MIRACLE CYCLES. 303

except where W is also parallel (or may be sup- posed to have been parallel when complete), so x is nowhere parallel to W except where C is also parallel. This of course proves, what indeed is evident at a glance, that C cannot have borrowed from x> nor W from C. But it also justifies our saying that in all probability W did borrow from Y, C from W, and x from C, if for the moment we allow these symbols to stand for types of text instead of individual manuscripts, actual or hypo- thetical. A table I have prepared will illustrate this part of my argument in a rather striking manner. I have assumed in drawing it that the rules which govern the relation of the texts where all four plays are available also apply where W is defective. But there is a third general fact to be noted namely, that C is nowhere parallel to W except where W is parallel to Y, and x is nowhere parallel to C except where C is parallel to W (if W exists). 1 In other words, assuming direct borrowing, while C only borrows from Y by way of W, and x only borrows from W by way of C, it is also true that C borrows nothing from W but what W borrows from Y, and x borrows nothing from C but what C borrows from W. This is a most remarkable state of things. How comes it that C and x> in borrowing from W and C respectively, avoid bor- rowing any original matter? It is this paradox 1 The table unfortunately does not illustrate this, since it does not show the portions of W, C, x which are not parallel to Y. But within the portion reproduced it may be observed in the parallel extracts.