‘Examination’ is the process by which one moves from the ‘raw materials’, as it were, the diverse forms revealed by collation, to an edited text of the work in question. Historically, this process has been central to discussions – and often a subject of great acrimony. Collating any text throws up variations: how does one decide which one, if any, actually represents the common source? (You may recall the note at the end of my introduction, pp. 13–15, in which I offered a skeletal analysis of an example from Cursor Mundi.)

Simply considering the matter abstractly, there would appear two ready courses to follow. The first might be defined as ‘taste’, the second ‘attestation’. In the first instance, one chooses, on the basis of whatever inner standards occur to one, the reading that one likes best. There might be various ways of articulating such standards, ranging all the way from considerations apparently rational, for example, the lexical argument I brought to bear on the reading from Cursor, to ones frankly intuitive, one's particular fascination with a certain word, for example. A procedure like this probably guided most editors down to the eighteenth century, notoriously reticent about their practices.

The limitations on operating in this fashion are fairly obvious. Individuals of intelligence and good intent may obviously display different ‘tastes’. Thus, the prospect opened by this way of proceeding has always been perceived as falling well short of ‘critical’. Inherently, it implies every person his/her own editor, a range of varying editions, and no agreement about the nature of the text or of its transmission. Unfortunately, as we will see, this may prove an unavoidable danger.

On the other hand, one might attempt to assess the strength of support for any single textual reading. How many manuscripts provide this version? What does one think of their general reliability? The term ‘attestation’ refers to the strength of support for any given reading: how many manuscripts attest to (offer a witness for) this textual version? This opens the possibility that one should simply examine the collation on a numerical basis. In this scenario, any reading attested in but a single copy is probably wrong; those attested by a majority of the witnesses likely correct. One would simply weigh numerical attestation and insert into one's text the most popular reading.