When Two "Gold Standard" Instruments Disagree, Who's Right? (Hint: it’s not the relevant question)
Part 2 of 2: The Fairness Attestation Gap by Eric Dorre, co-Founder - FairConnect
Part 1 of this series asked why weekly compliance testing can't keep pace with infrastructure that never stops trading. Part 2 asks the harder question underneath it: even when you do measure continuously, can you trust the measurement itself?
This isn’t about whether to use optical backscatter reflectometers (OBRs) or optical time-domain reflectometers (OTDRs); both are considered gold-standard instruments, with the OBR being the more precise and the OTDR able to see further. Both can have their place in a certification cycle. Nor is this an argument for which type of OBR or OTDR to use. Point two different machines at the same cross-connect and ask each to certify its length and latency. Both will give you an answer. Those answers will not always match, sometimes by several nanoseconds (and implicitly by several meters).
Precisely declaring where one fiber run ends and the next begins is a genuinely hard problem for either instrument, just for different reasons which are beyond the scope of this note. Different inferred lengths do not imply a defect in the instrument. It's the predictable result of software interpreting physics, making a judgment call under conditions that are inherent to any live data center environment. It doesn’t mean these instruments are unreliable; take an OTDR as an example, since it's the instrument most exchanges have relied on to date. It performs a few basic functions: some are well-suited to continuous latency monitoring, and others are more challenging for exchange operators (but there’s hope!).
OTDR Function 1 - Measurement:
First, OTDRs inject a light pulse into a fiber cable and map a fiber's loss and reflectance by timing light that returns, creating a “signal trace” for that fiber. Operators can configure sampling parameters such as pulse width to manage trade-offs depending upon network topology. Examples include optimizing between “dynamic range” against “dead zones”: longer pulses reach farther but blur near-end events; shorter pulses resolve closely spaced events but cut range. The good news: once settled on a configuration, OTDRs create very consistent and reliable signal traces, so consistent that we consider them “fingerprints” (we will come back to this later).
OTDR Function 2 - Interpretation:
A second function embedded into OTDRs is a convenience feature for their operators: “event-distance interpretation.” The key word here is “interpretation”. OTDRs are most often handheld and used “in the field”, where the operator wants immediate feedback for the result of their test. Handheld devices interpret the signal echo described above, and reasonable algorithms can disagree about exactly where one spliced, daisy-chained run of fiber "ends" and the next connector begins. Every OTDR vendor's auto-detect logic makes that call a little differently. None of them is lying, and none of them is even wrong, exactly. It’s a matter of “interpretation,” and interpretations often differ.
Which raises the first question this piece is really about: When two certified instruments disagree, which one do you believe?
The honest answer is that it's the wrong question. Getting two machines to auto-interpret the same absolute length is genuinely hard and arguably beside the point, at commissioning or anytime after. And it makes for stifling, often unhelpful conversations between venues and their clients that impede progress towards the end-goal. But that question and the resulting debate happens between trading firms and exchanges because nobody can answer the most relevant question, starting at commissioning and every day after: is my cross-connect (always) the same latency as everyone else's?
That's a repeatability problem, not solely an accuracy problem. An OTDR, when the device is rack-mounted and does not require a human’s physical interaction with the fiber optical cables, repeatedly handles it well and at scale, creating reliably consistent signal traces we discussed above. Even better, we no longer rely upon the interpretive function of the OTDR. Once a baseline trace is established for each line at commissioning (pick your favorite OBR or OTDR for that task), comparisons of current-state traces to the baseline answers the persistency question over time. If nothing changes, we assume the line remains as it was certified; if something has changed, the rack-mounted, in-line device will trigger an alert to investigate further. An exchange is equipped with information to support an investigation and help suggest appropriate actions to remediate the differences.
That distinction is the difference between an expensive, arguably unsolvable problem and a tractable, buildable one. This is what FairConnect is built to facilitate, and it empowers the network engineers in the data center to work on an exception-basis, rather than:
wasting countless hours searching for exceptions and relying on inconsistent interpretations that are difficult to document in the compliance audit trail, or
never getting around to continuous latency monitoring and validations, at all
A point-in-time certificate, however carefully measured, answers a question about the day it was issued. It doesn't answer the one that matters to a trading firm sizing up a venue right now: is this still true today…for EVERYONE? The fix isn't a more precise OTDR or OBR. The fix is treating every cross-connect as a living asset with a signal trace “fingerprint” checked continuously. It’s about accuracy and repeatability at-scale. This is what FairConnect has built for and automates for venues and shared network providers.
Fair, in other words, has to be a status you can check everyday. Not a certificate handed once, a while ago, and never looked at since.
If your last cross-connect measurement was a one-time event rather than a standing practice, you already know which side of this you're on. The most sophisticated trading firms are no doubt keeping you honest about your latency certification of the lines they lease from you; what they are all wondering is whether their peers’ lines remain within the same equalization tolerance. If repeating this process and documenting it for auditability is more challenging than it should be and a distraction from other important tasks, we'd welcome the conversation to support you.