Across the three New Jersey exchange pairs in Borrill's table, light needs 90 to 143 µs to cross in vacuum. A trader using the consolidated tape has to live with that floor. Borrill builds a larger claim from it: the National Best Bid and Offer is physically undefined, and a new exchange design could remove the co-location premium. We accept the floor. The text does not carry the larger claim.
Regulation NMS, adopted in 2005, requires a consolidated best bid and offer across 16 exchanges; Rule 611 protects it against trade-throughs. Comparing quotes from separated venues requires a decision about which quote is "current." Special relativity makes the order of events outside each other's light cones depend on the observer's frame. Borrill names the concealed choice "engineered simultaneity." His definition has three conditions: the system compares spacelike-separated events, uses an implicit convention to do so (arrival order at the SIP), and presents the result as an objective measurement. Under Theorem 1, quote events at different exchanges separated by less than their light-time can yield different computed NBBOs in different inertial frames. He connects this to Lamport's happened-before relation: "The NBBO requires a total order. Physics provides only a partial order."
The dollar case rests on cited work. Aquilina, Budish and O'Neill estimate about $5 billion a year in global latency-arbitrage extraction. A follow-up study by the same authors puts high-frequency traders' win rate in stale-quote races at 96 to 99%. The printed abstract promises more than a diagnosis: adjudication on each participant's own worldline, with simultaneity "constructed by a completed bilateral handshake," would take the prize out of the co-location race. The paper supplies no dataset, model or backtest of its own.
How long are New Jersey quotes disconnected?
The light-time table earns its place. Carteret to Secaucus spans 27 km, or 90 µs in vacuum and 135 µs in fiber. Mahwah to Carteret spans 43 km, or 143 µs and 215 µs. Nasdaq and NYSE quotes updated within 143 µs of each other cannot have caused each other. Direct feeds run at O(10) µs, while HFT execution takes under 10 µs. Fast traders operate inside the window Borrill identifies.
Figure 1, his flagship example, uses the wrong venues for this claim. It pairs an NYSE quote with a CME Group futures quote at Aurora, 1,180 km and 3,940 µs away. CME futures do not enter the equity NBBO. The paper describes five primary data center clusters for U.S. equity exchanges, yet its fifth is CME Group at Aurora. Its own table calls that site a futures reference.
Theorem 1 also asks the reader to move between frames. Suppose quotes across Mahwah and Carteret arrive 10 µs apart. Our arithmetic puts the speed needed to reverse their order at more than 10/143 of light speed along that line, about 7% of c. Every exchange, the SIP and every trader share one rest frame far more closely than that. The frame-dependence is genuine physics; no market participant experiences it.
What the SIP actually orders
"The SIP does not observe the market; it observes its own input buffer." Borrill is right there. Arrival order at one point, though, is frame-invariant. The printed abstract accepts the same property for matching engines, describing their arrival order as "a total, local, frame-invariant order." The SIP is likewise a single point. At any given sequence number, its published NBBO has the same value in every frame.
The body says the NBBO "is a function of arrival times at a single location, not of emission times at multiple locations." Location and rule are the dependencies it identifies. Its arrival-order rule is "a valid synchronization convention but not a unique one," and "a SIP located elsewhere, or using a different tie-breaking rule, would produce a different NBBO." Agreed. Moving the SIP changes the distances quotes travel to reach it, so arrival order may change. Every inertial observer agrees on that new order. Receiver latency explains the first change; a design choice explains the second. Both remain frame-invariant. Relativity of simultaneity enters neither. The best price across all exchanges at a given instant has no frame-independent answer, but the SIP does not compute that quantity.
The reported delays strengthen the latency reading. The paper puts SIP delivery at about 1,128 µs after emission, versus tens of microseconds on direct feeds, a ratio "exceeding 50:1." It expressly rejects the ordinary interpretation. "This is not a speed advantage in the conventional sense," it says, calling the gap a "frame advantage." We disagree. The 1,128 µs delay is roughly eight times the 143 µs light-time across the widest New Jersey pair. Processing and network hops account for most of the gap: a timelike delay at the receiver. Calling it a "frame advantage" changes the label on information latency. Relativity permits a consolidator much faster than today's.
The handshake promised in the abstract
The abstract promises adjudication on worldlines, market centrality that becomes relative, a falsifiable prediction distinguishing the two designs, and limits arising from light speed, fragmentation and incumbent rents. The body beneath it has another title, "Engineered Simultaneity," and carries the mark Version 3.0, March 2026. We did not find the handshake protocol, the prediction or the fragmentation analysis there. Its conclusion describes Theorem 1, the Lamport connection, the $5 billion gap, and engineered simultaneity as a concept extending beyond financial markets. Recognizing the mistake is, it says, "a prerequisite for principled system design." This is the close of a diagnostic paper. It leaves the printed title and abstract's design promise unanswered.
The missing design matters to a trader. Across 16 venues, two counterparties can see quotes in different orders. The abstract never explains what a bilateral handshake does with those orders. Until it does, the proposed disappearance of the co-location prize lacks a mechanism.
Five billion dollars, borrowed from London
The paper identifies the source of its $5 billion figure: London Stock Exchange message data, extrapolated globally by the cited study. That study observes races about once per minute per liquid security, with a modal duration of 5 to 10 µs. Its messages come from one exchange, the London Stock Exchange. Borrill nevertheless attaches the global estimate to the U.S. NBBO.
The cited U.S. result is $14.4 million and measures a narrower difference. Bartlett and McCrary report that the SIP NBBO matches their "true" NBBO 97% of the time, with $14.4 million in aggregate gains on $3.7 trillion of volume. Borrill gives no period for that figure. His response is that "they have not measured the gap between convention and reality; they have measured the gap between two conventions." That works as an epistemological objection. The $14.4 million remains a measured dollar amount on U.S. trades. Borrill also says magnitude has no bearing on the impossibility, then identifies the 3% of mismatches as the place where $5 billion is extracted. The 3% describes U.S. SIP mismatches; the $5 billion is a global estimate extrapolated from London data. The sentence combines different samples.
We could not test any of this. Our data are minute bars: one bar covers 60 seconds, about six million 10 µs windows. A test of the design would require venue-level quote and message-arrival timestamps. Nor can a change to matching rules be replayed on prices produced under the old rules.
A specified handshake protocol, paired with a prediction that the 96 to 99% HFT win rate would fall under it, would change our view. For now, Borrill has a sound warning about the SIP's input buffer. The exchange design remains in the abstract.