Roman Britain and the wider empire · AD 108 to about 165; the legion's last attested decade is the 120s
Where, when and how did the Ninth Legion, last recorded building a fortress gate at York in 107/108, cease to exist?
What happened. Some time between December 107 and December 108 the Ninth Legion (in Latin, Legio IX Hispana), a unit of about five thousand Roman soldiers, finished a gate of its fortress at York in northern England. It had been in Britain since the Roman invasion of AD 43; it had been mauled by the rebels of the British queen Boudica in 60 and refilled with two thousand men from Germany, and it had been attacked by night in its camp in Caledonia, now Scotland, in the 80s. Part of the dedication slab for the new gate, carrying the titles of the emperor Trajan, survives (it is RIB 665, its number in Roman Inscriptions of Britain, the corpus of every Roman inscription found in Britain), and it is the last dated text that names the legion anywhere. Soon after 165 the army's legions were listed on a column in Rome, thirty-three of them (ILS 2288, its number in Hermann Dessau's selection of the most important Latin inscriptions); the Ninth is not among them, and no ancient writer records where or how it ended in the half-century between.
Why it is a mystery. The evidence is indirect (inscriptions of officers who served in the legion and went on to dated posts, roof tiles stamped with the legion's name, and a few sentences in later writers about wars in which Roman soldiers died), and historians have read it in four ways. The oldest view, going back to the German historian Theodor Mommsen in 1885, is that the legion was destroyed or disbanded in Britain, and Emil Ritterling, whose 1925 encyclopaedia article on the legions is still the standard account, put that end after 120. In its modern form, argued by Nick Hodgson, an archaeologist of Hadrian's Wall (the frontier barrier built across northern Britain from 122), in Britannia, the journal of Romano-British studies (volume 52, 2021), and explored by the Dutch archaeologist Erik Graafstal in the same journal (volume 56, 2025), the loss falls in fighting that followed the visit to Britain in 122 of Hadrian, emperor from 117 to 138. The second view is that the legion left Britain for the Rhine, on the strength of three stamped tiles found at Nijmegen, a Roman fortress on the Rhine in the Netherlands, and was lost in the Bar Kokhba war, the Jewish revolt of 132 to 136 in Judaea; the third, that it left Britain and was lost at Elegeia in Armenia in 161, where Cassius Dio, a historian writing in the early third century, records (Dio 71.2) a whole Roman legion destroyed by the Parthians, Rome's eastern rivals, a view proposed by Werner Eck, a German historian of Roman inscriptions, in 1972, and one towards which Duncan Campbell, a historian of the Roman army, leans; and the fourth, that the legion was broken up or merged into another legion, so that no record of an end would exist.
How it was judged. Every inscription naming the legion was collected: all 133 records from Britain, and every entry for the rest of the empire in the Heidelberg and Clauss-Slaby databases, the two large online databases of Latin inscriptions. The careers of the three senators who served in the legion and whose later posts are dated (T. Aninius Sextius Florentinus as its commander, and L. Aemilius Karus and L. Novius Crispinus as its senatorial tribune, the young senator who was second-in-command) were worked back to the latest year the legion must have existed, and that arithmetic was run as a simulation; the Nijmegen finds were counted against the fortress's other finds; and five pieces of evidence (the officers' careers; the absence of any base or trace of the legion outside Britain after about 122; the British crisis of about 122 to 125; Dio's record of a legion destroyed at Elegeia in 161; and the absence of any building stone of the Ninth on Hadrian's Wall) were weighed against the four endings, each weight given as a range from a base reading to a sceptical one, the sceptical end being what a determined critic of the British reading could reasonably assume. The result is that the legion certainly existed in 122 and almost certainly in 124, that nothing places it outside Britain after that, and that Britain in about 122 to 130 is the best-supported ending, at about 50 to 65% on central readings of those ranges. That probability is a judgement, stated so that it can be varied and tested, and the answer is probable but not proven; any inscription naming the legion, or a serving soldier of it, dated after about 130 would move the answer east.
The legion most probably ended in Britain, destroyed or defeated and then disbanded, between about 122 and 130, in the years after Hadrian's visit of 122. On central readings of the evidence the British ending has a probability of about 50 to 65%.
The hypothesis was argued by Hodgson (2021) and explored as a possibility by Graafstal (2025), and the analysis supports it by weighing the evidence they used, without testing it independently and without new evidence. It is not the agreed view of the field: Campbell, A.R. Birley, the historian of Roman Britain's governors, and Lawrence Keppie, the historian of the legions, hold other positions, and no published response to Hodgson exists. The probability of a British ending runs from 85% at the base reading to 36% at the sceptical one, and every figure is conditional on the four endings and the likelihood ranges chosen. The strongest opposing case, Campbell's book of 2018, could not be read.
The legion is last dated as a unit at York, 10 December 107 to 9 December 108, building the fortress gate (RIB 665). The careers of its last officers require it to exist until about 123/124: Florentinus, whose tomb inscription is CIL III 87 (CIL is the Corpus Inscriptionum Latinarum, the Berlin corpus of all Latin inscriptions), was governor of Arabia, the province annexed in 106, roughly modern Jordan, by 2 December 127, after governing Narbonensis, Roman Provence, in 123/4 or 124/5; and Karus (ILS 1077) was consul, one of the two senior magistrates of the year, in office on 19 March 144. Crispinus (ILS 1070), whose tribunate falls about 124 to 128, probably carries it to about 125 to 128, the end of his service. Those careers give the latest date the legion is required to have existed, and they cannot give the date it ended: the officers alone require the legion in 124 at 88%, in 126 at 51% and in 130 at 6%, and a legion still in being in 128 leaves no more trace in these careers than one that ended in 135. The list of legions carved in Rome after 165 (ILS 2288) omits it. The Nijmegen finds come from a detachment of the legion: three tile stamps from one die (the carved stamp pressed into the wet clay) against 196 stamps of the vexillatio Britannica, a task force drawn from the British legions that held the fortress, in a fortress of 16.5 hectares whose structural occupation ended about 125 to 130, and no legion-sized base was free on the Lower Rhine after 122. In the same years Britain suffered a war with heavy losses, recorded by Fronto, tutor to the emperor Marcus Aurelius, in a letter of 162/3; Hadrian answered it with 3,000 legionaries drawn from three legions in Spain and Upper Germany, the province on the middle Rhine, and with the Sixth Legion Victrix (VI Victrix), which took over York as a one-for-one replacement. None of those sources names the Ninth; together they show a British crisis in the same years in which the officers' careers cease to attest the legion, and a Roman response of the kind that the loss of a legion would call for.
Every percentage is conditional on two choices: that Britain, the Bar Kokhba war of 132 to 136, Elegeia in 161 and a residual ending elsewhere or without record are the only four, and the range of weight given to each of five pieces of evidence, from a base reading to a sceptical one (the weight, or likelihood, of a piece of evidence is how probable that evidence would be if the ending were true, and the prior is the probability given to each ending before the evidence is considered). With every value at the base reading Britain is 85%; at the sceptical reading, 36%; at the sceptical reading with the prior for Britain lowered from its base value of 30% to 15%, 19%. Over 100,000 random draws across the ranges the median, the middle value, is 52% when every value within a range is treated as equally likely and 65% when the base reading is treated as the most likely value, and Britain ranks first in 97% of the equal-weighted draws and in all of the base-weighted ones. Dropping any one piece of evidence from that 52% moves Britain to 29% (without the careers, or without the absence of a base for the legion elsewhere), 43% (without the British crisis) or 73% (without the absence of any building stone of the Ninth on Hadrian's Wall, which counts against a British ending). The percentages are judgements, printed with their ranges so that a reader with different judgements can see how far the answer moves; the ranges come from those judged inputs, and more runs of the simulation would not narrow them.
The strongest sustained case for an eastern ending is Campbell's book The Fate of the Ninth (2018), and it could not be read: it has no preview in any bookseller's or library database, no scanned copy exists, and no journal has reviewed it. His case is reconstructed from what could be read in full: his 2010 article in the magazine Ancient Warfare, all 57 posts of his blog and all 25 of his posts on RomanArmyTalk, an online forum on the Roman army, to February 2026. That case can be answered at three points. The absence of evidence for a British war of 119, the war in which the traditional account, dating the loss to 117 to 119, had the legion destroyed, is accepted, since the interval given here is about 122 to 130; the one Nijmegen tile found in a datable layer, a rubble layer of the fortress's third building phase, shows men of the Ninth there, and a tile in such a layer dates the moment the layer formed and says nothing about how long the legion had been there, so the quantities carry the argument for a detachment; his "unknown unknowns" objection, that an eastern base may exist unrecorded, is what the sceptical end of the weighting allows for, and accepting it in full brings Britain to 29% with Bar Kokhba ahead of it. Any argument that exists only in the book is unknown here and is not answered. Hodgson's 2021 paper, which argued the British reading, is behind the publisher's paywall and was read in its abstract and through Graafstal's citations and quotations.
The shares come from a weighting of five pieces of evidence, each given a likelihood from a base reading to a sceptical one under each of the four endings, with the result computed over 100,000 random draws across those ranges; the bar shows the middle of each range. What moves them is the reading of the evidence: Britain is 85% at the base reading, 36% at the sceptical one and 19% at the sceptical reading with the prior for Britain lowered to 15%, and it ranks first in 97 to 100% of the draws. Every figure is conditional on the four endings and the ranges chosen.
Every inscription naming the legion was collected: all 133 records from Britain in Roman Inscriptions of Britain, the corpus of every Roman inscription found in Britain, read in full, and every entry for the rest of the empire in the Heidelberg and Clauss-Slaby databases, the two large online databases of Latin inscriptions, 245 rows in all. The literary sources were read in Latin and Greek, and the scholarship from the German historian Theodor Mommsen in 1885 to Erik Graafstal, the Dutch archaeologist whose article appeared in Britannia, the journal of Romano-British studies, in 2025.
Two modern arguments could not be read directly. Duncan Campbell, a historian of the Roman army, wrote a book, The Fate of the Ninth (2018), of which no preview, scan, review copy or journal review exists; his case is stated from his 2010 article and from all 57 of his blog posts and 25 forum posts, so arguments that exist only in the book are not answered. Nick Hodgson's 2021 article in Britannia, which argues for a British ending, is behind a paywall and was read in its abstract and through the citations and quotations in Graafstal's article.
The careers of the three senators who served in the legion and whose later posts are dated (Florentinus, its commander, and Karus and Crispinus, its senatorial tribunes, the young senators who were second-in-command) were worked back to the latest year the legion must have existed, and that arithmetic was run as a simulation 200,000 times. The legion's three stamped tiles from Nijmegen, a Roman fortress on the Rhine, were counted against the fortress's other finds. Five pieces of evidence were then weighed against the four proposed endings, each weight a range from a base reading to a sceptical one, with the result computed over 100,000 random draws across those ranges.
Figure 1 lays out the dated evidence for the legion: the last dated text, the York gate of 108, the fixed points such as the Sixth Legion's arrival in Britain, and for each officer a band spanning the middle 90% of the simulated values for his service in the legion, with the red line the median, the middle value; the shaded columns mark three of the proposed endings, Britain about 122 to 130, the Bar Kokhba war and Elegeia in 161. Figure 2 turns the same simulation into a curve: the probability that the officers' careers alone require the legion to exist in a given year, a floor on its existence rather than an estimate of when it ended.
The legion certainly existed in 122 and almost certainly in 124, nothing places it outside Britain after that, and Britain in about 122 to 130, in the years after the visit to Britain in 122 of Hadrian, emperor from 117 to 138, is the best-supported ending: about 50 to 65% on central readings, 85 at the base reading and 36 at the sceptical one. Every percentage is conditional on the four endings and the ranges assigned, and the probability is a judgement, stated so that it can be varied and tested. Any inscription naming the legion dated after about 130 would move the answer east.
Already established by others. Positions already argued by others that this analysis supports: the impossibility of 117 to 119 (Ritterling 1925; A.R. Birley 2005); the detachment reading of Nijmegen (Haalebos 2000, by Jan Kees Haalebos, one of the fortress's excavators; van der Veen 2025, a study of the fortress by P.J.M. van der Veen); the reading of the Numisius stone (CIL XI 5670, the inscription of a tribune of the Ninth named Camurius Numisius Iunior) that makes the consul of 161 of the same name his son (Keppie 1989); the absence of any eastern evidence (Keppie; Menahem Mor, the Israeli historian of the Bar Kokhba revolt); and the British setting in the early 120s, argued by Hodgson 2021 and explored by Graafstal 2025. An analysis that reaches the same conclusion as its proponents by weighing the same evidence has not tested that conclusion independently: the agreement supports the reading of Hodgson and Graafstal without settling the question. What this analysis adds is the synthesis and the explicit statement of how much the answer depends on judgement.
What historians have said. The popular account has the legion destroyed in northern Britain about 117 to 119, a date the officers' careers rule out. Mommsen (1885) had it annihilated at York soon after 108; Ritterling (1925) saw from the careers of Karus and Crispinus that the end must fall after 120 and thought of a second British revolt, about 125, later than the war of 117 to 119 to which the traditional account had tied the loss. Eric Birley (1948, 1971), the Durham historian of the Roman army, built the eastern alternative on Cocceius Severus, a chief centurion of the Ninth who rose to camp prefect of the Tenth Legion Gemina and whose family name Birley read as a sign of a late date, on the undated tiles of the Ninth from Carlisle, near the western end of the Wall, and on the Nijmegen finds; Bogaers (1967), one of the excavators of Nijmegen, placed the whole legion there in the 120s; Eck (1972) took it to Elegeia in 161 on the Numisius diploma, a military diploma (the bronze discharge certificate given to soldiers on completing their service) that names a Q. Numisius Iunior as consul in 161, whom he identified with a tribune of the Ninth of that name. Keppie (1989) dismantled the eastern case item by item, and Haalebos (2000), the other excavator of the fortress, and the later Nijmegen excavators reduced the continental presence to a detachment of 104 to 120. Campbell (2010, 2018) holds that the legion left Britain by the 120s and was lost in the East, and calls a British loss "a very weak theory"; A.R. Birley (1981, 2005) has it alive to about 124 with the place undetermined. Hodgson (2021) argued for defeat followed by disbandment in Britain, in 122 or shortly after, and Graafstal (2025) explores the connection with a security crisis after Hadrian's visit as a possibility to be examined. Neither paper has yet drawn a published response, and the field has no settled verdict.
The analysis reaches the conclusion Hodgson argued and Graafstal explores by weighing the evidence they used, so it supports their reading without testing it independently, and it finds no new evidence. It does not show that the field favours the reading: Campbell, A.R. Birley and Keppie hold other positions, and no published response to Hodgson exists. It rules out the traditional date of 117 to 119, which A.R. Birley had already done in 2005, and it states, as the proponents do not, how far the answer depends on judgement: 85% at the base reading, 36% at the sceptical one, about 50 to 65% on central readings.