An Average Is Not a Picture

Freight rail may be the most thoroughly measured industry in America that cannot see itself.

Every week, the Class I railroads file their operating performance with the Surface Transportation Board. Train velocity. Terminal dwell. Cars online. Original estimated time of arrival. Industry spot-and-pull. Public, standardized, comparable across carriers.

None of it tells a railroad what is happening on its network right now — which is the whole case for real-time operational intelligence for freight rail. The industry is not short of measurement. It is short of measurement that arrives while the decision is still open.

What metrics measure freight rail network performance?

The Surface Transportation Board requires Class I railroads to report train velocity, terminal dwell, cars online, original estimated time of arrival, and industry spot-and-pull performance. Carriers manage against operating ratio, asset utilization, cost per carload and return on invested capital. Together these describe the network’s health — after the reporting period has closed.

An average is not a picture

Terminal dwell is reported as a mean. So is velocity. That is the correct way to report a network to a regulator — it is comparable, auditable, and resistant to being gamed. It is a very poor way to run one.

Consider what a dwell average of twenty-six hours actually conceals. It conceals that most cars cleared in eleven. It conceals that a few hundred sat for four days behind a crew shortage that resolved itself before anyone escalated. It conceals which of those cars was the one carrying a customer’s entire month of inbound, and it conceals which one was a tracked vehicle bound for a seaport of embarkation on a movement order with a date attached to it.

The average is true. It is simply true about a week that has already finished. And a railroad cannot recover a missed connection retroactively, any more than a refinery can un-shut a pipeline.

This is not a criticism of the Surface Transportation Board’s metrics, which are well designed for the job they were designed for. Regulatory reporting exists to make carriers comparable and accountable over time, and a mean is the right instrument for that. The error is entirely on the operator’s side of the table — it is the quiet assumption that because a thing is measured for the regulator, it is therefore visible to the railroad. Those are two very different things, and the gap between them is where the week goes.

Every hour a car sits is capacity, service and capital standing still.

And the economics of that sentence are unforgiving in a way outsiders consistently underestimate. Railroading is a fixed-cost business measured by a ratio — operating ratio, expenses over revenue — in which a point is an enormous amount of money and everybody in the building knows it. A car that sits is not merely late. It is an asset earning nothing, occupying track that something else needed, consuming yard capacity, and degrading a service metric that a customer is contractually watching. Rail freight visibility and risk intelligence is not a reporting nicety in that context. It is the difference between managing the network and reconciling it.

Why a leaner network needs real-time operational intelligence for freight rail

Over the last decade the industry ran one of the largest operating experiments in its history. Precision scheduled railroading — whatever an individual reader thinks of it — did what it set out to do to the ratio. Assets came out. Trains got longer. Cars and locomotives were retired or stored. Networks were rebuilt around scheduled, point-to-point service rather than tonnage-driven departures.

Set the merits aside entirely; the operating consequence is not in dispute and both sides of the argument agree on it. A leaner network has less slack. That is the objective, not a side effect — slack is the technical name for assets that are not earning. But slack is also what a network absorbs surprise with, and a railroad that has removed it has, by the same act, removed its own margin for finding out late.

Which changes what visibility is worth. In a network carrying spare locomotives, spare crews and spare capacity in the yards, a problem discovered on Wednesday could still be absorbed by Friday — expensively, inelegantly, but absorbed. In a network run to a tight ratio, the same problem discovered on the same Wednesday propagates. The car misses the connection. The train departs short. The crew hits its hours. The terminal congests. And the number that finally surfaces the whole cascade is a dwell average, published the following week, describing damage that is already done.

The lean network is not a worse network. It is a network with a much higher return on knowing sooner — and a much steeper penalty for the weekly report being the first place anyone finds out.

Six systems, one network, no single view

The obstacle is not that railroads lack instrumentation. It is that a modern Class I runs six or seven instrumented domains that were built at different times, by different vendors, to answer different questions — and were never asked to answer one question together.

Each of those is competently run. The gap is not in any of them. It is in the seam — which is nobody’s metric, nobody’s budget, and nobody’s job title.

And the seam is exactly where the customer lives. A shipper does not experience train velocity; a shipper experiences whether the car showed up. When a plant runs out of inbound and takes a line down, the cause is almost never a single dramatic failure. It is a crew gap that met a congested terminal that met an equipment defect, three domains and two days apart, each visible to somebody and none of it visible to anybody at once. The shipper then hears about it from their own production line, which is the worst possible place to hear about it, and calls a railroad that is still assembling the explanation.

The overlay model is not a preference in this industry; it is a precondition. A dispatch system is a safety-critical installation under federal oversight. Positive train control took the better part of two decades and a congressional mandate to deploy. Any vendor whose proposal begins first, replace your dispatch platform has not met a railroad. The picture is built on top of what is already moving trains, or it is not built at all.

Forty-one thousand miles of it is a national security asset

There is a dimension to this network that most rail technology never touches, and it is the one USADG was built for.

The Department of War designates a Strategic Rail Corridor Network — STRACNET — of more than 41,300 miles of track serving 141 defense sites, maintained with the Federal Railroad Administration and last updated in 2023. The reason is blunt: heavy and tracked vehicles deploy by rail. When armor moves to a seaport of embarkation, it moves on the same rails, through the same yards, behind the same crews, as everything else the railroad is carrying that day.

Which produces a strange and underexamined situation. A movement with a national-defense obligation attached to it is, inside a Class I’s operating systems, mostly just a car. It dwells in the same yard, waits on the same crew, and is subject to the same congestion as the load of lumber ahead of it. The railroad is not being negligent. It is doing exactly what its systems were built to do, and its systems were built to optimize a network, not to distinguish a priority movement inside it.

The Department of War has understood the exposure for fifty years — the Railroads for National Defense program exists precisely because the rail system is privately owned, commercially optimized, and indispensable to mobilization all at once. It reviews abandonments and mergers for their effect on defense sites and works through the Surface Transportation Board to mitigate them. It is careful, long-running, structural work.

But it operates on the map, not on the day. It tells you which corridors must exist. It cannot tell you that a particular movement, on a particular Tuesday, is sitting in a yard in Illinois behind a crew shortage, four days from a sailing that will not wait. Nobody is watching that — not because anyone is failing, but because the map and the movement are two different problems and the industry has only ever instrumented one of them.

Defense supply chain rail risk monitoring is the discipline of making that distinction visible — to the railroad, and to the program office on the other end that has a date and a ship. It is the same problem the sustainment and readiness conversation runs into everywhere: readiness is computed from assumptions about a supply chain that nobody is watching in real time.

Between the detectors, the network is guessing

There is a quieter version of the same problem, and it sits underneath every metric discussed so far.

A railroad knows the condition of its rolling stock at the points where it has installed something to measure it. Bearing detectors, wheel impact detectors, equipment identification readers — fixed installations, spaced along the line, each producing an excellent reading at the instant a car passes over it. Between those points, the network is not measuring anything. It is interpolating.

That is a perfectly rational design, and it is how the industry has always worked, because instrumenting fixed points was the only thing that was economically possible. But it means a railroad’s picture of its own assets is a series of snapshots taken at intervals, with inference filling the gaps — and a bearing does not fail at a detector. It fails wherever it happens to be, which is almost always somewhere else.

Every operator in this industry already knows what that gap costs, and none of them needs it spelled out.

Where the picture becomes a capability

Visibility alone is a dashboard, and Class I railroads are not short of dashboards. What a network actually needs is two things, and USADG builds both.

ReflexOS™ is the operating picture. Dispatch, yard, equipment, track and defense-movement telemetry, resolved into one live view on top of the systems already running. Nothing is ripped out. It exists to close the interval between something going wrong on the network and somebody being able to do anything about it.

The quantum sensing layer is what feeds it physical truth instead of inference. This is the same instrument capability USADG applies to subsurface exploration, pointed at a different problem: telemetry from the asset itself, continuously, rather than a reading taken at the moment it crosses a fixed point on the ground.

Those two things are not a bundle. They are one architecture. An intelligence platform with no instrument beneath it is a layer of opinion built on somebody else’s assumptions — and a sensor with no platform above it is a data feed nobody is reading. The sensing companies in this sector do not operate an operational intelligence platform. The platform companies do not build sensors. USADG operates both, which is why they are described here as one product rather than two.

Where a railroad wants it, Sigma Shield cybersecurity resilience layers on top of that — and the words on top of are doing real work. A Class I already has a security program, a security team, and a stack it has spent years building. Sigma Shield is an added overlay to that existing cybersecurity, not a proposal to replace it, and it is pointed at the boundary where the corporate network meets signaling, wayside and dispatch. It is the same operational-technology seam every heavily instrumented industry has, with different acronyms and identical consequences. It is an option, not a precondition — the picture and the sensing stand on their own.

What that yields is the same thing it yields in every other estate USADG works in — and it is the same structural problem, whether the asset is a refinery running six unreconciled systems or a Class I running seven. Continuous measurement, resolved into one picture, on top of what is already there: the real-time operational intelligence platform, tuned to an industry where the asset is measured in hours and the network is measured in a ratio.

The report tells you what the network did last week. The railroad has to decide what it does in the next four hours. Keep the network moving.