are lower because ETH, while more volatile, presents a different pool of liquidity providers.

This pattern carries a disclosure implication. Under SEC Rule 606, broker-dealers must disclose order routing practices including Payment for Order Flow arrangements. Digital asset markets carry no equivalent obligation. A broker-dealer whose headline spread appears tight at retail sizes but routes large institutional blocks through external liquidity at super-linear cost faces no requirement to disclose that architecture. The slope is observable in the data. The cause requires direct inquiry.

At sub-$1K, all broker-dealers cluster inside a basis point. At >$100K, one holds 0.98 bps while others reach 6.17 bps. The gap is invisible in any pooled number. It appears only in size-bucketed analysis — which most institutional evaluation frameworks never request.

‍

2.

Cross-Asset Routing: BTC and ETH Are Not the Same

The cross-venue ranking is not asset-invariant. Every venue's relative position shifts between BTC and ETH, and the magnitude of the shift is largest precisely where it costs the most — at institutional trade sizes and in the AMM and PropAMM categories.

‍

2.1

BTC — Cross-Venue Ranking and Trade-Size Sensitivity

• Below $1K. CEX-2 (0.01 bps), CEX-1 (0.01 bps), and BD-2 (0.03 bps) are essentially tied. Hyperliquid (0.08), BD-1 (0.13), BD-3 (0.35) follow inside a single basis point. SolFi PropAMM (2.84) and Uniswap (2.15) are already an order of magnitude wider before any sizing pressure has been applied.
• $1K – $10K. CEX-1 widens to 0.15 bps; BD-2 to 0.27; Hyperliquid to 0.38; BD-1 to 0.36; BD-3 to 0.83. SolFi (10.30) and Uniswap (6.11) widen disproportionately faster.
• $10K – $100K. The order-book-based group is still inside 2 bps (CEX-1 at 0.89, Hyperliquid at 1.11, BD-3 at 1.58, BD-2 at 1.36); BD-1 is the standout at 0.31 bps. SolFi prints 8.80, Uniswap 18.46.
• Above $100K. CEX-1 at 3.30 bps, Hyperliquid at 2.96 bps, BD-3 at 5.97 bps, BD-2 at 6.17 bps, BD-1 at 0.98 bps — BD-1 remains the tightest broker-dealer at large size, and Hyperliquid is competitive with the deep CEX. SolFi at 13.26 bps and Uniswap at 165.54 bps are structurally unsuitable for institutional-size BTC flow.

‍

2.2

ETH — Cross-Venue Ranking and Trade-Size Sensitivity

ETH preserves the gross ordering of venue categories but reshuffles individual venues and tightens some of the slopes. Two findings are structurally significant:

• Hyperliquid inverts vs. BTC at institutional size. At >$100K ETH, Hyperliquid (6.39 bps) is the cheapest book in the entire dataset — meaningfully tighter than the deep CEX at 12.72 bps. On large BTC, Hyperliquid is mid-pack. A routing rule that treats both assets the same will pay the wrong rate on one of them every time.
• SolFi PropAMM is effectively broker-dealer-quality on ETH at mid-sizes. At $1K–$10K on ETH, SolFi at 1.09 bps is mid-pack with the order-book venues for the first time in the entire dataset. At institutional sizes, however, SolFi converges back to 14.36 bps.

‍

2.3

BTC vs. ETH: Three Cross-Asset Patterns

• CEX spreads run roughly 2x wider on ETH at every size bucket. CEX-1's multi-fill mean is 0.56 bps on BTC versus 1.13 bps on ETH. Any cross-venue ranking framework requires an asset-conditional intercept.
• AMM execution quality flips with the underlying asset's chain. Uniswap on Ethereum is dramatically wider on BTC than on ETH (roughly 3x at the >$100K bucket) because BTC on Ethereum is a wrapped synthetic with shallower native pools. AMM and PropAMM execution quality is a property of the (venue, asset) pair, not the venue alone.
• The broker-dealer tier is consistent across assets in ranking, shape, and leader behavior — but differs in level. BD-1, BD-2, and BD-3 land in the same order on both assets. The difference is that on ETH, absolute spreads are tighter and the gap between the best and worst broker-dealer narrows.

Any framework that evaluates venue quality without specifying the asset is answering a question that was not asked. The cross-asset shift at the institutional end is a 2x–10x reordering, not a noise effect.

‍

3.

Time-of-Day Effects

All broker-dealers in the dataset tighten during the Asia window and widen through the US session, regardless of where the operator's desk is regulated. The pattern is consistent enough that it is likely a property of the broker-dealer category as it interacts with the daily volatility cycle.

BD-1’s comparative flatness across the day is consistent with its internalization model: pricing off internal marks rather than CEX top-of-book means less exposure to US-session volatility. The Asia-tight/US-wide pattern in BD-2 and BD-3 may reflect thinner one-way taker flow in Asian hours reducing adverse selection, or denser cross-exchange arbitrage keeping prices aligned during those hours - the mechanism is not directly observable from candle-level data, but the empirical pattern is robust across 89 days.

The session table compresses the hour-of-day pattern into three overlapping windows. The Americas-widest/Asia-tightest ordering is the dominant pattern but is not universal. Hyperliquid on BTC is a notable exception: its European overlap (07–16 UTC) is the tightest single window at 0.71 bps, reflecting deeper book activity from European market makers. Hyperliquid is technically not accessible from the US, which makes this effect structurally plausible. SolFi PropAMM on BTC inverts the pattern entirely - its Americas session is its tightest window at 13–15 bps flat - but this is a local minimum inside an always-expensive venue, not a routing solution.

Sessions overlap by design. A candle in the 07–09 UTC window contributes to both the Asia and Europe session rows; a candle in the 13–16 UTC window contributes to both the Europe and Americas rows. This preserves the natural overlap of global trading desks rather than forcing artificial boundaries.

One broker-dealer shows a 3–5× intraday spread ratio on BTC. Its Asia session is its best. Its US session is its most expensive. The mechanism is not fully observable. The cost is.

‍

4.

Day-of-Week Effects

Crypto markets trade seven days a week, so any weekday/weekend dispersion is likely real microstructure rather than calendar noise — particularly when the analysis spans 89 days of continuous data. The CEX and onchain CLOB venues tighten on weekends; broker-dealers widen. This is the clearest empirical signature in the dataset that the two categories serve structurally different client bases.

BD desks quote tighter when institutional clients are at their screens; CEX and onchain venues see retail-heavy weekend flow that is price-taking across both sides and tightens the realized range. An alternative explanation is that weekend tightening for CEX and onchain CLOB reflects fewer market makers and arbitrageurs active - likely those tied to more traditional financial institutions - leaving a cleaner retail-dominated order flow.

Uniswap on ETH shows the clearest AMM weekend anomaly: Sunday is the widest day at 20.50 bps mean, while Saturday is the tightest at 12.70 bps. The spread between them  -  7.80 bps  -  is wider than the entire multi-fill mean spread of any broker-dealer in the dataset at any size bucket. Sunday appears to function as the DeFi ecosystem’s market open: the day used for weekly portfolio repositioning, with large block trades through shallow constant-product pools driving the spike.

A routing engine that treats Saturday the same as Tuesday compounds a systematic, avoidable cost across every weekend of the year. Weekend routing should be category-aware: CEX and onchain CLOB flow is structurally cheaper on weekends; broker-dealer flow is structurally more expensive.

‍

5.

The Single-Axis Benchmark
Blind Spot

The findings in Sections 1 through 4 share a common characteristic: none of them are captured by a standard pooled per-venue spread. A single-axis benchmark is constitutionally incapable of detecting what the data shows, because the cost is conditional - it varies with size, session, day of week, and asset simultaneously.

The standard approach: compute a pooled per-venue spread, averaged across all trade sizes, times of day, days of the week, and both assets. The number is technically accurate. It is also operationally useless for any trade that deviates from the average - which is to say, every trade that actually matters.

Any single-axis benchmark — a pooled per-venue spread, a category-level league table, or a one-size-fits-all routing heuristic — will mis-state execution cost for a non-trivial slice of flow. A defensible BestEx regime must evaluate cost conditional on (venue, asset, size, time) jointly, not marginally.

‍

5.1

Regulatory Context

The regulatory environment is accelerating in ways that make the benchmarking gap increasingly consequential. MiCA Article 78 is already in force across the European Union. The SEC's 2026 examination priorities explicitly name execution quality as a review focus. The Atkins NMS roundtable in September 2025 reinforced that best execution must be multidimensional, not rule-based. The CLARITY Act cleared the Senate Banking Committee in a 15-9 bipartisan vote on May 14, 2026 and now awaits a full Senate floor vote.

An examiner asking to demonstrate best execution on a $300,000 BTC block routed at 2pm US session on a Saturday through a broker-dealer of record will not be satisfied by a pooled headline spread. The tokenized securities development deserves particular attention: securities trading simultaneously on onchain and offchain venues cannot rely on NBBO-style infrastructure that does not yet exist for onchain markets. Institutions building a conditional EQ framework now for BTC and ETH will have the architecture ready when the next asset class demands it.

‍

6.

Cost Summary

The figures below represent direct execution costs only — the observable spread gap between what was paid and what was available. The base case applies to a hypothetical institution running $1B in annual digital asset volume, with a BTC/ETH split of 85/15 — consistent with the distribution observed across broker-dealer flow in this dataset. Large-block fractions (>$100K notional) follow the same dataset distribution: 47% of BTC flow and 25% of ETH flow at institutional size.

The base case applies to a hypothetical institution running $1B in annual digital asset volume, with a BTC/ETH split of 85/15 - consistent with the distribution observed across broker-dealer flow in this dataset. Large-block fractions (>$100K notional) follow the same dataset distribution: 47% of BTC flow and 25% of ETH flow at institutional size. Scale linearly to your own book.

‍

6.1

AMM Routing — Large BTC Flow

On a $1B annual book, approximately $400M of BTC flow falls in the large-block (>$100K) category. The gap between Uniswap and the best available order book at that size is 162.58 basis points — analyzed, not estimated.

‍

6.2

Findings 2–5 Combined

The basis point gaps in findings 2 through 5 are real and empirically documented. Their significance is not primarily the per-trade cost - it is that they are systematic, consistent across 89 days of observed data, and increasingly the subject of regulatory examination under MiCA Article 78 and the SEC’s 2026 examination priorities.

Finding 2 dominates at $207K - the cost of choosing the wrong broker-dealer for large BTC trades. That choice is indistinguishable in pooled headline spreads. It is visible only in size-bucketed analysis, which most institutional evaluation frameworks never request and most vendors never volunteer. On a $5B book, this finding alone produces over $1M in annual avoidable cost. At 10% AMM allocation, finding 1 adds another $3.2M. Combined: $4.2M per year, none of it on any fee schedule, none of it in any standard TCA report.

‍

6.3

The Benchmarking Multiplier

Every figure above is recoverable: identify the routing pattern, update the policy, capture the spread going forward. Finding 6 — the static benchmark — is what prevents that identification from occurring. A pooled per-venue spread will not surface the 162.58 bps AMM gap at large size. It will not surface the 5.19 bps broker-dealer selection gap. It will not surface the session differential or the weekend effect. It reports a number that is technically accurate and operationally blind.

‍

‍Conclusion

Optimal execution quality in digital asset markets is conditional on four variables simultaneously: venue, asset, size, and time of day. This is the architecture of a mature market structure framework — the same conditional logic that underpins best execution in equities, fixed income, and FX, adapted to an asset class that trades across five structurally different venue types, twenty-four hours a day, seven days a week, onchain and off.

The principles have not changed. Fairness, transparency, diligence — these still define what good execution looks like. What has changed is how those principles must be evidenced and operationalized in a market without a consolidated tape, without protected quotations, and without a mandatory routing competition layer.

In March 2026, Solidus Labs published Rethinking Best Execution in Digital Assets — introducing the industry's first Execution Quality (EQ) framework, designed to translate traditional BestEx principles into the fragmented reality of digital asset markets. The data in this report is the empirical foundation that framework requires: a systematic, cross-venue, cross-asset measurement of what execution quality actually looks like across eight venues, five categories, and 89 days of live market data.

The institutions that build conditional execution quality frameworks now — evaluating cost across venue, asset, size, and time of day jointly, not marginally — will not just reduce execution cost. They will build the audit trail, the governance posture, and the regulatory defensibility that the next phase of institutional digital asset adoption demands.

A defensible execution quality regime is not a theoretical ideal. The data to build it exists. The methodology to apply it exists. The regulatory pressure to adopt it is no longer approaching — it has arrived.

This report is a companion to Solidus Labs' March 2026 publication, Rethinking Best Execution in Digital Assets — the industry's first institutional framework for execution quality in digital asset markets. The Solidus EQ product operationalizes the principles described in both reports: objective measurement, contextual intelligence, and supervisory alignment across on-chain and off-chain venues alike.

‍

Methodology

Study Design

This study analyzed realized intra-trade price dispersion across a heterogeneous set of crypto execution venues. The motivation is to compare execution quality on a like-for-like basis across centralized exchanges, broker-dealers, onchain orderbook venues, automated market makers, and proprietary market makers  -  venues that do not share a single consolidated tape, and that in several cases do not publish an order book at all.

The full dataset covers February 1 through April 29, 2026 (UTC), encompassing 89 calendar days of continuous taker-side trading activity across eight venues.

Core Metric: Intra-Candle Spread

For each venue, taker-side trades were grouped into one-second time windows (candles). For each window, a spread was computed as:

spread (bps) = (high − low) / midpoint × 10,000
where midpoint = (high + low) / 2

This captures the realized price range over which the venue actually transacted within the window - not the quoted bid-ask the venue advertised at the top of book. The two are related but not identical: a venue may quote a wide spread but execute within a narrow realized range if all trades occur near the mid, or may quote a tight spread but move its transactable price significantly within a second if large flow hits. Single-trade candles produce high = low and therefore a spread of zero by construction. These are excluded from multi-fill averages. The time-regime cuts restrict to multi-trade candles before averaging, which sharpens the question: when this venue moved its transactable price within a second in this regime, how far did it move?

Venues

Trade-Size Buckets

Time Regime

Candles were aggregated along three time dimensions simultaneously:

Sessions overlap by design. A candle in the 07–09 UTC window contributes to both the Asia and Europe session rows. A candle in the 13–16 UTC window contributes to both the Europe and Americas rows. This preserves the natural overlap of global trading desks rather than forcing artificial boundaries.

Limitations

• Realized range vs. quoted spread. The candle high−low blends venue microstructure with short-horizon price volatility. A quiet venue with wide quotes may appear tight if the underlying price did not move within the window.
• Taker-side only. Maker fills are excluded by design. The result is an execution-cost view (what the price taker paid), not a market-making view.
• Cross-quote-pair comparison. Venues are compared across different quote currencies (USD, USDT, USDC). The stablecoin basis can move independently of the underlying crypto price.
• Equal-weighted aggregation. Candles are averaged with equal weight rather than notional weighting.
• Coverage period. Results reflect February 1 – April 29, 2026 only. Findings may not generalize to different volatility regimes, market conditions, or materially different liquidity environments.