The Illusion of Best Execution in Fragmented Books

Best execution is often treated as a clean metric. Match the best price. Minimize slippage. Execute quickly. But in fragmented markets, this logic frays. The price is no longer a number—it’s a moving target scattered across venues, feeds, and latency windows. Execution quality becomes probabilistic, not absolute.

An order hits the book. At that moment, a different venue shows a better quote—one that the trader didn’t see or couldn’t reach in time. Technically, the order wasn’t filled at the best price. But operationally, the system acted as intended. This tension between intent and outcome defines the illusion.

Fragmentation by Design

Exchanges are not built to cooperate. Each venue publishes its own feed, runs its own matching engine, and manages its own latency profile. Some rebroadcast data from others. Some don't. Some offer colocation; some throttle access. This creates a fractured surface where “best” is always relative to visibility and reach.

In FX, this fragmentation is an accepted constraint. In crypto, it's amplified by inconsistent APIs, fluctuating latency, and unpredictable throttling. An aggregator tries to reconstruct a synthetic book, but it does so with partial information. That book is never complete. The moment it’s formed, it’s already out of date.

Routing Isn’t Neutral

Smart order routers are supposed to solve this. They analyze depth across venues and direct flow accordingly. But these routers face structural limitations. Some venues impose hidden delays on inbound orders. Others reject aggressive cross-venue strategies. Some simply don't update their books fast enough.

Even in cases where a router identifies the best quote, reaching it in time is a different task. Market data and order placement often travel on separate paths. Even a few hundred microseconds of drift between the two creates mismatches. A quote that looks live is already gone by the time the order lands.

The Hidden Cost of Partial Visibility

Market makers adjust spreads based on what they see. If one venue shows a wide spread and another shows a tight one, the difference may reflect more than liquidity—it may reflect asymmetry in access. A trader without access to the faster venue sees a worse market. A trader with direct access can act on information before it decays.

This is where best execution breaks down. The benchmark assumes a shared reference. But in practice, every trader sees a different version of the market. One based on local infrastructure, stale data, and asymmetric access rights.

Liquidity Isn’t Fungible

A quote on Venue A doesn't guarantee fillability for someone plugged into Venue B. Latency matters. Queue position matters. Even if both venues show the same price, fill probability is not equal. A 500 BTC bid on an exchange might fill instantly for a colocated participant and reject a cloud-based one milliseconds later.

Some firms internalize this and adjust their fill models accordingly. They discount quotes that are likely to decay. They model hit ratios by venue. They prioritize execution certainty over theoretical price advantage. Others stick to best price and eat the slippage.

The illusion persists because the benchmarks ignore these dynamics. They assume every price is reachable. They assume every fill is fungible. They assume the book is static long enough to measure against. None of these assumptions hold.

Regulatory Lag

In traditional finance, Regulation NMS attempted to codify best execution by mandating routing to the best displayed price. But even there, fragmentation creates edge cases. Dark pools, hidden orders, stale quotes, and SIP latency distort the surface.

In digital asset markets, no such framework exists. Exchanges decide what they report, how often, and with what level of granularity. Some hide volume. Some aggregate order flow. Some update in bursts. The result is a market that claims transparency but operates on opaque mechanics.

Modeling Around the Gaps

Quant teams with execution mandates often rebuild their own metrics. They simulate what “best” means under their latency and connectivity constraints. They treat price as a function of access. They backtest slippage against reachability, not just displayed quotes.

These models often diverge sharply from vendor-provided execution quality reports. That’s not a bug. That’s a better alignment with how execution actually works. A fill that looks worse on paper may be optimal under real constraints.

Local Optimality over Global Illusion

Execution engines that chase a global best often trade against ghosts. A system optimized for reachable price—factoring in venue behavior, latency window, and cancel-replace behavior—delivers better realized outcomes. Not necessarily prettier fills. Just more predictable ones.

Some desks split order flow accordingly. They allocate fast flow to venues where responsiveness matters. They allocate passive flow to venues with deep queues. They don’t chase the best quote—they chase the most reliable execution window.

This approach doesn’t align with traditional best execution metrics. But it aligns with profit and risk containment.