One of the curiosities of textual criticism is its assumption of continuous processes: It is usually assumed that a text, once it started in a direction, just kept going in that direction. So the Alexandrian text just kept getting shorter, the Byzantine smoother, etc.
It should instantly be evident that this notion contradicts most theories of the text. They assume that most major variants arose before the manuscript era. But if they predate the manuscript era, then there was a change in process: The production of variants stopped.
It is quite likely that the history of manuscripts is not a continuous process, but rather a complex history of destruction and reconstruction -- of copies getting gradually worse with each generation and then periodically being subjected to a systematic improvement.
Consider: It is universally agreed that the most common variant in copying a manuscript is haplography -- a loss of certain words or individual letters. If this process continued unchecked, every late text would be short. Yes, manuscripts were corrected after copying -- but correctors don't catch everything. Even if only half a dozen haplographies sneak through one copy, run such errors down a dozen generations and you get a short, badly corrupt text.
And yet, our late manuscripts, whatever else they are, are not short and show none of the errors of this sort of repeated bad copying (for a text that does show this sort of problem, look at I Samuel).
The logical conclusion is that Biblical texts have been subjected to reconstruction -- that is, that the old copies have been carefully examined and improved to correct the various losses.
The meaning of "destruction" is probably obvious. Scribes make haplographies. Pages may be lost from their exemplars. (This is demonstrably true in manuscripts of Arian, but it may also explain the loss of Mark 3:28-4:4 in 579.) A word or two may be damaged by damp. Errors will naturally multiply.
Reconstruction is a more complicated matter, which gets little attention. Critics admit two levels of attempts to repair texts: Correction and recension. Reconstruction is neither of these; it falls somewhere in between.
Correction is a relatively feeble process. At best, correction can only improve a text to the measure of the standard against which the document is compared. That is, if Y is a copy of X, and after correction, Y is compared against X, this process can only find places where Y deviates from X. It cannot produce better readings than those found in X. And if Y is corrected against something other than X (call it Z), it still can't produce anything better than Z.
And chances are that Y won't be even as good as X, or Z, because the scribe making the corrections probably missed some things.
We can see this in action, by looking at, for instance, Codex Claromontanus. This manuscript started with a "Western" text. It was corrected, repeatedly, against the Byzantine text. I examined the readings of Colossians (as found in the NT auf Papyrus.) All told, I found 121 places where D* went against the clear reading of the Byzantine text. 105 of these readings were eventually corrected -- after two major and sundry minor corrections of the manuscript. That still means that more than one error in eight went uncorrected -- and the correctors introduced some few errors of their own. Plus, Claromontanus was copied before the final correctors worked, and the scribe who copied it had difficulties with some of the correctors' notations. So Dabs, intended to be a Byzantine manuscript, wound up with dozens of deviations from the Byzantine text -- most but not all of them in the direction of the "Western" text. Simple correction, no matter how many times repeated, cannot prevent destruction of the text. It merely slows the process. To give an analogy: Correction alone is like giving transfusions to a man dying of blood loss. It slows the death. But unless the wound is closed, the bleeding will continue until the victim dies.
Thus there is need for the rehabilitation of texts. Sometimes this rehabilitation is the result of recension: The detailed comparison of multiple texts to produce a full-blown new edition intended for widespread publication. We know that Alcuin and Theodulf produced recensions of the Vulgate. It is also extremely probable that the Kr edition of the Greek Bible is the result of recension.
But recension is a very major undertaking. It entails gathering several sources, comparing them, producing a composite edition -- and convincing others to adopt it. This takes both resources (access to multiple copies, plus a good deal of time and material) and prestige (a recensional text produced by someone with no authority isn't likely to be widely promulgated).
What's more, recension implies a very strong goal: To impose one's corrected text. It's not likely that most scribes had such lofty expectations. They just wanted a good text for their own use. For this purpose, they wouldn't go out and compare a dozen manuscripts; instead, they would take what they already had, and compare it with perhaps one other, or go over their text and look up particularly troublesome passages.
This is where knowledge of items other than the Bible can help. We have very many instances of this phenomenon in other works. Take, for example, the traditional song "Boney on the Isle of Saint Helena." This particular song, about the death of Napoleon Bonaparte, is fascinating because -- although recent by folk music standards -- it has gone very badly to pieces. I've had occasion to examine ten collected versions of this song, no two of which were identical. It happens that two of these were collected from the same singer, eighteen years apart. The second collection differs substantially from the first, notably in the inclusion of an additional verse. It appears that, in the interim, the singer had listened to additional versions of the song (very widespread in his area of North Carolina), and built up his own text. The result was the fullest text of "Saint Helena" known to me -- but also, based on the evidence, the best. It wasn't a recensional product -- but it was the result of working over other versions as the singer came across them.
We see something similar in certain Shakespeare plays. As an example, consider Titus Andronicus. The editions of this play reveal quite a bit. There was an early printing (Q1) from 1594. This printing served as a basis for a printing in 1600 (Q2). However, the copy of Q1 used to set Q2 was damaged, and the compositor of Q2 emended it conjecturally. Q2 was used as the basis of Q3 (1611). Q3 was used as the basis for the First Folio (F1) printing. However, someone (perhaps even Heminge or Condell, the actors who promoted the publication) seems to have noted a missing scene. As a result, F1 contains, for the first time, a text of Act III, scene ii. In general, F1 has a late and inferior text -- but it has been reconstructed at this point, and is superior to all other witnesses.
That is not to claim that reconstructed texts are generally superior to unreconstructed texts. They are merely longer. Consider, for instance, the case of Codex Vercellensis (a) of the Old Latin. Here we can literally see reconstruction taking place. The old text of the ending of Mark has been excised (with a knife!) and a new text supplied. It is believed that a in its original state lacked Mark 16:9-20. So a vulgate text of these verses was supplied. We note that the result has absolutely no critical interest or value (we have plenty of copies of Jerome's version of Mark 16:9-20, and none of whatever text existed in a). But it shows a reader examining the text, being concerned, and attempting repairs. Multiply this by dozens of instances (from the careful work done on 1739 to the likely use of conjectural emendation on D/05) and you see why New Testament manuscripts, despite the general tendency for texts to decay, managed to stay quite full until the very end of the manuscript era.
I can, as I write this, feel the fans of the Byzantine text latching onto this description with glad cries and preparing to use it to condemn the Alexandrian text. It's not that simple. I am prepared to allow that the Alexandrian text is almost certainly too short. That does not make it inferior. A crucial question is, when did reconstruction begin? If the Byzantine text is reconstructed from the Alexandrian (which is possible), then in general the Alexandrian text is still superior. It's defective, but it has not had the additional layer of bad reconstruction we find in the Byzantine text. (In Hort's view, for instance, the Byzantine text came about, in effect, by reconstructing the Byzantine text using the "Western" text as a source of variants. Only if the Byzantine text is a result of reconstruction beginning before the current condition of the Alexandrian text does it have independent value. And even then, it is merely independent value.
We should note that reconstruction is not really a single process. Some manuscripts, like 1739, have been reconstructed by comparison with other texts. Others, especially early in their history, were probably reconstructed by conjectural emendation. Other forms of reconstruction might occur in special cases -- e.g. a one of the synoptic gospel might be compared against another gospel (one wonders if this might not explain some of the heavily harmonized "Caesarean" texts), or against the Diatessaron, or even against a version in another language.
In the history of most ancient texts, including the New Testament, there were several points at which reconstruction was almost imperative: The times when new "features," such as accents, breathing, punctuation, or word division, took place. In addition, there was the conversion from uncial to minuscule. When any such process is undertaken, the copyist must examine the text in detail, deciding where to put the features. This will force removal of ambiguities. In some cases, the scribe will do it by reference to another copy, though there will probably be instances of conjectural emendation also. Another possible inspiration to reconstruction might be the preparation of commentary manuscripts: If the editor who inserted the scholia observed that they differed from the text of the manuscript, he might adjust the manuscript. Or a scribe copying a commentary manuscript might level the differences.
Commentary manuscripts offer another opportunity for reconstruction: The time when the commentary was added. Indeed, the addition of almost any sort of marginal equipment would encourage reconstruction. If a scribe is adding the Eusebian apparatus, for instance, this encourages the scribe to look at the text to see just where the markings go in.
For a true commentary manuscript, with marginal scholia of some sort, the temptation must have been even stronger, and there are suddenly two possible sources of variants: The text of the manuscript supplying the scholia and the scholia themselves. The tendency to level would have been great -- and not necessarily confined to the text being modified. If the copyist found that both the text before him and the scholia assumed one reading, but the text of the original commentary manuscript read something else, might not the corrections go the other way?
If it be objected that we have no evidence of this, I will admit that this is true. But this process took place mostly in the "silent centuries": The sixth through seventh centuries, from which we have almost no substantial manuscripts. From the fifth century and earlier, we have a variety of full manuscripts, with at best intermittent reader helps, and a variety of text-types. When the dark age ends, with E and L and their followers, we have manuscripts well endowed with the reader helps. We also have a much more Byzantine constellation of witnesses. Coincidence? Maybe. We have no way to tell.