Near Valence, in the Rh\u00f4ne valley between Lyon and the Mediterranean, the Route nationale 7 follows the track of the Via Agrippa. The Roman road was surveyed around 20 BC using a bronze cross hung from a pole with four plumb bobs \u2014 a groma \u2014 which let the surveyor project a straight line across uneven ground. The modern road was surveyed using satellite positioning. Both produced the same line.<\/p>\n\n\n\n
Neither engineer was consulting the other’s work. One was solving a military logistics problem for the Roman Empire. The other was solving a traffic problem for a modern republic. They arrived at the same answer because the answer was always the same answer, and the valley that made it the only answer in 20 BC has not changed since.<\/p>\n\n\n\n
Call it imitation if you want. Or coincidence. But there’s a more precise explanation \u2014 one that makes Roman road persistence feel less like archaeology and more like physics, and one that has direct consequences for where EU infrastructure money is going right now.<\/p>\n\n\n\n
The gromatici were Roman military surveyors, and their job was to lay roads through conquered terrain as fast and as permanently as possible. Their instrument, the groma, was elegant in the way useful things are elegant \u2014 a bronze cross suspended from a pole with plumb bobs hanging from each arm, allowing the surveyor to establish right angles and sight along straight lines over open ground. Not primitive guesswork. A specific technical solution to the problem of projecting geometry onto uneven landscape at scale.<\/p>\n\n\n\n
But the geometry was only the expression of something more durable: engineering principles derived from geography that hasn’t changed.<\/p>\n\n\n\n
Three principles governed Roman road placement, all of them attested in the archaeological record and all derivable from engineering common sense. Follow ridges and high ground where terrain permits \u2014 not for visibility, not for defensibility, but for drainage. A road on a ridge drains naturally after rain. A road in a valley floods, erodes, and requires constant attention. Roman roads in Britain appear on high ground with a regularity that wasn’t random; it was the cheapest maintenance decision available to engineers who’d be gone by the time the repairs were due.<\/p>\n\n\n\n
Cross rivers at the narrowest navigable point. Bridge-building is expensive; minimizing span length means minimizing cost and construction time while keeping the crossing commercially useful. The Thames crossing at Londinium was chosen because it was the first tidal-free point upstream where a ford, and then a bridge, was viable \u2014 making it simultaneously the best military crossing, the best commercial port location, and ultimately the capital of a country.<\/p>\n\n\n\n
Connect administrative nodes using the most direct geometry the terrain permits. Roman roads were built for military logistics, but military logistics and commercial efficiency are substantially the same problem. The network of nodes \u2014 provincial capitals, legionary fortresses, supply depots \u2014 mapped closely onto what would become the commercial centers of medieval and modern Europe.<\/p>\n\n\n\n
None of this has changed. Rivers are still in the same places. Ridges are still ridges. The Rh\u00f4ne valley hasn’t moved. The Thames is still tidal below London. When medieval engineers built roads, they largely followed existing tracks rather than conducting fresh surveys from landscape principles \u2014 but the existing tracks were following the same terrain. When 18th-century turnpike commissioners surveyed Britain for new toll roads, they arrived at the same corridors. When 20th-century motorway planners ran modern topographic surveys, they kept making the same choices.<\/p>\n\n\n\n
The Roman engineers weren’t being visionary. They were being correct. And correct, applied to a landscape that doesn’t change, turns out to last.<\/p>\n\n\n\n
\nThe surveyors' gap<\/strong>\n\nAfter Rome's administrative collapse in the 5th century, systematic geometric road surveying largely ceased in Western Europe for roughly a thousand years. Medieval road construction followed existing tracks rather than conducting fresh surveys from first principles, and the quality gap is visible \u2014 medieval roads are rarely as straight, as well-drained, or as well-sited as Roman ones. The organized surveying tradition the gromatici represented wasn't substantially recovered in Western Europe until 17th-century cartographic methods. This means the Roman network wasn't preserved, improved, or consciously continued. It was mostly abandoned \u2014 and then re-derived from the same landscape by engineers who had never heard of a groma.<\/code><\/pre>\n<\/blockquote>\n\n\n\nThree routes<\/h3>\n\n\n\n
Apply the mechanism to specific terrain and the pattern stops being abstract.<\/p>\n\n\n\n
Ermine Street ran from Londinium north through Lindum Colonia (Lincoln) to Eboracum (York), and the modern A1 follows the same corridor. From London, the A10 initially traces the Ermine Street alignment northward through Hertfordshire before the A1 takes over further north, where the route runs slightly west of the Roman alignment to skirt the low-lying ground near the Humber. Both roads converge on York by crossing the River Ouse at substantially the same point, because there is no other sensible point to cross it. The corridor has carried freight on this basis ever since the Romans established it \u2014 under the Roman name, then as the medieval Great North Road, then as a turnpike route, then as a modern A-road. The engineers who surveyed it for each successive iteration were not copying their predecessors. They were solving the same problem.<\/p>\n\n\n\n
The geographic case is straightforward: the route from London to York runs through a natural corridor defined by the Pennines to the west and the Fenland marshes to the east. Heavy wheeled traffic in either direction has one viable route. The river crossings at the Lea, the Nene, and the Ouse converge with the Roman road at the same points where the valleys narrow to a bridgeable width. Every road-builder who surveyed this terrain chose the same corridor, because the Pennine passes and the Fens aren’t alternatives \u2014 they’re disqualifications.<\/p>\n\n\n\n
The Rh\u00f4ne case is starker. The Via Agrippa’s southern axis ran from Lugdunum (Lyon) through Vienne, Valence, Mont\u00e9limar, Orange, and Avignon to the Mediterranean, following the Rh\u00f4ne flood plain south. The N7, and the A7 autoroute that now largely replaces it, follows this alignment explicitly; near Valence the N7 follows the track of the Roman road. The modern road runs through the same sequence of towns \u2014 Vienne, Valence, Mont\u00e9limar, Orange \u2014 partly because those towns exist where the Roman road established settlements, but mostly because east of the Rh\u00f4ne is the Alps and west is the Massif Central. The flood plain is the only viable north-south corridor for wheeled freight between Lyon and the Mediterranean that doesn’t involve a mountain crossing. Roman engineers, Napoleonic engineers, TGV planners, and autoroute designers all arrived at the same valley. There is no other valley. This isn’t path dependency in the sense of habit or inertia. It’s path dependency in the sense of a canyon.<\/p>\n\n\n\n
The third case is geographically coarser. The Roman road from Colonia Agrippina (Cologne) southeast toward Mogontiacum (Mainz), then east along the Danube through Castra Regina (Regensburg) toward Vindobona (Vienna), ran the length of the Limes Germanicus \u2014 approximately 550 km from the Rhine to the Danube. The modern A3 autobahn runs Cologne through Frankfurt to Regensburg; Austrian motorways extend the corridor to Vienna. The specific road-to-road alignment here is less precisely documented at map level than the first two cases, and the claim is properly a geographic one: the Rhine-Danube axis is Europe’s dominant east-west hydrological corridor, and any land route connecting the two basins through central Europe must transit Bavaria. The Alps enforce it to the south; the German highlands close the northern options. Roman military engineers building along the imperial frontier and 20th-century autobahn planners building for industrial Bavaria were working within constraints that haven’t changed. Different purposes, different centuries, same corridor.<\/p>\n\n\n\n
\nThe Fosse Way: what partial persistence looks like<\/strong>\n\nThe Fosse Way ran diagonally across Britain from Isca Dumnoniorum (Exeter) to Lindum Colonia (Lincoln) \u2014 a military frontier road built to supply a temporary boundary. Modern A-roads do follow substantial sections of it: the A46 between Lincoln and Leicester, the A429 through Gloucestershire and Warwickshire, the A37 between Bath and Ilchester, the A303 near Ilchester where excavations in 2024 confirmed Roman road material beneath the modern surface. But no major motorway spine follows the full diagonal. Exeter and Lincoln did not become dominant national freight nodes. The Fosse Way persisted regionally without generating the recursive lock-in \u2014 road to city-growth to more road to more city \u2014 that turns a Roman alignment into a permanent freight artery. The mechanism fired partially. The result is partial. Persistence was conditional, not automatic.<\/code><\/pre>\n<\/blockquote>\n\n\n\nThe city in the middle<\/h3>\n\n\n\n
The roads themselves didn’t survive. Most of the Roman road network in Western Europe was abandoned, looted for materials, or simply buried during the post-Roman centuries. What survived wasn’t the road. It was what the road had made.<\/p>\n\n\n\n
Every major Roman road required mansiones \u2014 staging posts at intervals of roughly a day’s march, approximately 37-44 km. These were maintained stations with stabling, food, and accommodation for official travelers and military supply chains. They were also, unavoidably, nuclei. Communities grew around these posts for the same reason they always do: foot traffic, buildings, an administrative presence to appeal to. The post became a village, the village a town, the town a city.<\/p>\n\n\n\n
The spacing was calibrated to the pace of a courier on horseback or an ox-cart with a load \u2014 the practical limits of a working day in the ancient world. But the spacing created something that outlasted the carts entirely: a hierarchy of settlements distributed across the landscape at regular intervals, positioned at the points of greatest topographic utility. The most important mansiones, at major river crossings and route junctions, became cities. The minor ones became market towns. The smallest became the kind of village that still, two thousand years later, sits on an English ridge with a name and a pub and no obvious reason to exist, other than that a road once stopped there.<\/p>\n\n\n\n
Londinium sat at the first tidal-free Thames crossing. Lugdunum \u2014 Lyon \u2014 sat at the confluence of the Rh\u00f4ne and the Sa\u00f4ne, the single most commercially important junction in Roman Gaul. Colonia Agrippina sat at the primary Rhine crossing for east-west traffic. Vindobona sat at the Danube narrows, the command post controlling movement between the western and eastern halves of the empire. Eboracum sat at the navigable Ouse, the northernmost major Roman logistics center in Britain. Every one of these is now the dominant infrastructure node in its modern national context.<\/p>\n\n\n\n
The recursive mechanism runs in one direction only: road creates city-seed; city grows; next road must serve city; city grows further; all subsequent infrastructure converges on city; city grows further still. By the time medieval road-builders were working, the cities were already there, and roads go where cities are.<\/p>\n\n\n\n
By the time 19th-century railway engineers were surveying routes, those cities were large enough to determine the course of entire national rail networks. By the time 20th-century motorway planners were working, the cities had planning departments, parliamentary constituencies, and the political weight to ensure the motorway came through. The Roman legions planted the seed. The compound interest ran without interruption.<\/p>\n\n\n\n
Guy Michaels and Ferdinand Rauch demonstrated this through a controlled comparison published in The Economic Journal in 2018. They compared Britain and France. After Rome’s administrative collapse, Britain’s urban system broke down almost entirely \u2014 Roman towns were abandoned, and medieval urbanization recovered at different locations: river crossings that gained new commercial importance, market towns, ecclesiastical centers with no connection to Roman sites. France, where continuous urban occupation persisted through the post-Roman transition, shows much stronger alignment between Roman road locations and modern city locations. The conclusion is stark: Britain’s urban network was reset; France’s wasn’t. Where the reset happened, road persistence weakened with it. The roads weren’t persisting on their own. The cities were pulling the roads along.<\/p>\n\n\n\n
This is also why the Middle East and North Africa provide such a clean negative case \u2014 which becomes clearer once the numbers are in.<\/p>\n\n\n\n
Measuring the inertia<\/h3>\n\n\n\n
Impressionistic historical argument has one persistent vulnerability: someone can always say the examples were cherry-picked. The Rh\u00f4ne valley, Ermine Street, the Rhine-Danube corridor \u2014 maybe Roman engineers just happened to pick routes that any modern engineer would independently choose. Maybe the overlap reflects geographic convergence rather than infrastructure inertia. The economics literature has been working on this.<\/p>\n\n\n\n
Luca De Benedictis, Vania Licio, and Anna Maria Pinna measured the statistical relationship between Roman consular road density and modern motorway density across Italian provinces. Their 2023 paper in the Journal of Regional Science found an elasticity of 0.36 to 0.39 \u2014 a province with 10% higher Roman road density has 3.6-3.9% higher modern motorway density, a relationship that survives controls for terrain, population, and economic development. They distinguish two pathways: direct construction on or alongside Roman roads, and a city-mediated route \u2014 Roman roads created cities, cities attracted motorways. The city-mediated effect is the dominant one. The mechanism from the previous section isn’t just plausible; it’s the larger of the two measurable effects.<\/p>\n\n\n\n
The stronger test comes from outside Europe. A 2022 study in the Journal of Comparative Economics \u2014 Dalgaard, Kaarsen, Olsson, and Selaya \u2014 found that Roman road density predicts current road density and current economic activity \u2014 measured via nighttime light intensity from satellite imagery \u2014 for European regions. The same relationship does not hold for the Middle East and North Africa. Roman road networks in MENA were just as dense as in Europe. But wheeled transport was effectively abandoned in MENA from the first millennium CE onward, replaced by camel caravans that don’t require paved roads. When wheeled transport returned with modern trucks, the caravan routes didn’t align with Roman roads, and the Roman road network predicts nothing about where MENA infrastructure ended up.<\/p>\n\n\n\n
Same Roman roads. Comparable geography. The difference was whether the chain of continuous use held. In Europe it held; in MENA it broke \u2014 not because the roads were inferior or the terrain less suitable, but because a different mode of transport supplanted wheeled freight before the urban nodes could compound. When camels replaced carts, the road network became irrelevant and slowly disappeared. When trucks arrived centuries later, they followed the camel caravan routes, not the vanished Roman ones. The chain had broken at the medieval junction, and no subsequent force reconnected it.<\/p>\n\n\n\n
Where it broke, geography alone was insufficient to recreate the pattern. The road mattered. The accumulated urban weight of two thousand years of compounding traffic mattered. The geography of the Rh\u00f4ne valley and the Pennines is necessary but not sufficient. Without the cities that grew at the junctions, it would have been just terrain.<\/p>\n\n\n\n
\nMapping the empire at scale<\/strong>\n\nThe Itiner-e dataset, published in Scientific Data in 2025 by Pau de Soto, Adam Pa\u017eout, Tom Brughmans, and 17 co-authors, maps 299,171 km of Roman roads with certainty metadata per segment. This is the data infrastructure enabling the recent wave of quantitative studies described in this section. The gap between \"we have Roman road maps\" and \"we can run a regression on them\" is not trivial \u2014 Itiner-e is what closes it, and the research program it enables is still accelerating.<\/code><\/pre>\n<\/blockquote>\n\n\n\nMeasuring the persistence is one thing. What it costs, and who pays it, is a different question entirely.<\/p>\n\n\n\n
What it costs to be right the first time<\/h3>\n\n\n\n
The A7 autoroute between Lyon and the Mediterranean is one of the most heavily trafficked freight routes in Western Europe. Truck volumes through the Rh\u00f4ne valley corridor are among the highest on any French autoroute; the stretch through Valence is a routine bottleneck that transport planners have been managing for decades. This is not, primarily, a planning failure. It is a geography problem with no available solution.<\/p>\n\n\n\n
East is the Alps. West is the Massif Central. There is no somewhere else. The valley that left Roman and modern engineers the same narrow option is also a valley where the capacity problem cannot be solved by rerouting \u2014 you can widen the road, add lanes, add rail alongside it, but you cannot move the problem to a different corridor because the different corridor doesn’t exist. The Lyon-Marseille TGV runs the same valley for the same reason the A7 does.<\/p>\n\n\n\n
Under Regulation (EU) 2024\/1679, the former Rhine-Alpine and North Sea-Mediterranean TEN-T corridors were merged into the North Sea-Rhine-Mediterranean corridor. The NSRM’s western axis runs from Paris through Lyon to Marseille and Fos-sur-Mer \u2014 the EU’s formally designated primary south-of-France freight spine, with multi-billion-euro infrastructure investment committed for road and rail upgrades. From Lyon south, this axis follows the Rh\u00f4ne valley, through the freight bottleneck at Valence, along the same ground the Via Agrippa crossed two thousand years earlier.<\/p>\n\n\n\n
The investment isn’t misdirected. The geography that made the Via Agrippa the correct answer in 20 BC makes the Rh\u00f4ne corridor the correct answer now. But the cities that grew along the route over two millennia have made the upgrade vastly more expensive and complicated than the geography alone would require. Every kilometer of widening, every new rail line, every noise barrier and tunnel requires negotiating with municipalities that exist precisely because Roman engineers chose this corridor in the first place. The Romans got the routing right \u2014 and in doing so, created the very urban fabric that now makes improving it so difficult.<\/p>\n\n\n\n
The Romans did not cause the congestion on the A7. They caused the cities that cause the congestion on the A7. Lyon exists at the Rh\u00f4ne-Sa\u00f4ne confluence because a Roman engineer chose it as the hub for a Gaulish road network. The subsequent two thousand years of urban growth at that point converted a military engineering decision into an economic fact that no empire, technology, or planning instrument has been able to revise. The engineers of 100 AD were solving their problem \u2014 efficiently, correctly. The planners of 2025 are managing the accumulated consequences of that solution, not because they’re trapped by reverence for Rome but because the terrain gave the same answer to both of them, and the cities grew up at the answer.<\/p>\n\n\n\n
That’s not legacy. It’s inertia. And there’s a difference: legacy can be discarded.<\/p>\n\n\n\n
Back in the Rh\u00f4ne valley, the trucks on the A7 don’t know they’re following the Via Agrippa. The Alps are to the east. The Massif Central is to the west. There is one road.<\/p>\n\n\n\n
Infrastructure inertia is not sentimental. It has no awareness of what came before. It is just the compounding weight of correct answers \u2014 the same constraint, running forward without expiry.<\/p>\n\n\n\n