Which pass did Hannibal take?

Carthage, on the coast of what is now Tunisia, fought Rome three times for control of the western Mediterranean. In the second of those wars Hannibal marched to Italy overland. What is known of the march comes from two ancient historians. Polybius was a Greek who wrote within about seventy years of it, questioned men who had been on it, and went over the Alps himself to see the ground. Livy was a Roman who wrote two centuries later and drew on the same Greek tradition. Between them they tell the following story.

Hannibal left New Carthage in Spain with an army that Polybius, from an inscription Hannibal himself set up, puts at forty-six thousand men, with thirty-seven elephants. From the crossing of the Rhône he marched four days upstream to a district the Greeks called the Island, a fertile triangle of land between the Rhône and a second river. Two brothers were quarrelling over its throne. Hannibal backed the elder, who repaid him with weapons, clothing and boots for the mountains and escorted the army with his own troops as far as the foot of the Alps, because the Carthaginians were uneasy about the next tribe on the road, the Allobroges. Ten days along the river brought the army to the first climb. The Allobroges held the heights above the defile by day and went home to their town at night; Hannibal occupied the heights after dark, the column was mauled in the defile next day, and he took the town, whose stores fed the army for two or three days. Four days on, tribesmen came out with branches and wreaths and offered to guide him. They led the army for two days, then attacked it in a steep gorge, rolling rocks down from above, and Hannibal spent a night with half his force on a bare white rock. On the ninth day of the climb he reached the summit and camped there for two days. The first snow was falling, and the Pleiades, a star cluster whose autumn setting was a calendar date for ancient farmers, were going down at dawn, which puts the halt in the last days of October or the first days of November. From the top Hannibal pointed his men to the plains of the Po, the flat land of northern Italy, and to the direction of Rome beyond.

The descent was worse than the climb. The path ran narrow and steep along a cliff, and a stretch of about 270 m had fallen away, an old break freshly enlarged. The previous winter's snow still lay there under the new fall. Hannibal camped on the ridge and set the men to building the path up. One day's work got the horses and mules through to pasture; three more days, with the Numidian troops working in relays, got the elephants over, "in a wretched state from hunger". Three days after that the army was on flat ground among the Taurini, the people around modern Turin, whose chief town it took after a three-day siege. The crossing had taken fifteen days and the march from Spain five months. Livy adds one story Polybius does not have: that the rock on the descent was split by fire and vinegar.

Neither historian names the pass. The route has to be worked out from the days, the distances and the landmarks, and the two place-names that would fix it are damaged in the manuscripts. Every modern translation calls the second river of the Island the Isère, which joins the Rhône at Valence; that is a conjecture made by the geographer Cluverius in 1624, and the manuscripts of both authors read something else. The height of the broken ledge, Livy's "1,000 feet", is a Renaissance restoration of a corrupt line. The argument has gone on since Josias Simler raised it in 1574. A count made in 1835 found ninety published opinions spread over five passes; Colonel Perrin added the Col du Clapier in 1883 by publishing the view from it; Peddie counted twenty-four modern writers in 1997 and found six for the Mont-Cenis, six for the Montgenèvre, five for the Clapier, four for the Petit St Bernard and three for the Col de la Traversette.

A pass, or col, is a gap in the chain low enough for a track to cross, and the candidates lie along the whole frontier between France and Italy. The sides divide first on the river. If the Island's river is the Isère, the army marched north-east up the Isère valley and crossed into the Susa valley west of Turin, by the Mont-Cenis or by the Col du Clapier beside it. The Clapier has been the usual answer, with a hedge, of ancient historians writing in English since Perrin; the Mont-Cenis was Napoleon's choice. If the Island lay further down the Rhône at Orange, as Livy's list of tribes implies, the army marched east up the Durance and crossed by the Montgenèvre, the easiest road into the Susa valley, by the Col de Larche further south, or by the Col de la Traversette, a high notch under Monte Viso at the head of the Po. The Traversette is the highest and hardest of the candidates and has had the most recent support: Gavin de Beer's books, a decade of earth-science papers led by William Mahaney, whose team cored a boggy meadow below the col and reported a layer churned by animals which they dated to 218 BC and identified as the passage of Hannibal's army, and a study of the energy cost of each route published in 2026. The two St Bernard passes to the north, which come down into the Aosta valley, carry the old Roman tradition that Livy himself rejected. No candidate has a majority, and the two most recent peer-reviewed contributions, one for the Clapier and one for the Traversette, point in opposite directions.

I read Polybius (book 3, chapters 39 to 60, and 34.10) and Livy (book 21, chapters 22 to 39) in Greek and Latin, in the standard Teubner and Oxford editions, which record what the manuscripts say, tabulated every number and place in one against the other, and traced the disputed readings to the editors who made them. I drew twelve routes over eight cols on a 90 m terrain model, measured them in stades (the Greek unit, about 180 m) and tested each against Polybius's day-counts; profiled each descent for ground steep and broken enough to match his cliff path; computed by sight-lines what each col can see of the Po plain; compared Polybius's old snow with the modern snowline and the glacier record; and computed the Pleiades' setting for 218 BC. I recalibrated every published radiocarbon date from the Traversette meadow and checked the rest of that team's claims against what their papers report. Then I gave each of fourteen lines of evidence a likelihood range for each of eight passes, argued item by item, and combined them. The nine items that come from Polybius count at half strength in the headline run because they come from one author, and the model was also run at full strength, with each source collapsed to one item, with each item dropped in turn, under seven sceptical scenarios, and by sampling inside every range.

The result ranks the candidates but does not identify the pass. The Col du Clapier comes first at about 40 per cent, the Col de la Traversette second at about 30 and the Mont-Cenis third at about 15. The ranking depends on the Isère reading of the Island: with the Island at Orange the Traversette leads at about 54 per cent. The physical evidence claimed for the Traversette is consistent with any large body of animals at the meadow between about 380 and 150 BC and does not test the crossing. Each percentage rests on the likelihood ranges I set, and the panel below shows how far it moves when a range is changed.

The page begins with the two narratives, Polybius's and Livy's, read in the original with their manuscript variants, from the march to the foot of the Alps to the descent into the land of the Taurini. It then takes up the Island and the approach; the days and stades of the ascent against real terrain; what each candidate col can see of the Po plain; and the snow and the setting of the Pleiades. A short section collects what the other ancient writers say. The Traversette meadow and its radiocarbon dates have a section of their own. Then come the eight candidates with the people who have argued for each, the weighing of the evidence and how its numbers move, what would overturn the result, what is new here and what is not, where the page falls short, and the sources, editions and code.

Two connected accounts of the crossing survive. Polybius wrote within about seventy years of it. He says he questioned men who had been present and had gone over the Alps himself to see the ground (3.48.12), and he took the army's numbers from the bronze inscription Hannibal set up at Cape Lacinium (3.33.17-18, 3.56.4). Livy wrote two centuries after the event. His Alpine chapters follow Polybius closely in their sequence of incidents and are generally held to draw on the same Greek tradition, through the Roman annalist Coelius Antipater, whose own source for the march was Silenus, a Greek who travelled with Hannibal. Everything else, from Strabo's list of passes to Ammianus, is a summary of these two or a one-line aside. So the two accounts are one tradition and do not count as independent witnesses to the route. Where Livy has a name that Polybius lacks, the name may come from Coelius, from Livy's own geography, or from a mistake.

The march to the foot of the Alps

Polybius gives the distances twice. In the geographical summary of the march he counts 1,600 stades from Emporion to the Rhône crossing, then:

The narrative then supplies the pieces. After the crossing Hannibal marched "along the river, away from the sea, as if towards the east, as if making for the interior of Europe" (3.47.1). Four days from the crossing he reached the Island:

The Island's elder prince, whom Hannibal had restored to power, re-equipped the army with weapons, clothing and boots "for the crossing of the mountains". And, "most important of all", he escorted it with his own force until it neared the ascent of the Alps, because the Carthaginians were uneasy about "the march through the Gauls called Allobroges" (3.49.11-13). Then:

Livy's version of the same stretch is the one passage in either author that names peoples between the Rhône and the Alps:

Livy then has Hannibal cross the Durance "in flood" and reach the Alps from the Durance "by a mostly level route, with the good will of the Gauls who lived there" (21.32.6: ab Druentia campestri maxime itinere ad Alpes cum bona pace incolentium ea loca Gallorum pervenit). That sentence is the Livian counterpart of Polybius's ten days in flat country, and it sits oddly with the Durance, whose valley above Sisteron is not level.

The ascent

The first fight is with the Allobroges, at a defile where the tribesmen hold the heights by day and go home at night to a town close by. Hannibal seizes the heights after dark, the column is mauled in the pass the next day, and he takes the town, which feeds the army for two or three days (Polybius 3.50.3-3.51.13; Livy 21.32.8-33.11 calls the town a castellum quod caput eius regionis erat, "the fort that was the chief place of that district"). He rests a day and marches on. On the fourth day "those living around the pass" meet him with branches and wreaths. Guided by them for two days, he is attacked in "a difficult and precipitous gorge" where rocks are rolled down from above, and spends a night with half the army "at a certain strong white rock" (περί τι λευκόπετρον ὀχυρόν, 3.53.5; Livy 21.34.9 has the night without cavalry and baggage but no rock). Next day the enemy withdraw and he "advanced towards the highest passes of the Alps" (3.53.6). Wherever the elephants are in the column, the tribesmen keep off (3.53.8; Livy 21.35.3).

The view from the top

The descent

Hannibal camped "on the ridge" (περὶ τὴν ῥάχιν, 3.55.6) after clearing the snow, and set the men to build up the cliff-path. One day's work let the horses and mules through to snow-free pasture; three more days, with the Numidians working in relays, brought the elephants over, "in a wretched state from hunger" (3.55.7-8). Livy adds the story of the rock split with fire and vinegar (21.37.2-3), which Polybius does not have. Then "on the third day from the cliffs" the whole army reached the flat country (3.56.1; Livy 21.37.6 triduo inde ad planum descensum). The crossing of the Alps took fifteen days and the march from New Carthage five months (3.56.3; Livy 21.38.1).

Where he came down

Livy then gives the only ancient discussion of the pass itself:

Where the Island lay decides which passes are possible at all. Polybius's Island is the land between the Rhône and a second river. If that river is the Isère, the army marched up the Isère and the candidate cols are those of the Maurienne and the Tarentaise: the Mont-Cenis, the Clapier and its Petit Mont-Cenis variant, the Petit St Bernard. If it is the Aygues at Orange or the Sorgue near Avignon, the army marched up the Drôme or the Durance and the candidates are the Montgenèvre, the Traversette and the Larche. No col is on both lists. So the first thing to settle is what the texts and the distances allow.

The river's name does not settle it. No manuscript of either author reads Isère: Polybius's manuscripts give Σκάρας or Σκώρας, Livy's sarar, bisarar or arar, and Isara is Cluverius's conjecture of 1624 in both texts (see the notes above). Holstenius's explanation, that ΟΙΣΑΡΑΣ was misread as ΟΣΚΑΡΑΣ, is plausible as palaeography, but it remains a conjecture. Conway's case for the Sorgue rests on Strabo's Σούλγας (4.1.11); Colin (1904) and Wilkinson (1911) had already put the Island between the Rhône and the Sorgue at L'Isle-sur-la-Sorgue, and Torr (1924) put it at the Durance. De Beer's Aygues has no ancient name to rest on at all. The river has to be found from the geography.

Polybius gives three measurements. The crossing was made "where the stream is single, about four days' march from the sea" (3.42.1). The Island was reached "on the fourth day" from the crossing (3.49.5). And the whole way from the crossing to the foot of the ascent was 1,400 stades (3.39.9), of which the last 800 were the ten days from the Island (3.50.1), which leaves 600 stades, 107 to 111 km, for the four days to the Island. Schweighäuser made that subtraction in 1789 and it has been the Isère party's main argument ever since. Measured along the valley floor on the route model, the distances are:

The 600 stades fit one combination only: a crossing near Roquemaure and an Island at the Isère, 111 km and 623 stades apart. That crossing is also the one about four days from the sea, since the delta mouth is some 90 km downstream of Roquemaure and about 55 km from Beaucaire. The Aygues fits four days of marching only if the army crossed at or below Beaucaire and then averaged 11 to 14 km a day on flat ground, and Beaucaire is two days from the sea, not four.

The same numbers can be turned round. Polybius's next stage runs 800 stades in ten days, 80 stades or about 14.5 km a day. At that rate four days is 320 stades, 57 km, which is the distance from Fourques to the Aygues. So one of Polybius's figures is wrong, or the four days from the crossing were long days and the ten days beyond the Island were short ones. The ten days were spent in flat country with an escort and no enemy, and 14.5 km a day is slow for that. I read the 600 stades as the more reliable figure, because it is the difference between two stated totals rather than a rate, and because Polybius's "four days from the sea" for the crossing points the same way. But the arithmetic is not clean, and the weighing in the model section counts the Island as a strong item and stops short of a decisive one.

Two other details favour the Isère. Polybius says the Island's third side was "mountains difficult to approach and to enter, almost inaccessible" (3.49.7): the Vercors and the Chartreuse close the Isère triangle in that way, while the small triangle between the Aygues and the Rhône has the low hills of the Baronnies behind it. And Polybius keeps the Island's people and the Allobroges apart. The prince of the Island escorts the Carthaginians "through the Gauls called Allobroges" (3.49.13), and the Allobrogian chieftains attack only when the escort has turned home at the foot of the ascent (3.50.2-3). Pliny's list of the Narbonese peoples runs inland from the coast "the Tricorii, and further in the Tritolli and the Vocontii and the Segovellauni, next the Allobroges" (Tricorium et intus Tritollorum Vocontiorumque et Segovellaunorum, mox Allobrogum, 3.34), and puts Valence "in the territory of the Cavares" and Vienne with the Allobroges (3.36). An Island in the Valence plain belongs to the Segovellauni or Cavares, with the Allobroges beginning across the Isère, which matches Polybius's distinction. An Island at Orange puts the Allobroges a hundred kilometres to the north, and de Beer had to make the "Allobroges" of the ambush a loose term for Alpine Gauls in general.

Livy points the other way. His Island is the same four camps from the crossing, but he makes the Allobroges live "near by" and makes the quarrelling brothers Allobroges, and then he sends the army "left into the Tricastini, along the outer edge of the Vocontii, into the Tricorii" and across the Durance (21.31.9). The Tricastini are at Saint-Paul-Trois-Châteaux, on the Rhône south of Montélimar. From an Island at the Isère that is a march back down the Rhône. From an Island at Orange it is a turn to the left, north, and then east up the Drôme, over the Col de Cabre or the Col de Grimone into the Gap basin, and down to the Durance: de Beer's route. Livy's itinerary is coherent from the southern Island and incoherent from the northern one. The question is whether it is Livy's itinerary at all. The names do not occur in Polybius, and Livy's sentence that after the Durance the army reached the Alps "by a mostly level road" (21.32.6) describes the Isère valley better than the Durance above Sisteron. The Oxford editors thought the reckoning of days excluded the Isère; the Clapier school from Perrin onwards has treated the Druentia passage as an intrusion from the annalistic tradition, and I could read Walbank's discussion of it only in snippets. I count it as a second tradition in its own right, of modest weight, and the model section runs the case in which it is trusted in full.

Ammianus (15.10.11) repeats Livy's peoples in Livy's order, "through the Tricasini and the far edge of the Vocontii to the passes of the Tricorii", and adds that Hannibal was guided by "natives from among the Taurini". He is a witness to the Livian tradition rather than to the route.

Polybius counts the days over the mountains too. From the foot of the ascent he gives nine days to the summit, two days there, a descent that begins on the next day and reaches the broken path, one day making it passable for horses and mules and three more for the elephants, and then three days to the plain (3.53.9, 3.54.4, 3.55.7-8, 3.56.1). The whole crossing is fifteen days (3.56.3), which the stages do not add up to, and the distance is "about 1,200 stades" (3.39.10). I tested each candidate against these figures on the Copernicus 90 m elevation model, with the route drawn as a least-cost valley-floor path between named places and the beginning of the ascent taken at the gate that each school itself uses: Aiguebelle for the Isère routes, Albertville for the Tarentaise, Sisteron for the Durance, Crest for the Drôme, Martigny for the Valais. The code and every intermediate number are in the files listed at the end of this page.

Three of the figures in the table tell the candidates apart and two do not. The 800 stades from the Island do not: along the Isère they end between Montmélian and Aiguebelle, where the Combe de Savoie shuts and the Maurienne begins, which is the first ascent of the Mont-Cenis and Clapier theories and equally the turn into the Tarentaise; from an Island at Orange, down the Rhône and up the Durance, they end 15 km short of Sisteron. Every school can claim the figure to within 10 or 15 per cent. The 1,200 stades for the crossing do not tell them apart either, and the reason is in the text: no line satisfies both of Polybius's totals. The Isère routes give 1,500 stades for the 1,400 from the crossing to the ascent and only 620 to 670 for the 1,200 of the crossing; the Durance routes give 1,140 and 940. The Drôme approaches that Livy's itinerary implies need 1,450 to 1,540 stades from Crest to the plain and 25 km a day for nine days over two intermediate cols, which is outside anything the narrative allows. One observation of my own, which I could not check against Walbank: 1,200 stades is exactly fifteen days at the 80 stades a day of the previous stage, so the figure may be a computation from the day-count rather than a measurement.

The nine days of ascent separate the candidates only a little. From the gates, the ascent averages 4 to 16 km a day for every candidate, slow but consistent with a battle on the first day, a captured town, a day's halt, a second ambush and a night on a rock. The four days of descent are more useful. Polybius's army came off the cliffs with its animals starving and reached the plain on the third day. From the Clapier the plain below 400 m is 25 km away, from the Traversette 28, from the Mont-Cenis 32: six to eight kilometres a day. From the Montgenèvre it is 51 km, from the Petit St Bernard 72 km to Pont-Saint-Martin, from the Larche 79 km to the Cuneo plain: thirteen to twenty kilometres a day downhill with a wrecked army. The difference is the length of the confined valley below each col, and it does not go away if the descent is measured to the first villages instead of to the plain.

The broken path separates the candidates most. Polybius says the descent was narrow and steep and that a stretch of about a stade and a half, some 270 m, had fallen away where the track ran along a precipice, an old break freshly enlarged (3.54.7); Hannibal "built up the cliff" (3.55.6). At 90 m resolution the elevation model can say where a ledge above a drop exists and where it does not. The Clapier has a 500 m stretch of the descent at 50 per cent one kilometre below the col, the Clarea headwall; the Traversette has 51 per cent half a kilometre below the col, on the east face above Pian del Re, losing 900 m in 2.7 km; the Grand St Bernard has 54 per cent at 1.8 km. The Mont-Cenis has its steep ground 13 km beyond the col, where the plateau drops through the Ramasse to Novalesa, after 13 km of near-level lake basin. The Montgenèvre, the Larche and the Petit St Bernard never exceed 23 to 29 per cent anywhere on their descents: there is nothing on them that a landslip could have made impassable to a mule. The model cannot say whether a particular ledge fell in 218 BC, only where the ground is of the kind Polybius describes.

Polybius sets Hannibal's speech at the camp on the pass and has him point out "the plains of the Po" (3.54.2-3). Livy moves it to the next morning and to "a certain promontory from which there was a far and wide view" (21.35.8). Advocates of every pass have claimed the view for their col, usually from personal experience on a clear day, and sceptics have called it a rhetorical commonplace. Whether a col sees the plain can be computed from the terrain model. For each col I cast rays every quarter of a degree out to 120 km over the elevation model, allowing for the curvature of the earth and standard refraction, and asked whether any ground of the Po plain, defined as land below 400 m east of 7.3°E and north of 44.4°N, was in sight. I then repeated the test at every 250 m of the first 10 km of the descent, from knolls 100 m and 300 m above each col, and from every cell within 1.5 km of the col and up to 300 m above it, the ground a general with a day to spare could walk to.

Only two cols see the plain at all. From the Traversette it lies in sight across a 63-degree sector from east-north-east to east-south-east, the nearest plain 26 km away at Revello and Saluzzo and the farthest 120 km at the edge of the model; it is in sight from every point of the first 2.25 km of the descent, and from the knolls beside the col across 22 to 26 per cent of the horizon. That matches both authors without special pleading. From the Clapier the col itself sees a two-degree sliver of plain 80 km away through the notch of the Val Clarea, beyond Turin; from one kilometre down the descent, on the lip of the Clarea headwall, it sees 5 per cent of the horizon, and from the ridge 1.4 km from the col, at 2,712 m, 8 per cent. That matches Livy's promontory on the first descent morning and gives Polybius a view to point at, though not a panorama. The Mont-Cenis sees nothing from the col, nothing from the first 10 km of the descent and a half-percent sliver from three cells on the ridge above; its famous prospect of the Susa valley is from the plateau's far rim, 12 km beyond the col. The Petit Mont-Cenis, the Montgenèvre, the Larche, the Petit St Bernard, the Grand St Bernard and the Col Agnel see no plain from the col, from any point of the first 10 km of descent, from the knolls, or from any of the 47 to 91 cells scanned around each of them.

The result is insensitive to the threshold for "plain" (raising it to 500 m adds the Cuneo terraces and the Canavese hills and opens no closed col), to the refraction constant, and to tree cover, which can only hide plain that the model shows. It does depend on the assumption that the observer stood within an hour's walk of the col, and it says nothing about weather. The Traversette has the view Polybius describes from the summit camp, the Clapier has the view Livy describes from a promontory on the first morning of the descent, and no other candidate has either.

Polybius dates the summit camp by "the setting of the Pleiades" (3.54.1), and Livy by the same star (21.35.6). For 218 BC at 45°N the true morning setting of Alcyone, the cluster's brightest star, falls on 29 October of the proleptic Julian calendar, and the first morning on which the cluster could be seen setting in the dawn falls between 4 and 9 November depending on the visibility threshold assumed, 6 November at the conventional 7.5 degrees. Two independent computations, Skyfield with the JPL DE406 ephemeris and PyEphem, agree to the day, and Columella's calendar of the first century AD, which has the Pleiades "begin to set at sunrise" on 20 to 21 October, "set" on 28 October and "set in the morning" on 8 November, agrees as closely as the 280 years and three degrees of latitude allow. So the army stood on the pass in the last days of October or the first week of November. The phrase is a farmer's calendar date, not a report of an observation made on the col, and its precision is about a week.

Polybius then says that on the descent the new snow of that year lay on "the snow that was there before and had lasted from the previous winter" (3.55.1), and Livy has "old untouched snow" under a shallow new fall (21.36.5). Whether last winter's snow survives to the end of October at a given col is a question about the snowline. In the modern climate the end-of-summer snowline of the French western Alps averages 3,035 m, ranging from 2,790 m in a cool year to 3,250 m in a hot one (Rabatel and colleagues 2013), and old snow survives in shaded hollows and avalanche cones a few hundred metres below it. Glacier records put 218 BC after the Iron Age advances had waned and within a century of a further advance around 100 BC, so the snowline was probably no higher than today's and may have been 100 to 250 m lower (Holzhauser, Magny and Zumbühl 2005; Le Roy, Ivy-Ochs and colleagues 2024). On those figures the Traversette, at 2,947 m, is at or above the snowline and old snow on its descent in late October is likely. The Clapier and the Grand St Bernard, at about 2,480 m, are 500 to 600 m below it and old snow on them is unlikely but possible in shade. The Petit St Bernard, the Petit Mont-Cenis and the Mont-Cenis, at 2,080 to 2,190 m, are 850 to 950 m below it, and the Larche and the Montgenèvre, at 1,990 and 1,854 m, more than a kilometre below it: old snow on those cols in late October would be exceptional in any climate the record allows. These bands are judgements rather than measurements; I found no station series of autumn snow patches for the southern French Alps. Two cautions on the text: Polybius puts the old snow on the descent below the summit rather than at the col, and his remark a few lines later that the heights of the Alps carry snow all year round (3.55.9) is a generalisation, so 3.55.1 may be a generalisation too. The model section runs that scenario.

Outside Polybius and Livy, the ancient evidence for the pass comes to six sentences. Each is given below in the original, with what it does and does not establish.

Two further scraps are sometimes cited. Strabo's description of the mountain of the Medulli (4.6.5), with "a great lake in hollow ground above, and two springs not far apart" from which the Durance and the Dora flow, has been claimed for the Mont-Cenis plateau and its lake and for the Savine basin below the Clapier. It is Strabo's geography of the Maurienne, garbled, and it does not mention Hannibal. Pliny's remark that the Po's source at Monte Viso is "a sight worth seeing" (3.117) shows only that the head of the Po valley was known and visited in the first century AD.

These passages, taken together, settle only a negative. The Petit St Bernard and the Grand St Bernard rest on a Roman tradition that Livy and Polybius both contradict, and on a folk etymology. They leave open the whole Cottian range from the Larche to the Mont-Cenis, with a lean towards the Susa valley from "through the Taurini" and a lean away from the Montgenèvre from Pompey's letter and Varro's order.

The Traversette is the one candidate with a physical case. Between 2011 and 2017 William Mahaney and a large team cored and sectioned a small alluvial mire at 2,580 m in the upper Guil, about 0.9 km west of the col and 370 m below it. They reported a churned, organic-rich bed at 38 to 50 cm depth, which they called a "mass animal deposition" bed, with elevated faecal lipids, a spike of Clostridia bacteria and a radiocarbon age model that "firmly places" the bed in 218 BC (Mahaney and colleagues, Archaeometry 2017, parts I and II; Mediterranean Archaeology and Archaeometry 2014 and 2017). Below the col on the Italian side they had earlier described a two-lobed rockfall which they identified with the landslip of Polybius 3.54.7 (2010; 2014). The authors called it "some of the first direct evidence of the route followed by the Punic Army", and the press called it solid evidence. I read all of the open papers, the 2020, 2024 and 2026 restatements and the critics (Kuhle and Kuhle 2012 and 2015, Hunt 2016, de Galbert 2016, MacDonald 2016), and recalibrated every published date against IntCal20 with my own code. The part that bears on the pass follows.

The dates and the bed are genuine, but they do not date the bed to 218 BC. In the original core the authors have two dates from the 40 cm level, 2070 ± 31 and 2530 ± 90 BP, which they say "bracket the invasion year". Each on its own excludes 218 BC at two sigma, in their own IntCal13 table (176 BC to AD 1 and 820 to 408 BC) and in IntCal20 (169 BC to AD 8 and 814 to 407 BC), and they are not two measurements of one age: the Ward and Wilson test gives T = 23.4 against a critical value of 3.84 (p = 1.3 × 10⁻⁶). Two dates from one depth that disagree by this much mean that the peat at that depth is a mixture of ages (which is what "churned" means), and a mixed bed cannot be dated to a year by any number of dates. The age model does not change this.

Table 1 of part I could be read more harshly. It prints, between the two 40 cm dates, a row "Hannibal: 218 BC, 2168 cal BP, ± 0", an event of known date and zero uncertainty with no agreement index beside it, which looks like the invasion year fed into the model as an input that the model then returned. I do not think the charge holds. A reference date printed in a results table does not show that the date was imposed on the model, and the paper's own plot of the model (Fig. 4 of part I) points the other way. OxCal's multiple plot draws a row for each element of a model as run, unless the user removes it; this plot shows eight radiocarbon dates in three groups ("Pre-invasion Beds", "MAD bed; Hannibal invasion", "Post-invasion bed") between four boundaries, inside a sequence named "Hannibal: Alps G5 and G5A sites", and no row for a 218 BC element. The 218 BC appears as a dashed vertical line and a dashed box drawn round the two 40 cm dates, with a label in another typeface. The simplest reading is that the table row is a reference entry and the line an annotation. I cannot confirm that reading either way. The OxCal input is not printed in part I, in part II or in the four later restatements (Mediterranean Archaeology and Archaeometry 2017 and 2020, Scientific Culture 2024, the 2026 paper), and no supplementary file is deposited with the journal, with either author's university repository or on Zenodo (I checked each). Without the input, neither the charge of circularity nor its denial can be verified.

What the paper prints is enough to decide the narrower question, because nothing in its own model output places the bed in 218 BC. The modelled ranges in Table 1 for the two 40 cm dates are 2464 to 1597 and 3002 to 1804 cal BP at two sigma. The modelled range for the boundary between the churned bed and the beds above it, the table's upper "Transition", is 2356 to 997 cal BP, which is 407 BC to AD 953. "Firmly places" and the caption's "high probability of >95%" rest on the agreement index, A_model = 97.6, which measures how well the measured dates fit the model's ordering; it is not a probability that any bed dates to any year. Part II and the magnetic-susceptibility paper of 2017 carry the same reading over: "linked to radiocarbon dates precisely at 2168 cal yr BP" and "2168 cal yr BP (i.e., 218 BC with a 95% confidence interval)". The table has a further fault. Its modelled columns cannot be the posteriors of the dates in the model plotted: for UBA-23647 the unmodelled range is 3895 to 3720 cal BP and the modelled range 3509 to 2109, with no overlap, beside an agreement index of 98.3, and an index near 100 requires the posterior to sit on the likelihood. I cannot tell what those columns report. The plot's row labels carry the calibrated means (for example "2040 ± 45" for UBA-24875, whose radiocarbon age is 2070 ± 31) rather than the radiocarbon ages, so plot and table cannot be matched row by row without the input either. The dating evidence does not independently establish a disturbance specifically in 218 BC, much less identify the animals responsible as Hannibal's.

The later sections do not repair this. Section G5B has no date within 500 years of 218 BC and its "218 BC bed" is interpolated between a seventh-century AD date and a ninth-century BC one. G5D's "218 BC bed" is at the level of a sample whose median is 400 BC. G5C has the one measurement in the whole series whose range includes 218 BC, UBA-30330 at 51 cm, 2190 ± 35 BP: it puts 17 per cent of its probability in 230 to 200 BC and 42 per cent in 355 to 282 BC. Nor could any measurement do better. At this point the calibration curve is nearly flat between 350 and 250 BC and falls steeply afterwards, so a sample formed in exactly 218 BC and measured to ± 30 years would calibrate to 380 to 197 BC; a sample of 207 BC, when Hasdrubal took the same road, or of 231 BC, when a Gallic host crossed, would be indistinguishable from it. The claim names a single year; the method, for samples of this age, cannot resolve better than about two centuries. In 2026 Mahaney reread the 2070 BP date as peat that grew back "nearly a century" after the event, with every older date reworked; on that reading there is no date from the event at all, only a Bronze Age floor and a second-century BC ceiling. The dates allow a disturbance in the late third century BC but cannot distinguish it from one at any other time between about 380 and 150 BC.

The other parts of the argument are weaker still. The lipid evidence shows herbivore dung at every depth from 15 to 60 cm, with a peak in the churned bed that is no larger than the peak at 15 to 25 cm which the authors themselves assign to medieval transhumance; by the authors' own criterion, "the presence of deoxycholic acid in the absence of lithocholic acid indicates a ruminant animal source", the bile-acid pattern they report points to cattle, sheep, goats or ibex rather than horses. The Clostridia are a class of bacteria that dominate the gut of every large herbivore and also live in waterlogged peat; the sequencing was done at class level, in one section, with no ancient-DNA authentication, and the genome programme announced in 2016 has not been published. The "horse tapeworm egg" of the 2016 press coverage has become "roundworm eggs" in the 2020 paper and has never appeared with a taxon, an image or a depth in any peer-reviewed text. The mire is a wet meadow of about sixty by forty metres on the French side, below the col: it cannot be the camp where Polybius's army spent two days at the summit, and Mahaney concedes that "the Col de la Traversette itself could barely hold 100 soldiers". The rockfall exists and is older than 218 BC, but its two lobes are dated by lichen cover and soil development, methods whose resolution is a thousand years or more, so they cannot show that the younger lobe fell "recently" before the crossing, and cannot show it did not; and Polybius's words describe a ledge-path broken away along a precipice, which Hannibal "built up", not a debris sheet lying across a slope. No artefact of any kind has been recovered from the mire, the rockfall or the col. The one physical thing that the fieldwork has established is that large herbivores have repeatedly churned and dunged a watering place below the Traversette over the last three thousand years, which is what one would expect of any high pasture with a spring.

Nothing published since 2017 changes this. A further field season in 2017 produced no reported artefacts; Mahaney's 2019 Archaeometry paper (online 2018) moved to sections in the upper Po valley with a similar churned bed at "about 2,200 years"; the 2020, 2024 and 2026 papers are restatements by the lead author; no independent group has re-sampled or re-dated the mire; and the only new item, Berti and Vollrath's 2026 energetics model that finds the Traversette the cheapest route in joules, is a length-and-gradient calculation that does not engage with the mire, the descent or the rockfall. The dating arithmetic above has not, as far as I can find, been published by anyone; the objections to the camp, the species and the uniqueness of the rockfall were made by Hunt, Kuhle and Kuhle and de Galbert in 2012 to 2016.

In the evidence model the Traversette deposit therefore enters as an item whose likelihood is 1.0 for the Traversette and 0.75 to 1.0 for every other candidate, because the deposit is what a crossing would leave and equally what three thousand years of pasturing would leave. A scenario in which Mahaney's reading is credited in full is run for comparison.

The argument in print begins with Josias Simler (1574). A count made in 1835 and reported by Bouché-Leclercq in 1874 found ninety published opinions: thirty-three for the Petit St Bernard, twenty-four for the Montgenèvre, nineteen for the Grand St Bernard, eleven for the Mont-Cenis and three for the Monte Viso, which is the Traversette (Azan 1902, p. 46). Nobody had proposed the Clapier until Colonel Perrin published the view of the Po plain from it in 1883. Peddie (1997) counted twenty-four modern writers: six for the Mont-Cenis, six for the Montgenèvre, five for the Clapier, three for the Traversette and, by subtraction, four for the Petit St Bernard. The Clapier has been the usual answer of ancient historians writing in English, always hedged: Walbank "hesitatingly" (1956), Lazenby (1978), Lancel (1995), Hunt (2017), Hornblower "probably" (2024). The Montgenèvre is the main scholarly alternative (Connolly 1981; Hoyos 2006 and 2015). The Traversette is the choice of de Beer's school (1955 to 1974), of Prevas (1998), of the earth-science papers (Mahaney and colleagues, 2007 to 2026) and of an energetics model in PNAS (Berti and Vollrath 2026). The Petit St Bernard survives as a French regional position (Bocquet; de Coninck 1992). Goldsworthy (2000), MacDonald (2015) and Lendering (2022) decline to choose.

The table gives each candidate, its principal advocates and its share in the headline run of the evidence model (the nine Polybian items at half strength). The cards below give the case each side has made, in the advocates' words where those could be read, and then (under Here) what this page found on each point. The full list of fifty-two works, with the Island each author chose and how much of each could be read, is in dossiers/scholarship.md.

What could be read. The public-domain studies from Whitaker (1794) to Torr (1924) were read from scanned copies, and Connolly, Goldsworthy, Hoyos (2015) and Hornblower from scanned copies of the books. De Beer, Walbank, Hoyte, Proctor, Lazenby, Seibert, Lancel, Prevas, Hunt and Mahaney's Hannibal's Odyssey could be obtained only as search snippets, catalogue records or abstracts, so their positions are given here as their reviewers, their citers and the snippets report them, with the page numbers those sources supply. Perrin's two books, Hennebert, Hyde, Bocquet, de Galbert, Lendering, Freeman (2022) and Kötter (2024) were not obtained at all. The access log lists every attempt.

The table below weighs eight candidates against fourteen items of evidence. Six are the cols that have had serious advocates. The seventh is the Grand St Bernard, which had the Roman tradition on its side. The eighth, "another col", stands for the passes nobody has argued for at length, the Malaure, the Agnel, the Croix and the Autaret among them. The Mont-Cenis row covers both cols of its plateau, the Grand and the Petit Mont-Cenis. Each candidate starts with a prior, the share it holds before any evidence is weighed: an even 14 per cent, the Larche and the Petit St Bernard slightly less and the Grand St Bernard 6 per cent, because Polybius's own list excludes the Aosta passes. The prior hardly matters. With a flat prior, the same share for every candidate, the headline shares change by less than one point.

The fourteen items fall into five families. The nine Polybian items are the narrative: the Island, the ten days along the river, the Allobrogian ambush, the nine days to the summit, the view, the old snow, the broken descent and the three days to the plain, the arrival among the Taurini, and the 1,200 stades. Livy's Druentia itinerary is one item. The testimonia are two: Varro's order with Pompey's letter, and the Roman tradition of the Graian and Pennine passes. The Traversette mire is one. Practicability is one. For each item and each candidate the table gives a range, my estimate of how likely that evidence would be if this col were the pass. The best-fitting candidate sits near 1; a candidate the item nearly excludes sits near 0.02. The ranges are argued in the sections above. A reader who disagrees with a cell can change it in code/evidence_table.py and re-run the model.

The items are not independent. Everything in the Polybian family comes from one author working from one set of informants, and Livy's account of the same events comes down the same tradition, so he adds no second witness. Multiply the nine Polybian likelihoods together as if they were nine witnesses and the Clapier goes to 70 per cent. Collapse each family to one item instead, by taking the geometric mean within it, and the view, the snow, the descent and the Island count as one fact because one man reported them, though they are different facts. That treatment gives a near-flat result in which the Larche leads at 18 per cent, because it is an easy pass and Livy's Durance suits it. The headline run takes a middle course. Every Polybian item enters at half strength, as if the nine were four or five independent facts, and everything else enters at full strength. That run gives the Clapier 40 per cent, the Traversette 29, the Mont-Cenis 16, another col 9, and the rest 6 between them. Those shares are conditional on the fourteen items and on the ranges set for them. When every cell is sampled inside its range 20,000 times, the Clapier comes first in 65 draws out of 100 and the Traversette in 29.

Dropping items one at a time shows what carries the ranking. Remove the practicability judgement and the Traversette leads, 45 to 32; remove the Allobrogian ambush and it leads 40 to 31; remove the Island, or the ten days along the river, and the two are level at about 35 each; remove Livy's Druentia and the Clapier rises to 49. No single Polybian item decides it. The view and the snow are the two items that most favour the Traversette, and each moves it by only a few points, because the Clapier has a partial claim to both.

The scenarios in the table below put the sceptics' cases in numbers. If de Beer's Island at the Aygues is right, the Traversette leads at 54 per cent. If Livy's itinerary by the Tricastini and the Durance is trusted in full, it leads at 46. If Mahaney's deposit is credited as the authors read it, it leads at 55; if the Traversette is granted to be as practicable as any other col, at 51. If the view is a commonplace the Clapier still leads, but the Mont-Cenis rises to 24 per cent; if Polybius's old snow is a generalisation, the same happens. If his numbers are distrusted altogether, nothing much changes, because the numbers are the least discriminating part of the case.

The table can be read in two ways. One is that the Clapier is the best single answer on the evidence as it stands, weighed as the table weighs it, and that its odds over the Traversette are about seven to five, a long way from ten to one. The other is that the question is which valley, the Isère or the Durance; that the choice turns on a conjectured river name and on how far the word "Allobroges" can be stretched; and that once the valley is chosen the terrain picks the col with little doubt, the Clapier for the Isère and the Traversette for the Durance. I hold both readings. That is why the verdict at the top of this page calls the Clapier the leading candidate and says that several passes remain possible.

The ranking would move if a witness to Polybius 3.49.6 or Livy 21.31.4 gave a river name other than a form of Isara, because the Island would move with it. The known manuscripts have been collated for two centuries and none does, but the Vatican Polybius has not been re-examined for this page and Cluverius's conjecture has never been tested against a fresh collation.

A find would move it further. A datable Punic or Iberian object, a coin of Carthage, a Numidian bit, an elephant tooth, from any col or its approach would outweigh every item in the table. None exists for any candidate. The Traversette fieldwork of 2011 to 2017 found none, and no comparable excavation has been made at the Clapier, the Savine basin or the Clarea headwall.

Ancient DNA from the mire could settle whether the churned bed is an army's. Host-specific sedimentary DNA from the churned bed, of the kind now routine in Alpine lake cores, that showed Equus in quantity and no ruminant signal at the right depth would make the Traversette deposit an army's, though still not one army's. A ruminant signal at that depth would point to livestock rather than an army.

A snow record for the third century BC would move the two leaders against each other. A reconstruction of the third-century BC snowline for the Cottian Alps 300 m below the modern mean would put old snow on the Clapier in late October and remove the Traversette's advantage on the snow item; one 200 m above would remove the Clapier's partial claim to the snow altogether.

A finer terrain model could change the Clapier's view. The view test was run on a 90 m surface. A 5 m model of the Clapier ridge and the Clarea headwall could open or close the Clapier's sliver of plain; it cannot close the Traversette's 63-degree sector.

Reading Walbank and de Beer in full could change the weights. The principal modern commentary and the principal modern case for the Traversette were read only in part (see below). An argument in either that this page has not met would change the weights, though probably not the ranking.

Most of what this page says has been said before. That the Isère reading of the Island, the Allobrogian ambush at the mouth of the Maurienne and the Clapier's summit plateau fit Polybius best was argued by Perrin (1883), Azan (1902) and Colin (1904), and was the conclusion, always hedged, of Walbank (1956 and 1957), Proctor (1971), Lazenby (1978), Lancel (1995), Hunt (2017) and Hornblower (2024), and of Kuhle and Kuhle's reply to Mahaney (2012). That the Traversette fits the view, the snow and the steep descent was argued by Torr (1924), de Beer (1955) and Prevas (1998). That the Grand and Petit St Bernard fail on the Taurini is Livy's own argument. That the mire's radiocarbon window admits several armies was said by de Galbert (2016), and that the Traversette has no summit camp by Hunt (2016). This page adds numbers to those arguments and shows which of them carry the result.

Five things are new here, as far as I can find, and all five are new analysis of evidence already in print. No manuscript, artefact, date or measurement is added to the record.

First, a computed viewshed for every candidate col, for its first 10 km of descent and for the ground within a walk of it, against a single definition of "the plains of the Po". Earlier claims for the view rest on eyewitness reports from particular days. The result is that only the Traversette and, marginally, the Clapier see any plain from the col or the descent; the Mont-Cenis has a half-percent sliver from three cells on the ridge above it, and the other cols see none.

Second, a recalibration of all fourteen published Traversette dates against IntCal20 with a statistical test of the "bracketing" pair; the finding that nothing in the published age model's own output places the bed in 218 BC, and that its "firmly" and ">95%" rest on an agreement index read as a probability; the finding that the model table's modelled columns are inconsistent with its own agreement indices; and the observation that at this point of the calibration curve no radiocarbon measurement could separate 218 from 207 BC, or from any year between about 380 and 150 BC. I found no published statement of this arithmetic.

Third, the textual point that the name Isère stands in no manuscript of either author and is Cluverius's conjecture of 1624 in both, the Livian manuscripts pointing to Sarar or Arar and the Polybian to Skaras; and the observation that the Livian "1,000 feet" of the landslip is a Renaissance restoration of a corrupt line, not a manuscript figure.

Fourth, the demonstration that Polybius's two totals, 1,400 stades to the ascent and 1,200 for the crossing, cannot both be satisfied by any single line over the real terrain, so that neither total can be used against a candidate; and the observation that 1,200 stades equals fifteen days at the 80 stades a day of the preceding stage.

Fifth, the evidence model itself, with its shared-source correction, its leave-one-out test and its named sceptical scenarios, which trace the disagreement between the Clapier and the Traversette to three cells of the table, the practicability judgement, the Allobrogian ambush and the Island.

I have not been able to check the second and fourth points against Walbank's commentary or against the recent Italian literature, which were not accessible; a specialist may know a prior statement of either.

Several books were not read in full: Walbank's Historical Commentary on Polybius I (1957) and his 1956 article in the Journal of Roman Studies; de Beer's Alps and Elephants (1955) and Hannibal's March (1967); Prevas, Hannibal Crosses the Alps (1998); Hunt, Hannibal (2017); Lancel, Hannibal (1995); Hoyte, Trunk Road for Hannibal (1960); Seibert, Hannibal (1993); Proctor, Hannibal's March in History (1971); Mahaney, Hannibal's Odyssey (2008); Kuhle and Kuhle's two Archaeometry papers; Mahaney's Wiley papers of 2010, 2013 and 2019. These were obtained only as abstracts, search-inside snippets or library catalogue records, and their arguments are known here through the open literature and one another. Where the page attributes a view to one of them, it is from those secondary reports.

The editions are second-hand at two points. Livy's Latin was read in Foster's Loeb text as transcribed on Wikisource, with the Oxford apparatus from scanned pages; the Puteanus is lost for this book and the manuscript readings are taken from the Oxford editors. The Polybian apparatus for 3.49.6 is known only through Schweighäuser and the Oxford editors, since the Teubner editions print none at that point.

The terrain model is coarse, at 90 m. It can say that ground of the kind Polybius describes exists on a descent; it cannot say that a particular ledge or its landslip existed in 218 BC, and its col heights are up to 35 m low. Route lengths are valley-floor lengths and are lower bounds by perhaps 5 to 15 per cent for the gorges the ancient track would have climbed around.

The snow bands are judgements. The bands for old snow at each col come from the modern snowline and the glacier record rather than from measurement; no autumn snow-patch series for the southern French Alps was found, and no temperature reconstruction fine enough for one decade of the third century BC exists.

Nothing here has been checked on the ground. The Clapier's "white rock", the Clarea headwall, the Savine plateau and the Combe du Queyras are known to this page only from the model and from others' descriptions.

The model is a set of judgements, one for every likelihood range. The shared-source correction is a rule of thumb rather than a derivation; the two extreme treatments bracket it and are shown.

Primary texts, as used.

• Polybius, Historiae 3.33-60 and 34.10, ed. Th. Büttner-Wobst (Teubner, 1882 and 1905), Greek text from the Perseus Digital Library XML; readings at 3.49.6 from J. Schweighäuser, Polybii Historiae V (1789) 594-595 and from the Oxford editors of Livy; English of W. R. Paton (Loeb, 1922) and E. S. Shuckburgh (1889) for comparison. Saved: texts/polybius_3_greek.txt, texts/polybius_34_10_greek.txt.
• Livy, Ab urbe condita 21.22-39, text of B. O. Foster (Loeb, 1929) as transcribed on la.wikisource, collated at the cruces with C. F. Walters and R. S. Conway, Titi Livi Ab urbe condita III (Oxford, 1929) and with Weissenborn-Müller and Riemann-Benoist from scanned pages; Livy 27.39 from thelatinlibrary. Saved: texts/livy_21_latin.txt, texts/editions_ocr/.
• Strabo 4.6.5-12, ed. A. Meineke (Teubner, 1877), with H. L. Jones (Loeb, 1923). Servius on Aeneid 10.13, ed. G. Thilo (1884). Sallust, Historiae 2.98 Maurenbrecher, from J. C. Rolfe (Loeb, 1921). Cornelius Nepos, Hannibal 3, edd. Fleckeisen and Halm (Perseus). Pliny, Naturalis historia 3.31-37 and 3.117-138, ed. Mayhoff. Ammianus Marcellinus 15.10, J. C. Rolfe (Loeb, 1935). Appian, Hannibalica 1-4, ed. L. Mendelssohn. Zonaras 8.23 in E. Cary's Dio (Loeb, 1914). Columella, De re rustica 11.2, thelatinlibrary. All saved under texts/testimonia/ and texts/terrain_sources/.

The Traversette papers and their critics.

• Mahaney, W. C., and 23 co-authors, "Biostratigraphic evidence relating to the age-old question of Hannibal's invasion of Italy, I: History and geological reconstruction", and "II: Chemical biomarkers and microbial signatures", Archaeometry 59 (2017) 164-178 and 179-190. Read in full (author manuscripts, Queen's University Belfast repository).
• Mahaney, W. C., and others, "Polybius' 'previous landslide': proof that Hannibal's invasion route crossed the Col de la Traversette", Mediterranean Archaeology and Archaeometry 14.2 (2014) 77-94; "Notes on magnetic susceptibility in the Guil valley alluvial mire", ibid. 17.1 (2017) 23-35. Read in full.
• Mahaney, W. C., "The Hannibal enigma of 218 BC", Scientific Culture 6.3 (2020) 7-24; "Polybius: decoder of Hannibal's alpine invasion route", ibid. 10.1 (2024) 59-67; "Futile route quest of the Hannibal invasion", London Journal of Research in Humanities and Social Sciences (2026). Read in full.
• Mahaney, W. C., and others, Journal of Quaternary Science 25 (2010); Archaeometry 55 (2013) and 61 (2019); Kuhle, M. and Kuhle, S., Archaeometry 54 (2012) and 57 (2015). Abstracts only.
• Hunt, P., in EARTH magazine (2012, 2016); Allen, C. C. R., The Conversation (April 2016); de Galbert, G., "Hannibal, les Alpes et la Traversette" (2016); Smithsonian and Queen's University Belfast press material (2016); Berti, E. and Vollrath, F., PNAS (2026). Read in full.
• Reimer, P. J., and others, "The IntCal20 Northern Hemisphere radiocarbon age calibration curve (0-55 cal kBP)", Radiocarbon 62 (2020) 725-757; curve file data/intcal20.14c. Ward, G. K. and Wilson, S. R., "Procedures for comparing and combining radiocarbon age determinations", Archaeometry 20 (1978) 19-31.

Terrain, climate and astronomy.

• Copernicus GLO-90 digital elevation model, tiles N43-N46 E004-E008, from the AWS open-data bucket. OpenStreetMap place and river data through Nominatim.
• Rabatel, A., Letréguilly, A., Dedieu, J.-P. and Eckert, N., "Changes in glacier equilibrium-line altitude in the western Alps from 1984 to 2010", The Cryosphere 7 (2013) 1455-1471. Klein, G., and others, "Shorter snow cover duration since 1970 in the Swiss Alps", Climatic Change 139 (2016) 637-649. Bender, E., Lehning, M. and Fiddes, J., Frontiers in Earth Science 8 (2020) 100. ARPA Piemonte, technical note on snow cover in the Piedmont Alps. Météo-France, "Le climat des Hautes-Alpes".
• Holzhauser, H., Magny, M. and Zumbühl, H. J., "Glacier and lake-level variations in west-central Europe over the last 3500 years", The Holocene 15 (2005) 789-801. Le Roy, M., Ivy-Ochs, S., and others, "Holocene glacier variations in the Alps", in European Glacial Landscapes: The Holocene (Elsevier, 2024) 367-418.
• Skyfield 1.55 with the JPL DE406 ephemeris; PyEphem 4.2.1 as an independent check.

Modern scholarship on the pass. The works consulted, with how much of each could be read, are listed in the accompanying notes (dossiers/scholarship.md); the principal ones are named in the candidates section above. The public-domain studies from Whitaker (1794) to Torr (1924) were read from scanned copies. The books of de Beer, Walbank, Hoyte, Proctor, Lazenby, Seibert, Lancel, Prevas, Hunt and Mahaney are in copyright and were obtained only as abstracts, catalogue records or search snippets.

The code, data and texts behind this page are kept together in one folder. dossiers/ holds the working notes of the four strands (the sources, the terrain, the geoarchaeology and the scholarship), texts/ the editions and papers, data/ the outputs, and code/ the scripts, which run in this order. The network is needed only for the downloads.

To change a judgement, edit the range in code/evidence_table.py and run the last two lines again. To test another col, add its waypoints to routes.py and its saddle to viewshed.py.