How Do Mistakes Enter Manuscripts?
Medieval monks made the same basic errors you make when you are typing from a handwritten model. The simplest is "dittography," repetition in type of the same word, produced because you looked up for a moment and then restarted with the word you already had typed. Computers make this easy when the text scrolls off the screen at the bottom, but it used to be rare in transcriptions on full-sheet manual and electronic typewriters. The marvelously named "homoeoteleuton" drops text from within a line (or many lines on a page if you're really careless) by the same process. Your eye notices a keyword in the line you just copied so you can find where you should start transcribing when you go back, but when you return to your exemplar manuscript, from which you're copying, your eye discovers another use of that same word further down the page. You'll never run into the missing text, and probably won't know it's not there. But your reader usually will notice it, so that's why they hired proof-readers or delegated that job to the head of the scriptorium who supervised the scribes. The most subtle kind of error would be produced because you couldn't properly read the exemplar's script, didn't know the language it was written in (as when you try to quote a foreign language you're not proficient in) or couldn't see in the dim light, which scribes often excoriate in marginalia lamenting their hardships (also cold, cramps from sitting in one place too long, poor pens or vellum, and obnoxious colleagues). The "misreading" error is detectable usually because it mangles the sense of the sentence, but how to correct it? You look at words which would look similar in the scribe's hand (i.e., ModE "handwriting") seeking a harder word that might have been mistaken for an easier one. Manuscript scholars long have assumed that such misreading mistakes almost always convert harder words to easier ones, and the principle of hypothetically correcting the backward to the more difficult reading is called "difficilior lectio potior" or "difficilior lectio" for short ("the more difficult the reading the more likely"). However, these errors generally produce only minor mistakes. Another sort of error can produce genuinely catastrophic problems.
If you paused in your copying and put a marginal notation in a new manuscript you were working on to indicate the proper "shelfmark" (like our Dewey Decimal or LOC notation) where you could later find the proper exemplar or source manuscript, and if that shelfmark was incorrect, but pointed to another very similar but incomplete manuscript, imagine what would happen the next day when your assistant started to finish copying the new manuscript. He walks to the shelf and reads along it to the right number, and picks out the wrong exemplar--voila, a medieval bungled hyperlink! He starts copying at the proper place in an improper manuscript and suddenly that version of Canterbury Tales winds up jumping from Man of Law's tale to Squire's tale, instead of Wife of Bath's, and maybe she gets stuck with the Shipman's tale or maybe tells no tale at all (if the second MS puts her before Man of Law, which you already finished copying. The patron will be very angry if s/he notices the omission, but maybe the supervisor of the scriptorium who checks the quality of the copying will notice the error and put a crux or "X" in the margin where Scribe 2 picked up the wrong manuscript. Then, instead of firing up FrontPage and altering the link, they'd have to use a pumice stone to erase the faulty lines, and they'd probably toss the remaining mistakenly copied leaves on the scrap pile for use in strengthening book bindings, where sometimes modern scholars will later find them.
Scholars are interested in any surviving medieval MS fragments because they might contain a unique "witness" to some previously unknown state of the text (e.g., Chaucer tries out an ethnic accent for the Wife? the Wife hits on Harry Bailly?). The manuscript fragments would be found because the old binding fell apart and revealed text in its "bands" or strips running across the binding, or in the papers lining its boards (cover). ,The fragments also could be found because they were in a collection where they systematically examine suspect bindings looking for that sort of thing The hard part is reassembling the pieces to recover the whole manuscript, since the books into which it may have been stitched might now be scattered from Moscow (very good Middle English collection) to the Vatican (ditto, esp. legal and theological texts, but also [by rumor] a secret library of dangerously naughty books sequestered by the popes in their inner library). They're also distributed in private and public collections scattered all over England, and in America and the other lost colonies where people might have taken their family libraries in the C17-20 colonial expansion. Online sites exist to help scholars locate scattered pieces of a manuscript of which they hold an example.
In the mean time, surviving manuscripts which contain such catastrophic copying errors generally have been detected by nineteenth- and early twentieth-century scholars who worked according to statistical and logical principles pioneered by German philologists who were tracking word meanings' evolutions over time. They borrowed their notion of "manuscript descent" from emerging Darwinian theories about species' evolution, and developed trees of manuscripts from exemplar to copies to copies of copies. Essential to their work were those very scribal errors which mangled the text, but which were like uniquely broken bits of genetic code in descendents who share the same defective parental DNA. Though this method has been subjected to criticisms and refinements, it still remains the basis for the Chaucer manuscript families which you see in the Riverside Chaucer on page 1121, and it underlies many editorial decisions about which words belong on the pages of RC as our best estimate of the text Chaucer intended to create. Thus, while English and American scholars were going through the theory wars over New Criticism's "Intentional Fallacy," medievalists just kept their heads down and committed the "IF" with grave deliberation, believing that it was their method in a nutshell. Only after they could reach consensus, for a few decades, about the proper form of Chaucer's text, which I would date roughly to the gradual acceptance of Robinson's second Houghton Mifflin edition (first pub. 1957), could scholars begin to play the theory games which had been preoccupying their colleagues in later, better-documented eras.
For Stephen Reimer's (U. Alberta) description of these and other kinds of scribal error, and the editing practices developed to emend them, click here. It's a demonstration site for a Manuscript Studies course that's not actually being taught because it's still in development, but it's a very thoroughly developed model and has a set of excellent links and bibliographies on the subject.