sandstone roughly squared with the hammer. Behind it for a width of seventeen feet were two parallel lines of oak branches laid close together at right angles to the ditch. Eight feet from the front of this layer of branches and eighteen inches above them was a single line lying in the same direction. From the front of the upper line other branches could be noted sloping down to the front of the lower line, but these were not placed so carefully as the horizontal layers. All over, the average thickness of the branches was from two to three inches, and some of them appeared to have been split.

We have here perhaps an example of something akin to the cervoli, or tree trunks with branches attached, which Hyginus prescribes for the base of a rampart of turf, where from the nature of the soil the sod breaks because of its excessive softness, and a fosse cannot be made without the sides tumbling in,[1] a method of construction of which we have an example in the Fort at Coelbren, in South Wales, which probably dates from the first century.[2] 'The section did not disclose any quite definite distinction between the outline of the earlier rampart and the clay forming part of the rampart of the enlarged fort subsequently placed above it. But on the side towards the ditch a darker band of colour seemed to indicate that the earth from the ditch below had been thrown on the top of the layers of branches. What further defence, if any, was added to the rampart and ditches we cannot tell. On the south side the sockets of large posts were met with in three places on the line of the rampart behind the ditch of the early fort, and it is possible that these may have been foundations of a palisade.

In front of the rampart were the ditches. A double line of them ran round the western half of the fort, but round the eastern half the line was only single. They were from four to five feet in depth, and nine to ten feet in width. All were of the ordinary V-shape, which was commonly used by the Romans. Much the most interesting point connected with them was the evidence they furnished as to the gates. Examination proved that they were not continuous, there being a clear break at each of the four entrances. Moreover, in all four cases, the ditch on one side of the passage was some forty feet in advance of the line of the corresponding ditch which advanced to meet it from the other side, while the extremity of the line that lay furthest out was curved inwards so as partly to cover the

1 Liber de Mun. Castr. c. 51.

2 Archaeologia Cambrensis, vol. vii. 1907, p. 129.