The twenty-eighth edition of the Nestle-Aland (hereafter NA) text differs from the previous edition only in the General (Catholic) Epistles, yet the 34 changes found there represent groundbreaking work done in New Testament textual criticism and led chiefly by Gerd Mink of the Institut für neutestamentliche Textforschung at the University of Münster. Thanks to Mink and his colleagues, we now have the Coherence-Based Genealogical Method, a powerful tool to assist in the construction of stemmata for textual readings in the New Testament. Credit specifically goes to Klaus Wachtel and Volker Krüger for writing the application. The impact of the CBGM is due to the fact that it is to be the basis for the NA text from the 28^th edition on.

Existing Greek New Testaments

Some explanation may be needed, beginning with background on the state of the NT Greek text up to the 28^th edition. Until very recently Bible translators and their publishers essentially had three choices of Greek texts: the one based on the oldest and acclaimed “best” manuscripts (called “Alexandrian” by proponents and opponents alike); the Textus Receptus or TR (used for the KJV); and the Majority Text or MT (similar to the TR but differing from it in significant places of especially weak manuscript support). The first text could be found either in the NA or in the Greek NT published by the United Bible Societies. Nearly all modern Bible translations were, and continue to be, based on the NA text. The NKJV, as its title implies, has followed the TR but noted the readings in the MT and NA.

We should also add by way of explanation that users of the NA text have always had access to a substantial collection of variant readings (including MT or Byzantine readings) supplied in footnotes, so translators have been free to choose readings rejected by the editors of the text. They also have had the valuable aid, A Textual Commentary on the Greek New Testament by the late Bruce Metzger, a complementary volume that provides brief explanations of many of the decisions made by the editors, and rates the level of certainty for each.[1] So translators have never been forced to follow a particular text, though they may have chosen to.

Setting aside the TR and MT for a moment, let’s focus on the NA text and what has led to the 28^th edition, current at the time of this discussion. Without going into all the history which is readily available elsewhere, the famous (or to TR advocates, infamous) scholars Westcott and Hort et al. defined the quest of textual critics as that of identifying the variant among other variant readings for a given passage which best accounts for all the others. That is, in any passage of which there is more than one version in the extant Greek manuscripts, the goal is to determine the original reading, which must have been changed into the other forms rather than the reverse. This is actually common sense of the reverse engineering variety. We know that scribes (copyists) at one or more points in time changed the original to something different for some reason(s); we want to reverse the process mentally by determining what they were thinking or failed to do properly, which will lead us back to the original reading, assuming that at least one of the extant manuscripts has it.

Of course, we cannot read the mind of any scribe, and we cannot debrief anyone who is deceased. We are forced to do a credible “profile” of an ancient scribe, and we begin with the assumptions that he is competent and conscientious, unless the manuscript he has produced is a mess. The high degree of agreement that we find among the extant Greek manuscripts tends to confirm these assumptions.

Differing Families of Manuscripts

Nevertheless, the manuscripts prove to disagree in many places, and we now know (thanks to CBGM research) that the level of agreement between all the passages that feature variants–in the General Epistles at least–averages about 88%. This agreement is called “coherence” in the new method. Textual scholars have necessarily focused on the disagreements, and the number and various kinds of disagreements led them to postulate that there were “families” of manuscripts with differing levels of quality. To simplify a complex situation, one can attempt to narrow the principal families to three, including the two already mentioned above: the early (Alexandrian) family and the later manuscripts, roughly lumped together as the MT or Byzantine family. There are very many more of the latter, leading to the name “Majority Text” by its advocates. The third family is usually called the Western text, best represented by codex D, Bezae.

A characteristic of the Byzantine family is that when there are differences with the earlier manuscripts, the Byzantine readings usually are “easier” for one reason or another: sometimes the reading is smoother, at other times it avoids a theological difficulty or some other problem. Westcott and Hort, in their quest for the original text, came to the conclusion that the Byzantine family was the result of misguided scribal attempts to correct or improve upon the original and therefore was to be rejected. They preferred the more ancient Alexandrian manuscripts. Advocates of the Byzantine family have argued that the scribes wisely corrected corrupt readings found in the earlier manuscripts. Westcott and Hort also rejected Western readings for the most part, which tend to be embellished.

Downgrading the Byzantine manuscripts has met with favor among most textual critics, and most have also preferred the Alexandrian family. However, Westcott and Hort went further in their admiration for these manuscripts by making the extraordinary claim that we have the original text whenever the two leading manuscripts of the family–Aleph (Sinaiticus or 01) and B (Vaticanus or 03)–agree.

In practice, this claim appears to have met with considerable acceptance, because one can see that Aleph and B tend to prevail in textual decisions seen in earlier editions of the NA (and GNT) text. But in more recent years a backlash to this position developed, and the NA editors made a concerted effort to minimize this effect and “dethrone” Westcott and Hort. As a translator, I saw the results in some places where the editors changed their minds about previous readings and placed other readings in the text that had the support of seemingly lesser manuscripts chosen over Aleph and B and even had some Byzantine support. Of course, it did not mean that I had to agree with the editors, and if they selected the “easier” reading with lesser manuscript support, I rejected it.

Thoroughgoing Eclecticism

Another challenge to Westcott and Hort, and to other traditional textual critics, has come from opponents who advocate what is often called “thoroughgoing eclectic criticism,” briefly discussed earlier. The keyword here is “thoroughgoing,” because in practice all textual critics do eclectic criticism to a greater or lesser extent, usually called “reasoned eclecticism.” Thoroughgoing eclecticism can be differentiated as “internal criticism” in opposition to “external,” because it is limited to choosing a reading based on contextual and transcriptional factors. Actually, it is easier to define this practice by pointing out what it does not include: dating of manuscripts, the quality of individual manuscripts, and where geographically they originated and were used.

For example, in traditional criticism, a reading found in B (Vaticanus) is taken seriously at the outset because B is an old (fourth-century) manuscript rated as very high in quality. In the thoroughgoing eclectic approach, B’s pedigree, as one might call it, does not matter. The reading is judged solely by how likely it seems to be the original, given the context and any transcriptional issues. This is entirely in line with Westcott’s and Hort’s quest for the reading from which the other (invented) readings came, and the same rules and guidelines are for the most part applied. For example, both thoroughgoing eclectics and traditional textual critics favor the harder reading over the easier one.

Another way to better understand thoroughgoing eclectics is through one of their complaints, which is also reflected in the word “thoroughgoing.” They often say that they will agree with reasoned eclectics in evaluating readings internally to determine the most likely original, only to find their counterparts rejecting the reading judged best from sound (internal) criteria and choosing another reading purely on the basis of external criteria. For example, J.K. Elliott voices his agreement with most of the TC guidelines taught by Metzger, but then he says this about many of the discussions he finds in Metzger’s Commentary:

Where I have found grounds for criticism is that the editors of the United Bible Societies’ The Greek New Testament (= UBSGNT) or of the allied Nestle-Aland testament (= NA) often jettison their own principles of internal criteria if they, or a majority of them, did not approve of the manuscript support for the reading that the internal criteria pointed to as original.[2]

 To critics employing reasoned eclecticism, this is simply a balanced approach, taking the external criteria of age, quality, etc. into account together with internal factors. Employing external criteria is also seen as a way to avoid wholly subjective decisions about readings.

An excellent example of this situation can be found in 1 Pet. 4:16, which ends with “…glorify God in this name,” as seen in most versions. It turns out that the word “part” in Greek is found in Byzantine manuscripts instead of “name,” which is supported by the oldest manuscripts, including Aleph and B. Indeed, on the basis of external evidence, the decision for “name” is easy, and this was the reading of the NA text down to the twenty-seventh edition. The problem is that “part” clearly is the more difficult reading; so we must ask, “Why would a scribe change ‘name’ to ‘part’?” It is easy to see why the opposite might occur. Contextually, therefore, “part” should be the original reading, and the editors of NA28 have placed it in the text, based on CBGM analysis.

You may infer from this example that the CBGM is a new version of thoroughgoing eclecticism; in a real sense, it is, though its advocates would probably protest this characterization. Nonetheless, consider that emphasis is placed on the fact that manuscripts play no part in the analysis; they are merely the physical carriers of text, and only the texts they carry matter. Consequently, the date of an individual manuscript is of doubtful or no importance, and users of the method are told to set aside previously held views of the quality of any manuscript, which can get in the way of doing an objective analysis of the text. Say goodbye to any shrines dedicated to Sinaiticus (Aleph/01) or B (03). Not only that; we are also told that the theory of a Lucianic recension resulting in the Byzantine text-type, or any such early recension, is not at all helpful and should be jettisoned as potentially harmful clutter, along with the now-obsolete concept of text-types.

It would certainly appear from these advisories that the CBGM aims to dethrone Westcott and Hort and that advocates of the Byzantine text have new reason to rejoice. I have to confess that I was more than a little skeptical of the method and might not have felt compelled to deal with it were it not for the fact that the NA text is to be determined by it, apparently for the foreseeable future. As a result, intellectual and spiritual honesty compels translators to learn the method, or at least to become sufficiently acquainted with it to be able to intelligently accept or to honestly reject it. For those who do the latter, I suspect that the NA text will become frozen in time at the 27^th edition (some may already prefer an earlier edition).

A Detailed Introduction to the CBGM

I trust that this background is sufficient to prepare the reader for an introduction to the CBGM, which probably will be the future of textual criticism from this point on. I can express its greatest strength in one adjective: meticulous. By contrast, an adjective that I would use to express the greatest weakness of textual criticism up to this point is: subjective. Because of the subjectivity, many textual critics have given up calling the discipline a science (if they ever did), or even a science-art. I do not mean to disparage the claims already made in this book for TC as a science or to dispute the subtitle of this book. In a general sense, one can still call textual criticism a science because it does have principles and rules that have found almost universal agreement among scholars of the school. It deals primarily with natural human behavior, as for example, psychology does. However, in comparison to a laboratory science, it is just an art, in that it lacks sufficient empirical measurements. This seems frightening when one considers that we are talking about identifying the original text of the Bible as the goal. Like dealing with life-and-death situations, we want to be as correct as we possibly can be.

The CBGM is possible because it is built on a computerized database, i.e. a collection of all the variant readings in carefully selected Greek NT manuscripts (well over 100 thus far) converted to machine-readable code for the purpose. A collation of the same data has been published separately under the title Editio Critica Maior 2^nd Revised Edition (ECM2),[3] to which we have already referred. Over 3,000 variant passages are covered there and by the CBGM. I’m sorry to add that there is one piece of very bad news: as I already indicated, only the General Epistles have been fully input and processed to date, and it looks like it will be a very long time before the rest of the NT is included. This is bad news, that is, if you decide that you like the CBGM. I am undoubtedly the bearer of good news if you don’t like it. We will probably continue to see new editions of the NA text in partial stages of completion, like NA28.

I think there is no denying that the CBGM has some major shortcomings, which I will discuss later. However, being able to manipulate so much data for the biblical books covered thus far has led to interesting discoveries under the sharp eyes of Mink and his colleagues. First, though, you need a basic understanding of the CBGM, beginning with the concept of “coherence” upon which the system as titled is based.

Coherence, the Basis of the CBGM

Coherence is the level of agreement between any two texts that are found in any two manuscripts, down to individual characters and spaces. The term “text” is very important. Essentially a text is what a scribe copies from one manuscript to another, which is a simple concept. If we are looking at a long historical chain of copying an ancient text, however, the text (i.e. the copied words) that we find in any manuscript may be much older than the manuscript itself. Think, for example, of buying a copy of the Declaration of Independence at a museum. I accept–at least for the purpose of this illustration–the museum’s guarantee that the text of the Declaration in my copy is the authentic eighteenth-century document, but the physical copy that I purchase is twenty-first century. My copy probably was one of hundreds or more that were printed by a machine using a printing plate with the image found on my copy, and like mine, every copy produced is a perfect reproduction in the sense that all the characters and spaces exactly match the original image. For example, there is not a single letter in the text that is different in some copies. In CBGM terms, this is perfect coherence. I could diagram it by representing the original as a circle from which arrows go out to all the hundreds of copies that were made. Every copy has the original as its one and only source.

Let me give another example of perfect coherence that is encountered in the CBGM. I could duplicate at home what the machine essentially did with the Declaration of Independence reproductions by using my home photocopier to run off a hundred copies of the copy I purchased at the museum. I would diagram the process the same way, with a hundred arrows going from the original (my copy in this case) to all the copies. However, I could also do it by making the first copy, then a copy of the first copy, then a copy of the second copy and so on. Each copy would then be the ancestor, as it is called in the CBGM, of the next copy, and the arrows would point from one copy to the next after starting with the original. That is, instead of a starburst of arrows (all copies directly from the original), I would have a straight line from one copy to the next. The text of each “manuscript” would be a perfect copy of its ancestor.

Mink’s Four Assumptions about Scribes

Now that we have a good grasp of perfect coherence, it follows that imperfect coherence occurs when texts do not entirely agree. This, of course, is the case most of the time in textual criticism since we are dealing with hand-copied documents. We will look at this in more detail below, but first, we have to profile our scribes. This Mink does by postulating four basic assumptions. I will list them much as he does, but with additional clarification:

First, a scribe wanted to copy his exemplar (German Vorlage) with fidelity.[4] To provide a clearer idea of each assumption and its probability, I will posit the opposite, like including an antonym as a means of clarifying the definition of a word. In this case, I will posit that the scribe did not care in the least whether his copy was accurate. We could assume that our hypothetical scribe was lazy and wanted to be done with the job as quickly as possible. If so, his copy would vary widely from the original and probably omit a good deal of it. Misspellings and other mistakes would be common. Since all these assumptions apply to the average scribe, we would expect to see vast differences between all the copies that are still extant, and few of them would be credible as representatives of the original. The fact of the matter, however, is that what we find in the extant copies is the opposite, supporting Mink’s assumption. Logically, we probably should expect it to turn out this way given that the exemplars were copies of Scripture.

Mink’s second assumption is that if a scribe intentionally introduced diverging variants (which led to imperfect coherence), he obtained them from another source (manuscript), rather than inventing them himself. The opposite of this assumption would be that whenever a scribe deviated from his exemplar, he made up the change himself, oblivious as to whether the same change already existed in another manuscript elsewhere. A corollary to this opposing assumption would be that the scribe most likely had only one manuscript as his exemplar from which to copy.

This second assumption is somewhat problematic in comparison to the first because we know that its opposite must ultimately account for the changes that we find in extant manuscripts; that is, for every variant reading, some scribe had to invent and introduce it the first time. As for the corollary I added, it is either true when such a change occurred, or if there actually was more than one manuscript available to the scribe in question, then the manuscripts agreed on a reading that the scribe rejected, or the scribe ignored or found any alternative readings available not to his liking. To some extent, however, the evidence from extant manuscripts which supports the first assumption also supports this assumption: over nearly two millennia there has been plenty of time for these kinds of changes to multiply dramatically, and in fact what we find are undoubtedly far fewer changes than we should expect if scribes exercised considerable freedom to invent changes.

Let me point out one more thing about the corollary: it seems unlikely that a scribe would actually treat more than one manuscript as his exemplar unless he were compelled to do so. One such scenario, which would not be uncommon, is that the exemplar would be missing some passages and another manuscript (or manuscripts) would be needed to fill in the missing section(s). The phenomenon of differing readings, however, forces us to infer that the scribe either invented the variant in question or found it in another manuscript. According to the second assumption, the scribe first consulted any other manuscripts available to him for an acceptable alternative to what he found in his exemplar. If he actually did find a reading he preferred, one would think that he would switch to that manuscript as the new exemplar if he were still in the copying process. Large blocks of texts that represent variant readings from a possible exemplar indicate that this may have happened to a degree.

On the other hand, the more typical variants of much shorter lengths indicate to the contrary that scribes kept to one manuscript as the exemplar even after resorting to a reading found in another manuscript. This suggests to me that the exemplar was favored for some reason and that the scribe was either assigned the task or considered it part of his task to, in effect, “correct” the exemplar when a reading seemed suspect. Whether or not he made a notation on the exemplar itself, the change became an integral part of his copy, making it a new descendant in the textual flow. In his chapter “Contamination, Coherence, and Coincidence,” Mink observes that a frequent feature of contamination is that a witness does not agree with its immediate ancestor in the CBGM analysis (the technical term is “closest potential ancestor”), but with the immediate ancestor of the latter.[5] This is just the sort of thing we should expect if an exemplar was favored–I dare say “popular”–but considered in need of correction on the ground that its scribe had deviated too much from the exemplar that he used.

The third of Mink’s four assumptions is that scribes used few rather than many resources. This assumption is the result of practical and psychological considerations. On the practical side, it seems doubtful that a scribe would have access to more than two or three manuscripts at a time from which to make his copy. If he were copying one manuscript mainly and making occasional changes, it seems most likely that he would introduce changes from one other manuscript, as in the scenario I just described. Psychologically, the job of having to make changes taken from more than one additional manuscript probably would seem taxing. The opposite assumption would be that scribes normally made their copies from “many” resources, and while “many” is hard to specify, I would think something in the neighborhood of a dozen or more would qualify. Pity our hypothetical scribe who is given a large stack of manuscripts and told, “There you are!” It is also hard to imagine “many” manuscripts of the same text being available at one place and one time.

In presenting the third assumption, Mink also makes much of an artificial limitation in the process of constructing “optimal substemmata” later.[6] These stemmata are best done with a minimal number of witnesses of readings, and the number of witnesses actually required to account for any given text in its entirety possibly depends on this assumption. It is at least a good working hypothesis because the number of witnesses essential for the text can logically be reduced, and this helps to make sense out of what otherwise might be a considerably more complicated genealogy. Historically of course, as Mink admits, we cannot know what actually happened. For example, extant manuscript X might be the one containing a text that has several particular readings found in manuscript Z. Therefore the text of X must be included in the optimal substemma of Z; however, in reality, the scribe who produced Z might have used one or two other manuscripts whose texts X reproduces. By the same token there may have been manuscripts overlapping in textual content that was actually used by scribes in the chain of the copying process, but manuscripts that duplicate readings serve no useful purpose genealogically and would only clutter the substemma. I should clarify again that in the CBGM terminology the textual content of a manuscript is also called a “witness,” closely associated with the manuscript but older (potentially much older) than it.

The fourth and final assumption is that the sources (manuscripts) used by scribes featured closely related texts rather than less related ones. As Mink notes, this assumption is a corollary of the first: that the scribe wanted to copy his exemplar with fidelity. The evidence of extant manuscripts, as we have noted, points to the truth of assumption one. What we find are manuscripts that are very similar in content rather than the opposite. The point of the copying process is to reproduce the original, so the “best” copies are those that are virtually identical to the originals, and the more a copy deviates from its original, the less good it is. As we noted earlier, the average agreement among texts that feature variant readings is about 88%, so this assumption seems well-founded. It proves to be very important to the CBGM because textual flow diagrams in the method and the stemmata of readings depend on it. That is, variant readings are assumed to be “inherited” from texts whose similarity to the text in question establishes them as “potential ancestors.” Indeed, the CBGM begins by creating lists of “potential ancestors” (a term that I will use without quotation marks henceforth) that qualify initially by their level of agreement with the witness in question.

I should note that the CBGM tools currently are available online in two versions, the original and version 2. I discovered that in comparison, the number of agreements between witnesses seems to decrease slightly in version 2 as seen in the “EQ” column. Not understanding how this could happen, I sent a question to the INTF, and Klaus Wachtel very kindly explained to me that two factors led to the changes: 1) corrections to the apparatus in v. 2 led to a decrease in the total number of variant passages by three; 2) in v. 1 of the CBGM 122 variant passages were undecided in the initial text and treated as lacunae (email from Wachtel 8/28/14). In v. 2 a decision was made in 79 of these passages. The first factor accounted for the changes I observed. What is especially important is that the number and percentage of agreements between witnesses are objective measurements, i.e. they are purely statistical and not in any way a product of human judgment (subjective evaluation). They are subject only to errors of the eye in their reading and to erroneous input or manipulation, which one hopes has been and will continue to be eliminated.

The four assumptions explained above are crucial to the success of the CBGM because they vastly reduce the complexity and the range of unknowns that the textual critic would otherwise face in dealing with the vast number of NT manuscripts now available. Therefore it is also crucial that the assumptions all be correct. The overall fidelity of the NT textual tradition indicates that they are. At the same time, however, I think it is fair to say at the outset that what is most likely true in general of scribal habits cannot be guaranteed for any particular manuscript. We must content ourselves with the expectation that whatever may have actually happened in the production of a given manuscript turns out to be only the logical equivalent of the process covered by the four assumptions.

Now then, I hope to offer an explanation of the CBGM that will be a little easier to grasp than most already published. Some observers have described it in statistical terms that can overwhelm readers without formal training in statistics. Mink himself uses statistical terminology from time to time, and a cursory look at the various tables and diagrams can be daunting.

However, as one of those without formal training in that discipline, I am happy to report that I do not recall encountering anything more challenging in the tables than a mathematical average or mean, in addition to percentages. In other words, the math could hardly be simpler. It is mainly the sheer volume of data that makes it appear complicated; the concepts, as I and others who have worked through the presentation and articles have learned, are sufficiently comprehensible. Moreover, the opportunity to obtain practical experience with them is available through the online tools that anyone can access at the INTF (Institut für Neutestamentliche Textforschung) website. A guide is included with examples that one can easily work through by using the tools. While hoping to offer an easier explanation of the CBGM, I do not mean to replace others already published, but only to supplement them. I strongly encourage the reader to read everything available for a fuller understanding of the method.

The Initial Text

There are at least two different places at which one could start even after examining the four assumptions about scribal habits: the so-called “initial text,” and the concept of coherence. Mink focuses on the latter and seemingly deemphasizes the former, but I think it is the “initial text” or Ausgangstext that is most likely to puzzle a novice to the CBGM, especially if he or she tries to follow the online presentation at the website.[7] There one finds the initial or “A-text” at the top of the heap of Greek witnesses to the NT.

One cannot fault Mink for ignoring disputes among textual critics over the meaning and nomenclature of the text that the discipline is trying to reconstruct, nor does he minimize the problems a novice or someone with a traditional understanding of TC might have with this “A-text.” On the contrary, in the early pages of the “Presentation,” he briefly mentions the leading theories of what the initial text might be, e.g. the author’s own text or the archetype of the textual tradition that perhaps differed from the authorial text in various places. The CBGM is not concerned with the theory, however, so he concludes by treating the A-text as the author’s text in order to simplify the situation. But he also emphasizes that the A-text is “a hypothetical reconstruction,” not “any actual historical reality.”[8]

Mink concludes the Presentation with, among other things, the extraordinary comment that the CBGM “does not make textual decisions,” it just reveals what a textual tradition will look like based on a given text-critical study of all the variants.[9] I found this most puzzling at first, given the fact that NA28 features changes based on the CBGM. So it would seem on the one hand that Mink treats the A-text as the autograph, while on the other, he distances himself and the CBGM from being any kind of final authority on the NT text.

I actually think now that this is a fair and accurate analysis. No extant manuscript on earth contains the A-text in all of its readings. The A-text is merely the product of a large number of textual decisions covering all the variant readings in the General Epistles, for now, and eventually the same will be true for the rest of the NT. These decisions were all initially made primarily on the basis of internal criteria, as opposed to the dating, provenance, and quality of individual manuscripts,[10] so in a sense, Mink is correct when he says that the CBGM does not make textual decisions. The A-text seems to have been established by textual critics as a preliminary to the operations of the CBGM. Despite its hypothetical nature, however, it is treated just like an extant text, and by default, it is assumed to be the autograph in CGBM analyses. That is, one has to physically select a different text as the “initial text” in the online tools to see a chart or diagram based on another text, and there is a problem with this that I will explain shortly below.

For the sake of greater utility, and I would also say objectivity, Mink has introduced into the CBGM the aforementioned option to choose any variant reading as the “initial text,” which for practical purposes can be considered the autograph. At the end of the Presentation, he even remarks that the CBGM can be used by those who prefer the Byzantine text. The significance of this remark becomes clear when one discovers that the A-text, as a result of the criteria used to establish it, is closely modeled by B (Vaticanus). Thus the Byzantine text, to many the original “A-text,” is the antithesis of Mink’s A (Ausgangstext)-text. So the application’s option of choosing a different initial text could be viewed as evidence that Mink is not using the CBGM as a means of delivering to us another Textus Receptus of his own making.

There is a significant flaw in this feature of the online application, however, to which I alluded: while one can choose any reading as the initial text, one cannot change any of the textual decisions that ultimately led to the genealogical relationships assumed by the application. It would be an even greater flaw if textual quality were linked to a particular text by the application, but the application itself does not do that. Any quality assessments are inferences by the user. Still, in light of the criteria of textual decisions, I think the claim that users from any textual position (notably Byzantine proponents) will find the CBGM a useful tool is questionable at best. If I favor the Byzantine as the original reading, for example, it follows that I also favor the easier reading as the original, uncorrupted form, and the harder (corrupted) form becomes the descendant. In the CBGM I can pick an initial reading different from A, but the application will work around my choice to present a stemma or flow diagram that is still consistent with the standard (non-Byzantine) TC criteria. I will be saying more about the criteria and related issues as we turn now to a detailed discussion of coherence, to which I briefly referred above.

Coherence and Contamination

Coherence, the first word in the title of the application, is a foundational concept developed by Mink to deal with the problem of contamination among the many extant NT Greek manuscripts. “Contamination” obviously is a negative term and not popular among all textual critics of the NT, but it is nonetheless useful in describing what we find. Some prefer a friendlier term, e.g. “mixture” or “crosspollination”.

As a technical term, coherence is a close relationship between two or more texts, bearing in mind that a “text” is not a manuscript but its written contents. In the CBGM, there are basically two kinds of coherence, “pre-genealogical” and “genealogical.” The first is the degree to which variant readings in any two texts agree verbatim in passages where variations are found among all the extant manuscripts. As noted earlier, the average level of agreement approaches 88%. However, in practice, the percentage at which agreement is considered significant by Mink varies in relation to the number of extant manuscripts found to be in high agreement with the text of a particular manuscript. This enables him to more easily categorize a text as primarily Byzantine or non-Byzantine by essentially giving him freedom in where to draw the line of distinction in the list of manuscripts. Some might say this is “fudging” to make the method appear more reliable than it really is, but then the concept of an average does allow some freedom of interpretation.

This pre-genealogical coherence is a highly objective measurement, essentially comparing the readings character-by-character, so it is entitled to our confidence–assuming as I cautioned earlier that there are no mistakes in the electronic input and no bugs in the application. Some technicalities could affect a decision, but the only subjective element, as I just implied, would be the level of agreement that is judged as the minimum to assess the texts of manuscripts as closely related.

Genealogical coherence, on the other hand, can be highly subjective. It is an evaluation of all the variations where there are discrepancies among the witnesses (texts) being compared. It seems to be based entirely on a set of “internal” criteria, i.e. criteria for determining the original reading from contextual (author-generated) and transcriptional (scribe-generated) factors. The exclusive use of these criteria has been called “thoroughgoing eclecticism” (previously discussed) and other names. They stand, not in opposition, but in contrast to “external” criteria, which are primarily the age and provenance (original geographic location) of a manuscript.

We have discussed all the criteria earlier, and we noted that there are two internal criteria that in a sense summarize all the others, like the two “great commandments” that summarize all the rest. Given alternative readings, that one which best accounts for the creation of the other(s) most likely is the original, and the harder(-est) reading most likely is the original.

Taking the hypothetical A-text discussed above as a reconstruction of the NT Greek text following the generally accepted internal criteria, every variant reading in the text of a manuscript is compared with it, and if the two agree, then the text of the manuscript is assumed to be the original or at least the “initial” text. When variants disagree both with the A-text and with each other, it is usually possible to reverse-engineer the scribal thinking that led to each variant by comparison with the others, and in these cases the decision is made as to which of two texts features the reading that was the source or “ancestor” of the other. These are all ultimately human judgment calls and are always subject to reevaluation. Indeed, the A-text itself is always subject to revision as more information comes to light, or as decisions on individual “variation units,” i.e. places where there are variant readings, are reevaluated.

This process is carried out for every variant reading common to witnesses (the texts of manuscripts) that are being compared. The results are input into the database, after which the witnesses are rated on the basis of genealogical coherence as queried by the user. That is, one can go online and have the computer output lists of ancestors and descendants for any witness in the database. Logically, one might think that the strongest genealogical coherence as an ancestor should be assigned to the witness that has the highest number of prior (“ancestor”) readings, and witnesses with lower numbers would then occupy lower places in the list.

If only it were that simple. The problem of contamination immediately becomes apparent in any of these lists, as we discover that virtually every potential ancestor has not only prior readings but also posterior (“descendant”) readings relevant to the witness being compared. It is as if the hypothetical process that I described above, in which a scribe creates a new copy by “correcting” some of the readings in his exemplar to match the readings of the exemplar’s ancestor, were actually the norm. Whatever the circumstances may have been, the outcome was the same: judging by internal criteria, manuscripts have contaminated or mixed texts relative to other manuscripts. A skeptic might argue that this problem simply points to flaws in the internal criteria or the application of them, but I am convinced that the criteria really do make sense.

In the CBGM, where very little seems simplistic, Mink (in my opinion) decided on the most basic standard possible to establish the ancestor-descendant relationship between witnesses: given any two, the one with at least one more prior reading than the other is the ancestor. It would seem that finding substantially more prior readings would lend more credence in identifying the ancestor. As Mink points out, however, this has an inherent drawback in that to have more of these readings, a witness also must have fewer agreements in variation units with the witness of comparison, which weakens their pre-genealogical coherence. If that is weak, then what are perceived as genealogical relationships between readings becomes doubtful. In the overall method, this means that agreements in variant readings between witnesses that are not related are most likely just coincidences.

In practice, the tables of potential ancestors reveal that the percentage of agreement in variation units is what establishes how close a given ancestor witness is to the descendant being compared. The basis for this evaluation is the fourth assumption about scribes, i.e. that they tended to use closely-related texts. Since the level of agreement is inversely proportional to the number of passages where one reading is the ancestor of its counterpart in the other witness, a greater or lesser majority of prior readings only affects the stability of the ancestor-descendant relationship. For example, if one of two witnesses being compared has only one more prior reading than the other witness–and a difference of only one or two more does happen–then the relationship is very unstable because if the editors who decided which readings were prior changed their minds about even a single reading at some future time, then the relationship will either be reversed or become undecided. How likely is this? As I will note below, we already have examples from comparing versions 1 and 2 of the CBGM, and the iterative nature of the revision processes is a given. One can only predict that at some point in the future, the processes will have been repeated to the point that further revisions will seem very unlikely.

When we examine the online potential ancestor tables, we find that the witnesses are all ranked not only by their positions in the list, which strictly follow the percentages of agreement, but also by rank numbers. The numbers are in order from one at the top (most closely related) to higher numbers running down the list, but with zero’s here and there for some witnesses. These are witnesses presenting exactly the same number of prior and posterior readings as the witness to which they are compared, and by this simple standard neither can (for the time being) be designated as a potential ancestor of the other. There are also cases of two (or more) witnesses with the same rank number when the percentages of agreement are the same or nearly so.

You may very well wonder what the real-world meaning of this evaluation is. It goes back to the assumption that the scribe wants to copy his exemplar with fidelity–unless he is also using another manuscript and taking some readings from it. Even then, we can expect to see very close resemblances between the copy and the main exemplar. The greatest deviations most likely would occur if the scribe copied a large block of text from the other manuscript, in which case the copy would closely resemble that text.

So then, the final product is a copy that should be very close in content to the exemplar. When we find two witnesses that agree with each other in variation units (and of course the remaining text agrees) in the high 90% range, as often happens, then by CBGM reasoning, we are probably looking at two texts that were on a scribe’s desk: the exemplar and its copy. As the percentage decreases, we are probably looking at a witness a few scribes or more removed from the ancestor to which it is being compared. The ranking in the potential ancestor’s tables represents this, though not in the strictest sense. Mink points out numerous times that in the genealogies of older witnesses, in particular, there are large gaps because so many manuscripts have been lost. The loss of these manuscripts obviously is beyond dispute, so the existence of large gaps is always available as an explanation for phenomena.

Connectivity vs. External Criteria

The ranking of witnesses as potential ancestors is one part of a very important concept in the CBGM that is the backbone of stemmata and textual flows in the method: connectivity. This is not inferred lightly, however; reasonable explanations must be sought for such agreements. Usually, or often, there is more than one instance of a coincidental agreement.

I think that the process of using internal criteria to determine genealogical relationships would strike most people as common sense once they understood it, but as it is undeniably subjective, students who are familiar with the more traditional use of both internal and external criteria (a.k.a. reasoned eclecticism) will wonder why the external (date, provenance, manuscript quality) are being excluded. One of the main problems with dating a text is that it is not the same as dating a manuscript, which can often be done with reasonable accuracy. To use a technical term, the manuscript’s date gives us only the terminus ante quem for the text, i.e. the date before which the text found in the manuscript must have originally been compiled. Every NT manuscript is of course based on an earlier or a contemporary copy, and the text that the later manuscript contains may, as Mink and others emphasize, be much earlier than the manuscript itself. Codex B (Vaticanus) is a famous example: originally thought by some to be a fourth-century recension (edited compilation), the discovery of P⁷⁵ (‘P’ for “papyrus”) proved that B’s text was at least early third-century. The more recent discovery of minuscule tenth-century manuscript 1739, containing a text often close to B, is another example.

Another major problem with dating is the fact that so many early manuscripts have not survived, which we already noted. As Mink observes,

It is impossible to trace the real filiation of surviving manuscripts in a contaminated tradition whose witnesses have largely been lost. All we can realistically aim at is finding genealogical structures in relationships between preserved witnesses, that is, between texts as transmitted by manuscripts, not between manuscripts as historical artifacts.[11]

So as thankful as we are for the early manuscripts like B (to which I will mainly refer as 03), we have no way to back-trace them to the autograph simply from what we know of the manuscripts (as opposed to their texts). Of course, Mink also takes an optimistic view of finding the genealogical structures that not everyone may share.

As for provenance, the geographical names given to so-called families or “types” of manuscripts at best suggested not that they were copied in those regions but were used there. However, modern scholarship has established that the different texts cannot be assigned with any confidence to geographical regions. For example, the so-called “Western” text linked to Italy, etc. is a “mixed” text with eastern elements as well, and “Alexandrian” readings are also found in non-Alexandrian manuscripts. The fact of such mixture, or contamination (to use Mink’s term), can be seen in the online CBGM tables and charts.

The quality of a given manuscript probably is not entirely discounted even with the CBGM, because one will often see kind words spoken or written of 03 and a few other manuscripts. The difference, however, between this appraisal and the traditional view is that in CBGM practice these manuscripts are valued only to the extent that they are dominated by readings favored by internal criteria, as reflected in the CBGM A-text. It certainly does not hurt that 03 is a fourth-century manuscript supported by a papyrus that predates it 150 years, but this is why minuscule 1739–which might be dismissed by traditional textual critics due to its late date–is also in the spotlight. If we suddenly discovered that Textus Receptus proponents were right and 03 (and all other “Alexandrian” manuscripts) were hopelessly corrupt, then the date and apparent quality of the hand that wrote 03 would suddenly mean nothing. As a matter of fact, many textual critics are of the opinion that the vast majority of changes to the original NT text took place by the end of the second century, so copies made in the third to fourth century were, at least in their view, vulnerable to most of the changes we find today. This could explain why we rarely do find an “easier” reading in a manuscript–such as 03–that normally displays the “harder” readings.

I noted above that proponents of the CBGM probably would protest a characterization basing it on “thoroughgoing eclecticism.” I should elaborate. Mink has stated that the editors of the ECM2, who include himself and Klaus Wachtel, are committed to the principles of reasoned eclecticism.[12] In most contexts, this would mean that external evidence, particularly the dating and quality of individual manuscripts, would have about equal weight with internal evidence, and often decisive weight depending on the editor’s philosophy of criticism. However, Mink, citing in particular 03, says that the “best known, highly esteemed” witnesses must be treated with caution. He notes that the ECM editors in practice actually treated 03 with skepticism. Nevertheless, when they examined 03 using CGBM tools and consulted it in constructing local stemmata, they judged it an outstanding witness and found it to agree with their own A-text more than any other witness.[13]

So essentially, Mink and his colleagues would never have taken the traditional approach of ruling in favor of superior external evidence over internal when the two are in conflict, nor would they allow traditional external criteria to favor even such a manuscript as 03 over other manuscripts with significantly lesser traditional credentials. Klaus Wachtel provides an enlightening example in his contribution “Towards a Redefinition of External Criteria” which, as its title implies, argues for genealogical coherence as a new, superior external criterion. The example is the problem of “Jesus” vs. “Lord” in Jude 5. This is one of my personal favorites because the choice of “Lord” found in NA27 and previously is based on an appeal to the exception provision attached to the harder reading criterion, i.e. that it is a reading so difficult as to be impossible. A plausible explanation of the assumed error is required (in my opinion at least), and the editors of the NA text suggested that the kappa of “Lord” in the nomen sacrum abbreviation apparently was mistaken for the iota of “Jesus.”[14] This is plausible because it entails a credible misreading of only one letter instead of several.

Wachtel reports, however, that the ECM editors instead adopted the minority position of Metzger and Allen Wikgren, who cited both superior manuscript support (including 03) and refrained from ruling out “Jesus” as an impossible reading, thereby making it preferable as the harder reading.[15] This certainly appears to be a decision consistent with reasoned eclecticism and traditional TC. Indeed, either reading could be justified by traditional criteria. However, Wachtel prefaces the analysis that he goes on to provide by saying that it was the coherence of the “Jesus” attestation that finalized their choice.[16]

The best way to understand Wachtel’s conclusion about coherence, and the process employing the CBGM, is to walk through diagrams provided by the application. Wachtel used version 1 of the application, but unless otherwise noted I will provide the results of version 2, which happen to be virtually the same (that is not always the case). The first diagram presented by Wachtel is the coherence of the reading “Jesus” at low connectivity (to be explained later):

We keep in mind that “A” is the A-text, the hypothetical text constructed by the ECM editors in which all the variant readings have been chosen by them. The numbers refer to the texts contained in the extant manuscripts designated by those numbers. This is a subtle distinction that will probably annoy some readers, but it is crucial for the CBGM, essentially eliminating historical dating of manuscripts as a factor in the genealogical evaluations. Identifying the texts by their manuscripts becomes strictly a matter of convenience, though one might protest that it is also somewhat inconvenient because it fosters the confusion of the two.

The slashes following witnesses 93 and 6 precede numbers that designate how far they are removed genealogically from A. In the case of 93, for example, A ranks as its fifth potential ancestor. The four closer ancestors of 93 all have the reading “Lord” with (in one case without) the article and are consequently outside the box in their own attestation. Witness 1501 probably looks out of place; its formatting indicates that it has the same reading as A, but does not have any of the other witnesses within the box as its ancestor at this level of connectivity. The CBGM is programmed to connect it to its closest ancestor even though that ancestor has a different reading, and this is indicated by the dotted arrow. I ran the query at the “absolute” level of connectivity (499) and confirmed Wachtel’s assessment that 1501 would inherit the reading from 665 only with 665 ranking as ancestor 12, which seems very unlikely. All of its closer ancestors have a different reading. The diagram above at lower connectivity suggests that the scribe of witness 1501 more likely miscopied witness 424, resulting in what Wachtel calls “spontaneous emergence” of the variant rather than “inheritance” of it from another witness.[17] In real-world terms as I see it, this is saying that the scribe of 1501 more likely did not have access to 665, or if he did, he ignored it in favor of other manuscripts at hand. If we check the Potential Ancestors table for 1501 just for Jude, we find that 665 ranks as only ancestor 91 with 83.6% agreement in variant passages, as opposed to 90% for the closest ancestor. Of course, we do not know if the scribe would have found 665 sufficiently defective in Jude to discard it.

The main point about the diagram is the high degree of coherence among the witnesses for this reading. Except for 1501, 93, 6, and 665, all of the witnesses are direct descendants of the hypothetical A or of the primary descendants 1739 and 81. But we can also see how this translates as a substitute for the external criteria of age and manuscript quality. If we accept the working hypothesis of A as the autograph or the initial text–however one may define that–then the first descendants of A become the logical equivalents of the oldest and most reliable extant manuscripts supporting the reading. I will say more about that shortly.

Wachtel also provides a diagram of the easier reading “Lord” as follows (CBGM version 1):

My apologies for the greatly reduced size, which is necessary to view the essential layout. This also leads me to recommend that the reader consult the online CBGM tools for a much better view, and to do that I must provide a brief excursus on using the tools.

Excursus: Finding a Reading in the CBGM

I am delighted to point out first that with the online publication of Acts in the CBGM tools the editors have provided a new interface which is point-and-click, and completely eliminates the need for a print copy of the ECM2 to conveniently access the tools. One accesses the main tools for Acts at Phase 4 of the Genealogical Queries, “Coherence and Textual Flow” (the first option). The direct URL is http://ntg.cceh.uni-koeln.de/acts/ph4/.

Unfortunately, if one is researching the General (Catholic) Epistles, the original, laborious code is still used that is provided in the expensive printed volume of the ECM2. Words have even numbers, while spaces are odd. A potential space is counted as ‘1’ before the first word in a verse, and the first word is counted as ‘2’.  The next space is ‘3’, the next word ‘4’, and so on.  Spaces are noted for words that are rejected for the text but found in other witnesses (superscript ‘T’ symbol in NA).

If you access one of the tools for the Catholic Letters, e.g. “Coherence in Attestations” at http://intf.uni-muenster.de/cbgm2/Coh1_4.html, you will have to fill in boxes to specify “first word or space” and “last word or space,” three boxes each. The first two are just chapter and verse, but the last is the code number for word(s) or space identifying the variant reading in the verse. If the variant you are researching is just single word or a reading not chosen for the text, then the code number will be the same for both sets of boxes. However, if the variant spans several words in NA28, the third box will have an even number followed by a higher even number in the sixth box.[18]$start$

You see one witness inside the box at the top left half, i.e. 617, from which all but one of the other witnesses branch off. Like 1501 previously, the remaining witness, 945, turns out to be eighth descendant to a secondary witness and its reading is assumed to be more likely an accidental or deliberate change from its immediate ancestor. What we mainly note, with Wachtel, is that 617 essentially changes the readings found in its two closest ancestors 468 (first ancestor) and A (second). We already know that A has “Jesus”; 468 has “Lord” and 617 and its descendants have “the Lord.” Additional checking reveals that 468 has A as its first ancestor. So the reading “the Lord” is very coherent beginning with 617, but it begins with a deliberate or accidental change from “Jesus” to “Lord” (the change of one letter in the nomen sacrum abbreviation). There is good reason to think that “Jesus” is the harder reading since the context would be the activity of the pre-incarnate Christ. That, together with the better coherence of the reading in “A”, led Wachtel to conclude that it is the initial reading and led to “Lord,” first without (468), and then with (617), the article (“the Lord”).[19]

Let me return now to my point about the first descendants of A in the diagram above. If we take the four that have A as their first ancestor, i.e. 1739, 81, 02, and 03, it appears from the diagram that they each have equal value. For many, 03 by itself with the harder reading probably would be sufficient evidence to accept the reading, and it helps to have the additional support of A/02 (Alexandrian readings outside the gospels). However, 1739 and 81 are tenth- and eleventh-century manuscripts respectively. They are probably too late to have value in the eyes of traditional students of textual criticism, though some might be surprised to learn that they feature mostly Alexandrian readings. However, should we ignore their late date and take the texts they carry as valid, early witnesses to the autograph? Why should we not instead assess them as late, flawed copies of fourth-century or later manuscripts and reject them as insignificant factors in our decisions?

As far as the CBGM is concerned, i.e. as a matter of its programming, these late manuscripts are just as important as the early ones because the texts (witnesses) that they carry test out as direct descendants of A. As a matter of the philosophy behind the CBGM, the dating of a manuscript establishes nothing more than the terminus ante quem for the witness it carries, as we noted earlier.

The dating issue aside, it may also surprise the reader to discover the degree to which the witnesses display a mixed text in terms of the old text-type classifications. This comes out in the tables of Potential Ancestors or Descendants. It is easy to view 1739 as an Alexandrian witness by CBGM criteria because it has only two ancestors: A[20] and 03 in that order. We are tempted–at least I am–to rule out A because of its hypothetical composition and simply assign 03 as the one and only ancestor of 1739. However, the Ancestors table for 1739 shows a significantly lower level of agreement in variant passages with 03 than with A and a relatively large number where it appears that the variant in the text of 1739 is prior to that of 03. I question the decisions of the ECM editors in those places since A is their own subjective compilation of the initial text, but I am inclined to think that I would agree with them more often than not, and the CBGM provides a way to confirm or refute that as I will explain below.

Witness 81 represents the problem of a mixed (or contaminated) text much more clearly and is more typical of witnesses than 1739 is. It has A and 1739 (together with 35) as its first and second ancestors for the General Epistles, yet we find Byzantine witnesses, including 617, as ancestors occupying several slots ahead of 03. You may wonder if the problem is not confined to late manuscripts. I wish I could say that it is, but then we have the case of 01 (Sinaiticus). Its potential ancestors start with A and 03 as one would expect. The problem is that the percentage of agreement in variant passages is only 90% for A, and perhaps more importantly only 87% for 03–more importantly because it is a real manuscript of undisputed quality. This percentage is much lower than we would expect, slightly less than average, and it reflects all those troubling passages where the two manuscripts are divided. For the unhappy traditionalist, the CBGM offers the unattractive prospect that 01 might be the product of several sources, some of which theoretically are found only in late manuscripts.

Indeed, we find data in the CBGM that is at times puzzling and conflicting, driving us back toward crude but simpler historical dating of manuscripts. It happens that 81 ranks as third potential ancestor to 01 for the Epistles, just below 03, but when we query its ranking for Jude alone, 81 jumps to the second position (A is first in both rankings), while 03 with a smaller percentage of agreement ranks only twelfth. Even more striking, 03 is known among CBGM users for having only A as its ancestor for the Epistles as a group. Yet when its ancestry for Jude is queried, it has three potential ancestors in addition to A, with 81 occupying the second place! Thus for Jude, 81 is a leading ancestor for both 03 and 01, and the CBGM is telling us that the witness (text) of 81 could have been used by the scribes of 03 and 01–based on the fact that the ECM editors have credited two more prior readings to 81 as compared to 03, and twelve more as compared to 01. In that case, it would necessarily have been available to the scribes in a much earlier manuscript, of course, the content of which was eventually copied to manuscript 81. Is this plausible? A skeptic of the CBGM would say that it is easier to imagine the scribe of 81 copying from 03 (or a reliable copy of it) to account for the high percentage of agreement (93% in Jude), and other earlier sources for the variations.

Fortunately, we are not left to simply trusting the editors on their conclusions about the all-important prior readings. Mink and his colleagues have made it relatively convenient to evaluate the data and conclusions through other online CBGM tools. One can run a comparison of any two witnesses, and then query the application for the actual passages and readings by clicking on “View Differences.” The variants are then listed, and one can examine each passage, along with its local stemma to see thereby something of the reasoning of the editors. Our appraisal of the CBGM clearly is helped by the fact that in Jude, 81 (and two other witnesses) tests out as a potential ancestor to 03, a scenario that is true only for that epistle and 2 Peter (where 665 is also an ancestor with one additional prior passage[21]). In the other epistles, 03 has only A as an ancestor, as we would expect. I should emphasize that the concept of a “potential ancestor” does not mean that the witness posited as such was actually used by the scribe of the descendant. Even when the direction of inheritance is stable, there may be too great a gap in the percentage of agreement, or too many intervening witnesses to support a realistic possibility of inheritance. For 03 in particular, inheritance for the prior readings actually goes directly to the hypothetical A, not to 81. Furthermore, the list of potential ancestors does not exclude any whose readings are all duplicated by others on the list.[22]

What is there in the data we are given that makes the CBGM model viable against the simpler one of a scribe producing a copy that is merely a descendant of earlier manuscripts according to historical dating, and not their potential ancestor as portrayed by the CBGM? Probably the best answer would be the fact that prior readings usually or often are the harder readings, though sometimes only marginally so. I can believe that a scribe would be inclined to change a harder reading to an easier one, as opposed to the reverse, though I think we have to be willing to entertain the possibility of the latter. Indeed, we know now to expect mixed texts. However, when there are differences in variant passages, I expect to see the easier (posterior) readings dominate in the descendant. If I were to encounter the opposite, i.e. repeated changes to harder readings, I would think that the scribe most likely used a different exemplar with a higher percentage of agreement, including the passages that were changed in what was thought at first to be the primary exemplar.

If we are dealing with two witnesses that share the highest percentage of agreement in variant passages, then by CBGM reasoning our only logical and practical option is to reverse the genealogical order between them if what we thought was the descendant proves to exceed the exemplar in harder (but acceptable[23]) readings. Thus the descendant becomes the new immediate ancestor since no other witnesses are extant with equal or greater agreement. In terms of manuscripts, this means that if the ancestor is much later (as is 81 compared to 03), then all early copies of its witness are lost or undiscovered, as would certainly be possible. Of course, this does not eliminate the problem; the new descendant now has the (harder) prior readings where it previously had the posterior readings. As I noted earlier,[24] in working through all this Mink and his colleagues came to the conclusion that a simple majority of prior readings dictates the direction of inheritance. The pesky prior readings in the descendant then have to be explained by contamination from other witnesses or “spontaneous or accidental emergence” (plausible reasons other than contamination).

 So then, we can examine the eight prior readings attested in Jude for 81 relative to 03 in our quest for enlightenment. One fact that soon becomes evident is that each of the eight is shared both by 01 and 02. One can confirm this either by consulting the Coherence in Attestations module of the CBGM or the ECM2, keeping in mind that a negative apparatus in the latter will not list 01 or 02, but for these passages includes them.[25] Thus we know at the outset that the scribe of manuscript 81 or a late exemplar of it might possibly have had access to an early manuscript containing the prior reading in question. Lacking more information, we might even have assumed, without doing adequate research, that 81 essentially was a copy of 01. Given statistics from the CBGM, however, we know that 81 is significantly different from 01 in Jude. It exhibits a relatively low 87.6% agreement in variation units and actually ranks as an ancestor to 01, with 16 prior readings against four posterior. Moreover, as we already noted, 81 is a very mixed text except in Jude, where it agrees with the hypothetical A-text at 96% and with 03 at 93%. In fact, for all 16 prior readings, it sides with 03 against 01. It also agrees with 02 at an even higher 94%, but the latter has significantly more posterior readings, marking it, like 01, as the descendant (rank five).

If we accept the theory that higher agreement indicates a closer genealogical relationship, then it would be more reasonable to assume from a traditionalist point of view that 81 essentially is a copy of 03 eclectically corrected against 01. But this leads us to the question of why the scribe of 81 would choose the eight prior readings in 01, 02, etc. over those found in 03 unless the ECM editors are mistaken, and the eight readings are not necessarily prior. If they are not mistaken, it follows either that the scribe–like a modern textual critic–was capable of rationalizing and choosing a more difficult reading over an easier one, or that the ECM editors are also correct about the transmission history, and the witness (text) of 81 is an ancestor to that of 03. Two potential ancestors of 03 in Jude, 623 and 33, have the eight prior readings as well with only one exception, but they have 81 as their first potential ancestor, so this is unremarkable.[26]

The CBGM application also tells us that 03 has six prior readings relative to 81 in Jude. Now since the value of the difference between prior and posterior readings is treated as relevant only to the stability of the genealogical relationship, while the percentage of agreement for variation units determines the genealogical ranking, the A-text is the first-ranked potential ancestor to 03 (almost 95% agreement), and 81 is second (93%). Witness 81 has only the A-text as its potential ancestor, at nearly 96% agreement. So, the six prior readings, all a-readings, are unremarkable, 03 agreeing with its first ancestor against 81, its second. They are also compatible with the traditionalist view, taking 03 to be the original reading.

For the eight posterior readings, we have the scribe of the text of 03 abandoning the potential ancestors A and 81, so classified by virtue of the fact that they contain two more prior than posterior readings. If I view this as a traditionalist, I can once again connect 01 with all eight prior readings and just consider them disagreements between the two manuscripts, and I would probably ignore 81. We are left, however, with the problem of 81 as described above. To appraise the situation, we must evaluate the prior readings case-by-case. As far as the CBGM is concerned, we would need only to reverse the inheritance of two readings in order to change 81 from an ancestor to another descendant of 03. However, I find it odd that 81 would have any clearly prior readings at all relative to 03, given the principle of the harder reading. Indeed, the decision by Mink and his colleagues to draw the line at a simple majority struck me as being unnecessarily cautious. So I opted to examine all eight.

The first three variation units are found in verse 9, beginning with the opening phrase of Jude 9, O DE MIXAHL (“But Michael”). In 03 the verse begins instead with OTE MEIXAHL (“When Michael”), the alternate spelling of the name being of no significance. What is interesting is that the adverb “when” (OTE) occurs at the beginning instead of at the middle of the first line in 01, 81, etc. It would be very jarring to find it repeated there in 03, which instead has “then/at that time” (literally “at that time disputing with the devil”) differing in Greek spelling only by the addition of a letter at the beginning of the word. That is the second reading in question. There is no indication in the manuscript of any correction as an afterthought by the scribe of 03, so either his exemplar already read this way, or he introduced changes to his exemplar as he originally penned his manuscript. For convenience, I will assume the latter.

The Greek words in question at the beginning of the verse sound very similar and could easily be mistaken under the right circumstances, so they were vulnerable to change. From my viewpoint, the reading of 03 can be considered more difficult than 81/01. It is abrupt, beginning without a connective, and the addition of “then/at that time” seems awkward and unnecessary. It would be helpful to know for certain what led the ECM editors to their decision. We do at least have the local stemma from the CBGM modules, and from it we know that the editors concluded that the reading of 03 was inherited directly from A, the text found in actual witnesses 81/01, etc. It is also worth noting that the editors agreed with the editors of the NA text, whose decisions have always been the same for this variant and all but one of the remaining variation units we are examining.

To avoid overthinking the problem, I assume that the editors in both camps viewed the reading of OTE in 03 as a simple mistake which our scribe, instead of correcting, compensated for by changing the OTE later in the verse to TOTE (“then”). As I suggested above, the reading of 03 can be considered the more difficult, but it can also be argued that it is both too difficult and that the reading of 81/01 better accounts for it than vice-versa. There is, moreover, the fact that 03 is alone in its reading, though I imagine that any scribe who had access to both readings would have preferred 81/01. So I would conclude that one could conceivably rate the decision as unclear, but I can understand the decision for 81/01 as prior.

The third reading in question is the status of a preposition before “you” in “rebuke you” at the end of verse 9. I should note that the preposition is all but illegible in the manuscript of 03, but what remains matches the preposition, and it is found in 1739, an important manuscript that usually agrees with 03. This reading is easier to evaluate because it is a Hebraism from Zech. 3:2 that came across as such into the LXX, which the NT writers routinely used. I think we can assume that the ECM (and NA) editors concluded that the original author quoted the angel’s exclamation from memory, without the preposition, and that a scribe later added the preposition to match the wording of the LXX. I find it a little surprising, however, that scribes did not change the position of “Lord” in the Greek to match the LXX as well if they were consulting it. Conceivably the author remembered the Hebraism and included it, but did not recall the position of “Lord,” or had a reason to move it. If so, a scribe (or scribes, if several made the change independently) could have omitted the preposition as a mistake without considering that it could have been taken from the LXX by the original author. Without consulting the LXX, he would have had no reason to question the position of “Lord.”

I would grant that scribes were more likely to “correct” quotations from the OT to their LXX wording than to “improve on” translation Greek from the LXX. Nevertheless, this is something we can only consider a tendency, so one might rate this decision as “unclear” as well.

This brings us to verse 12 and the fourth variation unit. This time we encounter a grammatical issue–an error actually. In 03 the Greek participle translated “carried along” is in the wrong gender (masculine) to go with “clouds” (feminine), and the explanation seems to be that a previous adjective referring back to “clouds” has an ending that looks masculine, but in reality, is dual-gender. The mistake is historically older than manuscript 03 because it is also found in the papyrus P⁷², where a scribe corrected the gender to feminine.

Ruling on this reading could be difficult for an especially conservative theologian faced with the challenge of making a textual decision. It is a virtual certainty that at some early stage a competent scribe would have corrected the masculine ending to feminine, while to do the reverse would be a major blunder. So was the original author inspired to make a grammatical mistake? Many if not most theologians would probably say that divine inspiration does not extend to proper grammar. For what it is worth, I do not believe that the Holy Spirit inspired even grammatical mistakes. I do believe that OT and NT writers were sometimes inspired to “break the rules” for special effects, such as to emphasize something. In the present case, however, I do not see any special effect or implication from the use of the incorrect gender. Moreover, I do not think this theological aspect of the problem can be treated objectively, much less quantified, as we would like.

Still, there is something to be said for the correct ending as the original. It seems far less likely that the original author would be distracted by his own writing than a scribe copying him would. That is, there is no reason why the author would think of the wrong form and actually use it, but we can see how a scribe copying the exemplar word by word could get confused and make the mistake. However, is this a case of a genealogically relevant, posterior reading in 03? No. As a blunder, we can expect it to “emerge spontaneously” from a number of different witnesses with the “a” reading to 03 and other b-reading witnesses, and this is what we see in the attestation for the b-reading at average connectivity:

We might keep in mind that by CBGM standards, it would require only two reversals of genealogical direction among the eight variation units to turn witness 81 from a potential ancestor of 03 to a descendant of it, as traditionalists would basically view it.

The fifth variation unit is found in verse 13; the variation is merely an alternate spelling of the word for “wandering,” either being legitimate words. We could assume that a scribe might have used whichever happened to be familiar to him without giving it much thought. However, this is another case where 03 is alone in its reading. Also, the a-reading has virtually perfect coherence. We might have expected many more spontaneous instances of the alternate reading. So the ECM and NA editors alike probably concluded (and I would agree) that the scribe of 03 momentarily veered from his exemplar in penning what we find here, and that the reading of 01 etc. is prior. But is this a case of genealogical inheritance? I think it is uncertain. The scribe might have corrected what he perceived as an error to the spelling with which he was familiar (creating a posterior reading), or perhaps he mechanically wrote the form he knew without giving the change any thought.

The sixth variation unit, also in verse 13, is the presence or absence of articles with “black” (or “gloom”) and “darkness.” Reading “a” has both articles, p⁷² and another witness, 1844, have the article only with “darkness,” and 03 has neither article. Unless it is translation Greek for a Hebraism,[27] I think the reading of P⁷² is the hardest and would expect the other two to represent corrections. The phrase occurs once elsewhere, in 2 Peter 2:17 where both articles are present. Since Jude and 2 Peter has a good deal in common, the stemma in the CBGM indicates that the ECM editors deduced the readings of P⁷² and 03 to be scribal errors for reading “a,” which is also found in 2 Pet. 2:17. We find again that 03 is alone in its reading, while the a-reading has perfect coherence through several generations. I think it is hard to account for the reading in 03, even as a blunder, though I would guess that was the view taken by the ECM and NA editors. If we make it the initial text in the CBGM modules, then the coherence of the genealogical relationship between its reading and that of 01, etc. is not very good (as we would expect), but it could be argued that a scribe would be inclined to “correct” the reading of 03, which appears rough–especially if he had 2 Peter before him.

This brings us to the seventh variation unit, which is more substantial. There are four different versions of the verb “keep” in verse 21, the a-reading being a second-person imperative. It makes good sense, following the second-person pronoun that opens verse 20. Strangely, 03 has instead the first-person plural subjunctive, used as a hortatory “let us keep ourselves.” The online stemma shows this c-reading (no “b” is listed because it was assigned to a defective reading) as derived directly and solely from “a,” and then the e-reading, a first-person plural future, as derived from “c.” A “d” reading, a nominative plural participle, is also shown as derived from “a”:

 

1. a)  τηρησατε
2. c)  τηρησωμεν
3. d)  τηρησαντες
4. e)  τηρησομεν

I find this stemma very odd. Of the four readings, “a” would seem to be the easiest for the context. The most difficult, bordering on impossible, would be “e,” and “c” next to that. Indeed, I can conceive of “e” as an error for “c,” and in that respect, I would agree with the stemma. If “e” was an error of the ear, then “c” might possibly have been the original reading, a seemingly awkward shift from the second plural (verse 20) to the first (hortatory subjunctive) intended to mitigate the force of a direct imperative. I, of course, concede that this is not the most credible of explanations; any scribe encountering such a change of person would have been inclined to change the verb to some form making it compatible with verse 20. Reading “d,” not as natural a fit as “a,” would have worked, the assumption being that one of the participles was functioning as an imperative. It is hard to imagine what might have led a scribe to change “a” to “c,” or even to “d.”

As I have noted, one of the features of the CBGM modules is the option to set a different initial reading. One cannot obtain a truly objective outcome because the initial conclusions of the editors and the existing genealogical relationships cannot be altered. But the results can be interesting. If we set “c” as the initial reading here, we obtain the following from the Coherence at Variant Passages module:

The module, unfortunately, will not supply us with a different stemma, but the resulting textual flow is closer to what I would expect. This is especially true if we eliminate or reverse the arrow from 617 (reading “a”) to 1845 (“c”). It turns out that if we test just for Jude and not all the epistles, 617 actually becomes a distant descendant of 1845, ranking only at 41 with 85.6% agreement. So then, if we can somehow accept reading “c” as difficult but not impossible, it can account for “e,” and “a” is a correction of “c.” This, of course, is another case where we would like to know the reasoning of the ECM (and NA) editors. Once again, to avoid overthinking it, I assume that they viewed reading “c” as a blunder (i.e. “impossible”) and therefore posterior. Nevertheless, I would like to know how they would reconstruct the process that resulted in the blunder. As far as I can determine from clear images of 01, 03, and P⁷², there is nothing perceptible to the eye (or ear) to account for it.

The eighth and final variation unit occurs in verse 23. It is long and complex, covering the first half of the verse with over a dozen variant readings if we follow the ECM2. In this case, thankfully, we are assisted by commentary from the late Bruce Metzger on the discussions of the UBS Greek NT committee, who chose the same reading as the ECM editors. Furthermore, our good fortune continues as we discover that both committees most likely came to the same conclusion for the same reason. This will become evident from the CBGM stemma:

1. a)  ους δε σωζετε εκ πυρος αρπαζοντες ους δε ελεατε εν φοβω
2. b)  ους δε σωζετε εκ πυρος αρπαζοντες τους δε ελεειτε εν φοβω
3. c)  ους δε εν φοβω σωζετε εκ πυρος αρπαζοντες ους δε ελεατε εν φοβω
4. d)  ους δε εν φοβω εκ πυρος αρπαζοντες σωζετε ους δε ελεειτε εν φοβω
5. e)  ους δε εν φοβω σωζετε εκ πυρος αρπαζοντες ους δε ελεατε εν φοβω θεου
6. g)  ους δε σωζετε εκ πυρος αρπαζοντες ους δε ελεγχετε εν φοβω
7. h)  ους δε εν φοβω σωζετε εκ πυρος αρπαζοντες ους δε ελεγχετε εν φοβω
8. i)  σωζετε εκ πυρος αρπαζοντες ους δε ελεατε εν φοβω
9. j)  ους δε εν φοβω σωζετε εκ πυρος αρπαζοντες
10. k)  ους δε εν φοβω σωζεσθε εκ τοινος αρπαζοντες
11. l)  ους δε σωζετε εκ πυρος αρπαζοντες εν φοβω
12. m)  ους δε σωζοντες εκ πυρος αρπαζοντες εν φοβω
13. n)  ους δε σωζετε εκ πυρος αρπαζοντες εν ζοφω
14. o)  ους δε εν φοβω σωζετε εκ του πυρος αρπαζοντες
15. p)  ους δε σωζετε εκ πυρος αρπαζοντες

I note with great relief for the reader that we need only concern ourselves with witness “i,” which is the reading of 03. As we can easily see above, the ECM editors decided that the i-reading was inherited directly from “a.” It will be helpful to quote Metzger’s comments at length (I will merely omit the Greek citation):

The singular reading of B . . . (“and those, whom you pity when they contend [or doubt], save and snatch from the fire, but some pity in fear”), can scarcely be correct, for it involves, as Hort admits, “the incongruity that the first [Greek term for “some”] must be taken as a relative, and the first [Greek verb] as indicative.” It is probable that the scribe of B accidentally omitted [Greek for “others”] before [Greek for “save”], in which case his archetype would have agreed with the text preserved in [list of manuscripts including 01].[28]

Please note that the first brackets (“doubt”) were Metzger’s; the rest are mine, to mark Greek words and special characters in the quotation.

Even if you do not know a single Greek letter, a careful look at the readings listed above will reveal that each one begins with “ους δε” except for i, the reading of 03. What Metzger and Hort (whom Metzger quoted) both noted was an awkward imbalance in the structure of the Greek in verses 22-23, which form a unit. Once again we are faced with the harder vs. impossible reading issue: either 03 supplies the original reading, i.e. the initial text for all the others, which the scribes of the other witnesses “corrected,” or the scribe of 03 committed a blunder in copying the manuscript containing witness A. Metzger and the other editors of the UBS text decided that the scribe committed the blunder, which in this case was of such a kind that nothing else had to be changed to compensate for it. The same decision was made for the 26^th edition of the NA text. Previous to that, the reading of 03 was actually chosen for the text, and I suspect that the reading was considered nearly impossible but not quite, rendering it the “harder” and therefore preferred reading, though no doubt reluctantly in this case. An extraordinarily high regard for 03 surely played a large role, and a preference for the shorter reading may have played a role as well. I note that for the other seven variation units, the NA/UBS editors have always agreed on the readings, with which the ECM editors later agreed. Nevertheless, the change of reading in the NA/UBS texts, in this case, is proof of the fine line that can exist between the very difficult and the “impossible.”

I also want to draw the reader’s attention to the last few words of Metzger’s comment above. In the list of manuscripts with 01 he includes 81 (and 33), so what he says about the archetype for 03 is coincidentally equivalent to the CBGM ranking of 81 as a potential ancestor of 03, at least for this variant. And if I may suggest a logical extrapolation, I suspect that Metzger assumed the same archetype for the other seven variants, whatever the reasons the scribe of 03 had for deviating from it. Metzger seems to be entirely in harmony with Mink and his colleagues when he speaks of the “text preserved” in the manuscripts.

I do not know what Metzger would have said in summary about 81 if he also were considering the nearly equal times that it displays posterior readings relative to 03, as reported by the CBGM. In going over the eight “prior” readings, I think we have found that whether there are really any indisputable genealogically prior readings in 81/01 etc. is a least theoretically a judgment call, though of course I do acknowledge the fact that for many years the UBS/NA editors have in effect confirmed seven of the eight decisions made by the ECM editors, and ultimately (after the 25^th edition) all eight. That fact alone is undoubtedly sufficient to convince many translators and students of textual criticism that the decisions for 81–or more importantly for the traditionalist, for 01–in these passages must be correct.

The Choice of a Harder Reading

In that case, I think that there is a basic text-critical problem in this process for which I have as yet been unable to find a satisfying answer: the unavoidable conclusion that in a number of passages–probably a significant number–the scribe must consciously have chosen a harder prior reading over an easier posterior reading. I am not referring to clear blunders, but to cases in which experts can disagree over what is impossible and what is very difficult, but still plausible. To put it another way, modern text-critical editors sometimes choose the harder readings for “a” (and the NA text) over those found in 03,[29] but the harder reading principle predicts the opposite for a scribe. Moreover, as I see it, neither the traditionalist nor the CBGM model ultimately escapes the problem. Whether we are thinking of the scribe of a late manuscript copying early extant manuscripts, or the scribe of a very early text (witness) found in a late manuscript (CBGM model), the challenge for the scribe(s) remains the same if there is a choice of variant readings, with the harder and the easier.

One can argue that the CBGM model avoids this for 81 because as a potential ancestor of 03, we expect it to contain harder readings that the scribe of the text of 03 changes. The problem, however, is the point of view. If we take that of 03’s scribe, still within the CBGM model, then the scribe (of the text of 03) chose six prior readings relative to those of 81, all a-readings as was the case for 81’s eight prior readings. If we scrutinize these readings in the way we did the other eight, we will arrive at similar conclusions, and most of the choices have again been supported by the NA editors. Moreover, in every case the only ancestor to account for the preferred readings with a higher priority than 81 is the hypothetical A. If we rule it out as a real witness, we are left with only 03 itself (none of its potential ancestors) for four of the six readings.[30] Of course the same is true for 81, which has only A as a potential ancestor for Jude.

Mink has a way to address and represent this situation graphically, using what he calls an “intermediary node.” He presents the case of an exemplar and two witnesses, one of which has more prior variants relative to the other but also has some posterior variants relative to it, a situation that he maintains is true of virtually any two witnesses within a highly contaminated textual tradition.[31] The following diagram is supplied:[32]

In the diagrams, B is the exemplar and witness C is a potential ancestor of D by virtue of the fact that it has a significantly larger number of prior readings compared to D (a fact not represented in the diagrams). The problem is that C has a handful of readings that are posterior to their counterparts in D. The diagrams relate only to these readings.

Mink points out that the left diagram is inadequate because it does not account for the few posterior readings in C. However, an arrow (the statistical term “edge” is preferred by Mink) pointing from D to C would be misleading since it would give the impression that D as a witness is predominantly prior to C when the opposite is true in this simulation. The box with “C < D” is the intermediate node, and it represents the statement that there are “posterior variants in C deriving from prior ones which are found in the non-ancestor D and which evolved from variants in B”.[33]

Substituting our Jude witnesses, B would be the primary ancestor A of both 03 and 81. Granting that 81 has two more prior readings relative to 03 according to CBGM determinations, C would be 81 and D would be 03. Then the diagram to the right in the figure above (figure 30 in Mink) would represent the six prior readings of 03 relative to 81. Essentially, as I understand it, the diagram states that the six readings in 03 have found their way there after “evolving” (Mink’s word) from A, while the corresponding posterior readings in 81 derived from these same prior readings at some intermediate stage, as opposed to deriving them from 03. This is an abstraction of what actually happened, which is unknowable. Thus the intermediary node seems to serve as a kind of placeholder, possibly to be replaced with more information in the future.

Mink sees this as a partial solution to the problem of conflicting (prior/posterior variants) data since structurally it provides a way for these conflicts to be acknowledged and represented without the prevailing genealogical direction between witnesses being falsified.[34] This strikes me as overly optimistic, however, because we still do not seem to have made any progress toward an explanation of why a scribe would choose harder (prior) readings over easier ones.

Another discussion by Mink is more promising in revealing how he views the problem. He notes that scribes evidently tend to make changes to their exemplars by choosing readings in the ancestors of the exemplars. This follows from his second principle, already discussed above, and he further notes that scribes tended to choose readings from the closest potential ancestors of their exemplars when they made changes.[35] For an example, we can view diagrams he provides in his article Problems (p. 50):

The diagrams are very illustrative of how witnesses and the scribal process are handled in the CBGM. The model featured is that of a scribe working with two manuscripts, B and D, in such a way that the outcome (E) forms a logically[36] circular process. The scribe basically copies D against its ancestor B, more or less deliberately returning changes from C back to their original form in B to the extent that E has more prior readings relative to C. The point of the diagrams is that it must appear sequentially or historically that C is an ancestor of E, but since the scribe has changed C-variants found in D to what he finds in B, E has thereby become a potential ancestor of C (as indicated by the dotted line or “edge”). Since this is just an illustration of the phenomenon, Mink does not discuss how the scribe manages to distinguish C-readings or the processes that lead to his decisions. We could hypothesize that the scribe also changes some D-variants of B back to their original B form, but not enough to make E a potential ancestor of D as well.

More importantly for our purpose, this illustration happens to touch on the case of a scribe clearly correcting what he finds in his exemplar to the readings of its ancestor, which must include at least some preferences for harder readings. In this model, 03 would correspond to E, 81 to D, and A again to B. To be consistent with the CBGM we must eliminate C because 81 has only A as a potential ancestor in Jude.[37] Nevertheless, what Mink says about C’s handling of B can be applied equally to D in a theoretical model without C. The merging of A and 81, as of D and B in the model, is not relevant for us but undoubtedly did take place according to CBGM figures (189 EQ passages for A vs. 187 for 81[38]). What matters is that in our D, i.e. 81, there are variants of A that the scribe of 03 changed back to what they were in A, six of which we find in Jude.

Excluding blunders, the scribe presumably had reasons for all these changes, but I infer that Mink is somewhat reluctant to discuss possible reasons. The clearest statement that I have seen from him on the topic is the following:

[A] copyist did not always follow his exemplar but sometimes preferred variants from another source. He may have preferred them for various reasons–perhaps because he considered the variant more apt linguistically or logically, perhaps because he had more confidence in the other source, or perhaps simply because his main exemplar was damaged or illegible at a passage. We do not have positive knowledge about this.

It is certainly most unlikely that a copyist consistently used several exemplars.[39]

I included the closing line that opens another paragraph because with this statement Mink eliminates another possible reason for choosing a different source, i.e. that a scribe would routinely veer away from his exemplar. The line before that about “positive knowledge” is also very important. One could call it a kind of excuse for not saying more about the problem, but aside from giving Mink the benefit of the doubt, it is clear that he and his colleagues view speculation as an enemy.

When Mink poses the possibility that the scribe found the preferred variant “more apt linguistically or logically,” he seems to be saying that it was a choice in the scribe’s mind of the easier reading. However, I would argue that the suggestion that a scribe may have had more confidence in another source he was consulting is key to the issue. A scribe might choose a harder reading if he had more confidence in its source than in his primary exemplar, at least for the context at hand. Of course, Mink might disagree, and he certainly would disagree with anything he considered an impossible reading, regardless of the source.

Nevertheless, I think we now have what we need to flesh out the important harder reading criterion that we previously discussed. As we have seen, the CBGM dictates that scribes sometimes consulted the ancestors of their exemplars. We know, too, that the CBGM assumes that related witnesses will have both prior and posterior readings relative to each other, yet one will be an ancestor of the other unless they are equal in these readings–at least as the CBGM application is designed and written. It seems to follow, then, that scribes did not always change harder readings to easier ones when they made changes reflecting the difficulty of readings. In some cases, scribes adopted the more difficult reading found in another source, perhaps because they had more faith in the source, or perhaps because they used the same reasoning modern textual critics do in choosing the reading most likely to explain other variants.

Does this mean that we should jettison the criterion? Not at all. When we find readings that differ significantly in difficulty, it still follows logically that a scribe at some point changed the harder reading to the easier rather than the reverse, barring blunders. However, as with virtually all other criteria, exceptions must be allowed. Especially in a witness that contains harder readings in other variation units, we probably should not rule out the deliberate choice of the harder reading in any given unit. At least this seems to be true for the General Epistles. If we must grant, as seems clear, that witnesses contain both prior and posterior readings, then in my opinion it follows that we must also grant the occasional choice of the harder reading by a scribe, early or late.

The CBGM and the Traditionalist

In any case, granting the presence of prior readings in a descendant witness, I think we encounter an important implication: the possibility of a traditionalist interpretation of CBGM findings for those who want to take that position, depending on the witnesses under study. If I continue with 81 as an example, we know from CBGM data that it always joins 03 when it has the “a” reading against 01 in Jude. Likewise, it joins 01 when it has the “a” reading against 03 in Jude. All other readings relative to 03 and 01 are “unclear,” “no relation,” or derived from 03 (sometimes 03 and 01). So instead of accepting the proposition that 81 is a potential ancestor of 03 for Jude, I think a traditionalist can maintain that the scribe of witness 81 could have copied from 03 and 01, correcting one against the other and infrequently choosing other readings derived from them. In all the other epistles 81 is a descendant of 03 by CBGM calculation.

Of course, this raises the question of whether the same phenomena occur between 81 and 01 for the other epistles. Conceivably the scribe of 81 could have had witness 03 for all the epistles and have lacked 01 for some or most of the others, in which case he would have needed another early source (or sources), from the traditional viewpoint, to account for differences with 03. Witness 02 could provide this to some extent, but it would benefit the traditional position if 81 joins 03 for prior readings in opposition to those of 01 in any epistle for which 81 tests out as the potential ancestor to 01. A comparison of the two witnesses in the CBGM shows the latter relationship for 1 and 2 Peter and 1 John. The genealogical direction is also very stable for the Petrine epistles, especially 1 Peter with 66 prior readings for 81 vs. 35 for 01.[40]

It happens that 81 does join 03 in these prior readings for 1 Peter 58 times, and given the inherently subjective nature of determining “a” readings, this is a very substantial figure. The figure for 2 Peter is 26 of 30 times, and for 1 John, 81 joins 03 an impressive 47 out of 49 times. Unfortunately, at this point, we must wait until data is analyzed and disseminated for the other books found in 81 (not to mention the great bulk of work yet to be done for all the extant manuscripts in the CBGM project). It is possible that very different outcomes will be discovered; but that seems unlikely, given these figures. Therefore I will try to show how the CBGM can be of value even for the traditionalist as I go through a few additional examples below.

Issues for Genealogical Direction

Given the complications that Mink identifies, it is perhaps no wonder that he and his colleagues settled on such a simple criterion as a bare majority of prior readings to establish genealogical direction between witnesses. I give them credit for bringing to full light the reality of mixed texts in quantifiable terms. It seems that what we have had prior to the CBGM are basic descriptions of manuscripts that broadly qualify them as relating to one family or another, or as “mixed” at worse. But I suspect that this new mathematical criterion for genealogical direction raises three potential issues in the minds of those evaluating the CBGM: 1) it seems questionable at best that a scribe would be conscious of such a small difference between witnesses nearly equal in prior and posterior readings, so the main exemplar probably cannot be identified; 2) the genealogical relationship between some witnesses will be unstable, and 3) the subjective division of some variation units may further contribute to the instability of relationships. Are these issues of consequence for the CBGM user?

I do not think that the first issue is a problem. I certainly reject the possibility that a scribe with two or more exemplars would have spent time sorting out which had the greater number of “prior” readings relative to the “posterior” before deciding on a primary exemplar, the first “ancestor” of his copy. However, what we are actually referring to is the product of all the choices made by the scribe, and not even the scribe himself would likely have known which manuscript proved to be the true “ancestor” until his work was done, if then. I suspect that any given scribe went through a process similar to the initial task of the ECM editors: work through a manuscript (the physical carrier of a witness) with or without some preconceived valuation of it, and come to a final, informed appraisal of the manuscript as a whole based on the variations in it.

I do think that the second issue is a problem. The fact that 81 has genealogical priority to 03 in Jude due to the former’s two additional prior readings is evidence. One can make the case that the preponderance of the evidence, including the agreement of the NA editors (if we count additional expert opinion as a positive criterion), favors the choices made in 81 and the A-text, and I do not necessarily disagree. Nevertheless, if we abide by the CBGM standard, we would need to make only two changes to reverse the genealogical direction, and I think I have at least pointed out that two or more of the decisions could be reversed. Indeed, the direction would be reversed even if three of the decisions were simply changed to “unclear.”

I have examined other witnesses in the CBGM and have found not only that it can take very little to change the genealogical relationship, but that a significant reordering already took place with the appearance of version 2 of the computer application. Various texts that were ancestors of other texts in version 1 became descendants of the same texts in version 2 (henceforth v. 1 and v. 2), and vice-versa.[41]

The third issue is closely related to the second, and as it happens, the variations in Jude 23, the eighth variation unit discussed above, provide an example. In the fourth edition of the UBS GNT, which agrees with NA26/27, the longer readings that were composed of two clauses were divided into two separate variant readings each, probably for convenience. The structure of the Greek and the actual reason(s) for this division are not important for our purpose. What matters is that the a-reading above (found in 01 and 02), which was chosen for the text then and has continued to be the preferred reading in the ECM2/NA28, could have been counted as two prior readings if the GNT had been followed.

The first two variation units that we examined in verse 9 are also interesting in regard to the third issue. You will recall that I suggested that the readings of 03 were probably treated as secondary because the first was viewed as a mistake, i.e. a blunder, and the second as intended to compensate for it. What is relevant for us now is that, while the two readings are counted separately, there are notes in the ECM2 connecting them, indicating to the reader that the scribe probably was influenced by his first change to make the second. I would suggest that this is another choice that could have turned out either way, i.e. the two readings might have been treated instead as one. In that case, there would have been one less prior reading relative to 03. Therefore, it seems that whenever any two related witnesses have very few prior readings establishing one as the potential ancestor of the other, there is cause for concern about the genealogical direction.

There is also the odd case of the “no genealogical direction” relationship.[42] Whenever pairs of witnesses are equal in prior and posterior readings, they do not count as ancestors or descendants even though they may have sufficiently high levels both of agreements and disagreements to rival any other pair of related witnesses that might only differ by one or two prior readings. For example, witness 436 has a 90% level of agreement with 03, together with eight prior readings and eight posterior readings. As a result, it shows up both in the list of potential ancestors and the list of potential descendants, with a ranking of “0” in both. The physical position of each witness in these lists depends on the percentage of agreement, so 436 at 90% is in seventh/last place in the ancestor list to 03, and it is in tenth place among well over 100 in the descendants list. If just one decision about any variant unit in 436 were changed in any way, it would then have a rank of five in one list or the other.

Unfortunately, I do not have a practical way to search for a situation in which two witnesses with a high level of agreement but no genealogical direction relate when one seems to be dependent on the other for its reading (high “connectivity” based on internal criteria).[43] However, without a ranking, no connection would seem possible because a zero ranking would appear prohibitive unless the application has a special routine to handle the problem.[44] It does seem clear that early on in the planning process Mink and his colleagues envisioned the existence of witnesses in high agreement (coherence) with each other but without genealogical direction, that could remain in that condition indefinitely. Mink discusses the possibility in his Problems, positing the inclusion of “intermediary nodes” to cover agreements between such witnesses. However, he concludes the discussion there by noting the necessity of bearing in mind that the stability of the textual flow in undirected coherencies has a value of zero. This is based on a simple formula developed by Mink and his colleagues, in which the value depends on the difference between the numbers of prior and posterior variants in witnesses.[45]

The problem I see in these cases is that any witness with a high percentage of agreement (coherence) but no genealogical direction would presumably be just as likely to be used by a scribe as another high-coherence witness with direction. It seems plausible, also, that such a non-directional witness might be necessary to account for a particular reading in a witness for which it is listed as a potential ancestor. Yet the application does not appear to have a way to handle this.[46] On the optimistic side, we can probably assume that non-directional witnesses are seldom needed for the purpose.

The main objective of the CBGM is, of course, to deal with the contamination found in NT witnesses. The process, again, begins with the identification of pre-genealogical and genealogical coherence among the witnesses, which we have already discussed. It bears repeating that with the exception of agreements in pre-genealogical evaluation, the rankings depend on judgments made by internal criteria. A high degree of skepticism would be justified if these initial judgments were limited to a sampling of variation units as has been done in other methods, but in the CBGM every variation unit in the database has been evaluated. So the ranking is not an inference or product of general induction, though it does have problems as we have already seen, and will discuss further below.

The identification of potential ancestor relationships among witnesses essentially is the CBGM solution to the problem of contamination, and it also replaces external criteria in the method. Whatever claims may be made about the application, the online tools base genealogies of readings–which lead in turn to genealogies of witnesses–on the potential ancestor data. There is no detectable algorithm or routine for employing the age of a manuscript as a criterion, its appraised quality (other than its percentage of agreement with the hypothetical A-text), or any geographical data. Traditional textual families are also ignored in the calculations, though there is some recognition of them at a very basic level. The A-text happens to line up closely with the Alexandrian family. In version 2 of the application, a witness’s agreement with the Majority Text is stated as a statistic in the header for potential ancestors, but it does not factor into genealogical calculations. Above all, there is the classic recognition of the Byzantine text in the ECM2, where it is treated as a subset of the MT. The latter is the text of readings found in the majority of all manuscripts whether they support the A-text or not. The Byzantine text is redefined as the text containing the variant readings which are both found in the MT and differ from the (hypothetical) A-text.[47] Yet in practice, one can find Byzantine witnesses listed as supporting the A-text. So what is going on?

The CBGM “Byz” text

To sort out the concept of the Byzantine text according to the CBGM, we begin with some background. We have already discussed the text as an inferior text-type earlier in the book, and it is typically viewed positively or negatively in opposition the Alexandrian manuscripts. Given that Gerd Mink and the other ECM editors have jettisoned traditional views and arguments about text-types in general, the Byzantine included, they have taken a clean-slate approach to the latter, and their finding is that witnesses exhibiting this form of “text” took on great uniformity after the ninth century.[48] So there is statistical validity to a Byzantine text at that point in time due to uniformity, whereas before then we are probably to assume that the tradition was too fluid to establish the text.

The editors put together collections of Byzantine witnesses of the General Epistles from this group of manuscripts, with the main criterion that the readings disagree with those of their own (hypothetical) A-text. This would seem to go a long way toward preserving the traditional opposition between readings preferred by standard text-critical criteria and Byzantine readings, and perhaps that was the goal. Otherwise, the editors could have had much the same outcome by comparing the Byzantines against Alexandrian witnesses. By making their own A-text the standard, one would think that the editors ensured that opposition between Byzantine readings and the A-text was a certainty. Designating the Byz-reading in each case was then just a matter of identifying the one with the majority (or greatest?) manuscript support, which would mostly coincide with the traditional selections.

However, this breaks down when individual witnesses (as found in manuscripts) are examined. It turns out that even the most “Byzantine” of the witnesses extant agree with non-Byzantine witnesses now and then. That is, if we isolated all the “Byzantine” readings from their witnesses and created a hypothetical text (like the A-text) containing them all, then that text would behave as expected. But no extant Byzantine text will cooperate to that extent, and as a result, a look at the ECM2 reveals cases of Byzantine witnesses supporting a-readings.[49] Therefore it appears that we have to consider the CBGM definition of “Byzantine” (i.e. disagreeing with the A-text) as actually valid for readings rather than for any single (continuous) extant witness. This is probably how we are to understand the description as “the form of text defined by those readings which…differ from the established text.”[50]

Another important element in the CBGM for the Byzantine text is the concept of a divided or undivided witness. For each of the General Epistles, a subset of seven witnesses/manuscripts has been selected by the editors to best represent the Byzantine text for that epistle. This is similar to another CBGM practice of ultimately thinning out all sources not judged strictly necessary in establishing genealogies for witnesses, done for convenience and efficiency but at some risk of accuracy (in my opinion).[51]

The seven selected witnesses are described as being “nearly pure Byzantine manuscripts which rarely depart from the group,”[52] i.e. rarely disagreeing with each other. They do nevertheless disagree sometimes, and–depending on the epistle[53]–if two or more have a different reading from the group, the Byzantine witness is considered “divided” and the unqualified symbol Byz is withheld from every reading, even one that has some (limited) Byzantine support.[54] However, such a reading will have a form of the symbol with a suffix to identify one or more Byzantine manuscripts that do support it.

To be a genuine CBGM Byzantine reading, the editors have ruled that the reading under consideration must have the support of six (or for James, five) out of the seven select manuscripts. It follows that the total number of these readings will be significantly fewer than if, for example, a reading were considered “Byzantine” under less rigid standards, and I think one can view this situation in various ways.

The ECM editors present the results in a very positive light for the Byzantine text. The observation is reported that in all clear choices of the A-text for James, the Byzantine text differs from it in only 61 of more than 700 variation units. It is added, quite logically, that research on the Byzantine text of James must proceed mainly from the 61 readings. Then an interesting appraisal is made: “Thus it [the Byz text] is, apart from these 61 passages, an important witness to the early text.”[55] That, in my opinion, is putting a very high polish on the lemon, given that it has just been conceded that the 61 readings are the heart of the Byzantine text of James. Virtually the same thing is said for the other epistles.

Moreover, assuming of course that the “early text” is the A-text, the Byzantine text has been defined as contributing nothing to it that MT witnesses do not already contribute. It is as if the MT were being treated categorically as a subset of the Byzantine text rather than the reverse.[56] That is, the ECM editors are telling us that the Byzantine text is a combination of readings, most of which (otherwise known as MT readings) agree with the early text and some of which do not, which is about equivalent to the definition for the MT stated in ECM2.[57]

I am not certain as to what the small subsets of seven Byzantine witnesses are meant to do, but it seems clear that they provide an objective means for categorizing other Byzantine witnesses that is consistent with the ways things are done in the CBGM, which is a kind of lowest-common-denominator agreement. For the CBGM, classification methods must be mathematically practical and objective in the evaluation process, allowing the use of computers. One can quibble with the rules chosen by the editors, but under those rules I dare say that the subsets of witnesses probably are not difficult to choose, and identifying Byzantine readings can likely be done with little or no human input, minimizing the subjective element by reducing decisions to their most basic levels.[58]

When we consider the outcomes, however, I see problems instead of the conclusions made by the editors. One problem I already alluded to is the minimizing of Byzantine readings. A comparison of the list of Byz readings for each epistle with the readings found in, for example, the Robinson-Pierpont GNT reveals that a significant number of readings preferred in the latter against NA27 do not rise to the level of being fully “Byzantine” in the CBGM, yet these readings differ from the A-text and have majority support. Relaxing the rules categorizing the readings would cast a wider net and yield a more distinctively Byzantine text, while somewhat diminishing the claim of its being “an important witness to the early text.”

Another problem for this construction of the Byzantine text is the extent of evident mixture, i.e. contamination. Even when we look at a historical stage of the text when it has supposedly been preserved in a “consistently and carefully controlled form,”[59] the ECM editors have perceived a need for further purification, and they weeded out non-Byzantine and some partial Byzantine readings, using their own criteria. Then they presumably identified seven manuscripts for each epistle with the highest inclusion of the true Byz readings. Yet even with a minimum number of readings, no manuscript emerges as a perfect representative of the readings.

Indeed, two manuscripts that were sufficiently Byzantine to be included in all of the subgroups–18 and 35–support the A-text reading most of the time (as implicitly acknowledged by the ECM editors), joined by most other Byz manuscripts but opposed by others. Examples are too numerous to mention at any length; I will give the reader a general idea by listing some of the simpler readings, however. I begin with James: manuscripts 18 and 35 side with the a-reading aÓpo\ in 1:17 (over a dozen Byz mss support another reading); ti÷ß in 3:13 (over a dozen for another reading, five for a third); and aÓfeqh/setai in 5:15 (three support the plural reading). Then in 1 Peter: 18, 35, and other Byz manuscripts support the a-reading uJma◊ß in 1:4 (four support the first person); uJpo\ in 2:4; ca¿ritoß zwhvß in 3:7 (too many support other readings to establish a “Byz” reading); kalu/ptei in 4:8 (another case where there is too much division for a Byz reading, and RP[60] disagrees); and uJma◊ß in 5:10 (five manuscripts have the first person). I will mention just two cases of 18, 35, and other Byz manuscripts supporting a-readings in 2 Peter: e˚autoi√ß in 2:1 (two manuscripts have two other readings respectively), and qeouv in 3:12 (three manuscripts support another reading, and two another).

For the Johannine epistles, we find all seven Byz witnesses selected for the ECM2 (including 18 and 35) supporting the inclusion of de« (a-reading) in 1 John 1:3. All seven support aÓggeli÷a in 3:11 (two Byz mss support another reading). Six of the seven support the a-reading oJmologh/sh in 4:15. For 2 John I will just note as an example that the seven witnesses support the a-readings of the first-person pronoun in verse 2, and in 3 John 3 the reading ga»r supported by all seven can be cited, among others. In Jude, to cite only one example, the a-reading omitting lo/gwn in verse 15 is supported by the seven witnesses among multiple variants.

Some readers might feel that this is not an entirely fruitful analysis since it is comparing the ECM committee’s own version of the Byzantine text with their own version of the initial text, and therefore the results may only be relevant to the ECM. In response, I note first that the readings I cited are all the same as in NA26/27, reflecting a traditional preference for Alexandrian witnesses. As for the Byzantine text, I can offer two sources for comparison: the GNT produced by Robinson and the late William G. Pierpont,[61] and the Byzantine Greek New Testament (hereafter “BGNT“), compiled and published by the Center for the Study and Preservation of the Majority Text. Both are available freely online.[62] Unfortunately neither text appears to be supplied with a critical apparatus at this point in time. Robinson and Pierpoint direct us in their preface to several existing apparatuses, and it happens that the most recent of these is the ECM first edition.[63] BGNT is more informative in that we are supplied lists of the collated Byzantine (K^r) manuscripts that account for the base text. We are told that the collation was done mainly by Wilbur Pickering, and the listing is book by book.[64]

I have compared Pickering’s collation with the ECM, and there is no overlap except for mss 18 and 35. This is due probably to certain preferences held by Pickering and outlined in the BGNT introduction.[65] In any case, ms 35 is highly valued by Pickering and is listed for every book in his collation. Manuscript 18 is also cited for all but a few books. I therefore am confident that the combination of the two is recognized by all as the best overall representative of the Byzantine text, and that ms 35 is especially favored.

We can thus expect both the RP GNT and BGNT to agree with the sample of readings I cited above from the ECM, and that is in fact what happens, with the notable exception of kalu/ptei in 1 Pet 4:8, where RP has the future kalu/yei instead. Not surprisingly, this disagreement reflects a split in the Byzantine witnesses. Four of the seven chosen by the ECM editors have the future tense, while the remaining three–including 18 and 35–have the present.

When we consider the high valuation of mss 18 and 35, we would expect them to be in agreement for variant readings, and indeed they are on the whole. Nevertheless, there are variant units where they diverge from each other. In James, for example, I find 18 supporting the second-person reflexive pronoun in 2:8 while the text of 35 supports the third person. Just as interesting, a sigma is written above the line in 35 to show either a correction to the second person or the latter as an alternative reading. Thus it is clear that a scribe or instructor noticed the difference and considered it important. RP and BGNT both have the second person. I also thought it would be interesting to compare readings with the Textus Receptus (TR), which has the second person here as well. I note that we must assume a fundamentally different process for the “choice” of readings in the TR: in most, if not all of the cases we examine here, there probably was no text-critical choice of reading at all. Therefore, I will not be suggesting any critical deliberation behind TR readings, but only what we encounter there, which may be nothing more than a routine copying process.

That said, the variation unit in 2:8, unfortunately, was not selected for inclusion in NA28 (or earlier editions), and even when some of the examples I am citing are included, the information in the apparatus is incomplete. Consequently, convenient access is only available at present through a paper copy of the ECM2. The reader who does not have access to that but is willing to make extra effort to obtain the data can do so online by using the CBGM tools. The particular tool to access is “Coherence in Attestations,” and an online guide is available for assistance. In this particular case, the user needs version 1 of the tool and should input “2,” “8,” and “28” into the first and second set of blank spaces. For the “Variant,” “a” will provide a diagram of witnesses for the second person pronoun (with ms 18), and b will provide the third person variant (with ms 35), judged by virtually all critics as the secondary reading.

In James 2:11, 18 and 35 disagree on the form of the second occurrence of the verb moiceu/seiß (future indicative vs. aorist subjunctive).  RP, BGNT, and the TR agree with 35 on the future tense. In 3:14, 18 has the compound verb katakauca◊sqe while 35 has the simple form, with kata written above the line as a correction or alternative reading. RP, BGNT, and TR have the compound in agreement with 18.

In James 4:10 18 has the a-reading touv kuri÷ou«, while the text of 35 omits the article and it is shown as a correction or alternate reading.[66] RP, BGNT, and TR all agree with 18. Of greater interest, 35 appears to have i¶dete corrected to ei¶dete in 5:11 in agreement with 18, and RP prefer the former while BGNT and the TR have the latter which is also the a-reading and that found in 01 and 03. The seven Byz witnesses selected by the ECM editors are divided as well. The difference in meaning (imperative vs. aorist indicative) is significant.

I found several disagreements between 18 and 35 in 1 Peter, but most were problematic or of little significance for our purpose here. We find 35 supporting the reading aÓnagkastw◊ß (also RP, BGNT, and TR) in 5:2 while 18 diverges from most of the Byz witnesses, displaying aÓnagkastikw◊ß instead. In verse 10 ms 18 has the second person pronoun after katarti÷sai while 35 omits it and has a correction above the line. RP, BGNT, and the TR all favor the reading of 18.

An interesting variation unit in 2 Peter is not noted in NA28: in 2:2 the relative pronoun following di∆ has the masculine gender in 35 and other Byz witnesses, but the feminine in 18, in which case it refers to the extreme behavior rather than to the people committing it. The masculine with its superior support is chosen for NA, the a-reading in the ECM2, and RP and the TR have it as well. Yet the feminine was selected for BGNT.

Moving to the epistles of John, I first discovered 35 differing from 18 in 1 John 1:5, where 18 and other Byz witnesses read aÓggeli÷a while 35 has e˙paggeli÷a, corrected to aÓggeli÷a. RP, BGNT, and TR agree on aÓggeli÷a. Then it is 18 that departs from the majority by omitting the article before zwh\n in 2:25. RP, BGNT, and TR again follow the majority reading. In 3:15 18 sides with one of the other six main ECM witnesses in supporting the personal pronoun after e˙n while 35 and the others display the reflexive pronoun, also preferred by RP and BGNT. The TR has the personal pronoun in agreement with 18, however, which is also the ECM a-reading.

Another divergence between 18 and 35 occurs in 3:23, where the seven Byz witnesses are quite evenly split between the aorist subjunctive pisteu/swmen and the present subjunctive of the same verb. The text of 35 displays the latter and has a sigma written above the line either as a correction or an alternative reading. As is sometimes the case, the split here between 18 and 35 is paralleled by a split between the Alexandrian manuscripts 03 (with 18) and 01 (with 35). RP, BGNT, and the TR agree with 18 (aorist subjunctive), and I note that RP does not include the variant in the apparatus.

Just one verse later nearly the same thing happens, but with important differences. 35 sides with 03 this time reading kai« e˙n tou/twˆ with a correction, while 18 agrees with 01 in omitting kai«. Just as significant, or even more so, for our discussion, BGNT has the reading found in 18, while RP and TR prefer 35. However, the choice is not noted in the RP apparatus. I found no other disagreements between 18 and 35 in 1 John, nor in the remaining epistles of John.

A search of Jude revealed the omission of e˙k ghvß in 35 in comparison with the other Byz witnesses for verse 5, with a correction written above the line. The omission was judged an error by the ECM editors, however, no doubt because the resulting reading (e.g. “a people of Egypt”) would be nonsense in the context. A legitimate divergence occurs in verse 7, where 35 reads tro/pon tou/toiß in agreement with 01 and 03, while 18 agrees with most of the other Byz witnesses in transposing the two words. That is also the reading found in RP, BGNT, and the TR. The text of 35 has been corrected to that of 18 by the addition of a beta and alpha written over the two words in that order, to show that they have been “reversed.” Thus the corrector seems to have respected the fact that the discrepancy was the result of two different exemplars being copied.

Another significant divergence occurs in verse 12, where the main ECM Byz witnesses including 18 joins 01 in supporting the simpler reading without the article following ei˙sin (second word in the verse). The article originally was in 35 but was rubbed out as a correction. RP, BGNT, and TR all display the reading of 18 and 01, perhaps preferring the simpler wording.

Finally, in Jude 19, 18 and the majority of the leading Byz witnesses join the Alexandrian witnesses in having no object with the participle aÓpodiori÷zonteß, while 35 agrees with many other witnesses in displaying a reflexive pronoun as an object. As might be expected, the letters of the pronoun are all dotted to indicate doubt about their legitimacy. RP and BGNT follow witness 18; the TR agrees with 35, however.

All witnesses/manuscripts, it seems, fail to present a continuous Byzantine text in full, and depart from that text frequently. Nevertheless, the continuity of the text in extant manuscripts has been the dominant argument for the Byzantine text as maintained by Robinson.[67] One wonders how this argument can be sustained under close examination, given the data now available to us in the ECM2 (as I just illustrated) and disagreements over readings among such Byzantine advocates as Robinson and the editors of the BGNT. We would need a minimum of two significant, relatively independent manuscripts with identical readings that were all acceptable to Byzantine advocates upon which to build the argument. It might be possible for a different picture to emerge when a much larger portion of the NT is completed for the ECM, but it seems unlikely.

Furthermore, I do not see how agreement with the early text (or ECM A-text) is beneficial for the Byzantine text, as the ECM editors seem to imply. It is, of course, important to analyze and evaluate Byzantine readings that appear to be prior to variants in the older Alexandrian manuscripts. Even these readings, however, present questionable support for the Byzantine text in my mind because they run counter to the text-critical criteria favored for Byzantine readings, such as rejection of difficult readings. I think we see this problem at the root of Robinson’s reluctance to accept the harder-reading criterion without major qualifications.[68] If these readings are corrupt (from that perspective), it follows that they must have been introduced into Byzantine witnesses by scribal blunders, unless no simpler option was viable. Conversely, when Alexandrian witnesses display a reading that is, for example, easier than others, this poses a problem because it is out of character with those witnesses. We do not congratulate the scribe(s) responsible for the reading for having finally chosen wisely. Instead, we treat it as an early corruption, for want of a better explanation.

The transmissional history of the NT text that comes to my mind in relation to the Byzantine text is complex and muddy in detail, but if I may be allowed to oversimplify, I see the original text being edited in no organized fashion during the early centuries, to solve perceived difficulties. In the absence of organization and coordinated direction, the scribes were inconsistent in what they considered acceptable, and this could account for the overlap of the Byzantine text with readings that we mainly find in the Alexandrian witnesses. Indeed, we see the same inconsistency in modern editions of the GNT to a lesser extent.

Advocates of the Byzantine text can, of course, reverse this overlap and view the Alexandrian witnesses as sharing original readings with the Byzantine text, with corruptions mainly resulting when the Alexandrians are over-edited and conflict with Byzantine readings. So one must decide what criteria and sets of corollaries to follow or prioritize.

Using the CBGM When Evidence Is Divided

One of the gold nuggets that conservative textual critics and translators have given the public over many years is that we have every word of the New Testament existing somewhere in the thousands of copies of ancient Greek manuscripts extant today. We have also assured the public that we have the means to determine the original reading, and as a classically-trained scholar, I can maintain that there is no need to make conjectures, or educated guesses, as to the correct reading in NT passages in comparison to classical Greek literature, where sometimes less than a dozen manuscripts of a work have survived. According to J. K. Elliott, thoroughgoing eclectics take the same position,[69] so it does not necessitate a particular TC philosophy or methodology.

I continue to hold to these assertions, but I have to admit that there are passages where the choice of the original reading is very difficult. Among the worst, for a traditional conservative textual critic at least, are those where the “best” manuscripts are divided, and none of the alternatives is persuasive as the original by internal criteria. The fact that such passages exist has always left me with an uneasy feeling, and that was before Mink and others drew attention to the weaknesses in traditional textual criticism. So, in these passages, I find readings supported by combinations of manuscripts that provide me no objective grounds for a decision, and little else to go on.

Then there are also those rare passages like 1 Pet. 4:16 where the harder reading is supported by the inferior manuscripts against the best manuscripts, which should not happen in traditional TC. It seems that the only explanation for these troubling anomalies is the contamination or mixture of texts by scribes. If that were not enough, it is also a fact that a large number of potentially important manuscripts have not been taken into account in past decisions by editors of the Greek text. Translators have worked with what they had, relying on the NA apparatus and commentaries for information about alternate readings.

While it is true, as already noted, that decisions on the readings are originally made before CBGM operations start, identifying genealogies of readings theoretically are another way of determining the original text. Provided that the problems with the CBGM can be managed, this genealogical method potentially is another useful tool even for the traditionalist. It is something like version-tracking the hard way. With a modern word processor, all I have to do to find the original is to locate version one of the document. I do not need to back-trace the flow of thought through the different versions to identify the original. If, however, I forget to turn version-tracking on, I have a task more like that of the CBGM. Eventually, using the computer and internal criteria, I work through all the versions of the NT, back-tracing the flow of thought, and I develop a genealogy of witnesses (the texts of individual manuscripts). If the genealogy makes sense, then the text identified as the original ancestor most likely has the original reading, whether it is a choice that could be made by logical means or not. The choice of reading always needs to be reasonable, of course. As I indicated above, however, there are situations where there is not enough information for the decision to be comfortable by traditional methods, let alone firm.

I will offer four examples, not only following CBGM principles but also attempting to apply a traditional perspective where applicable. The first example is supplied by Mink in his “Contamination” chapter. It is an excellent choice for our purposes because it is difficult, due to our having too little textual information. It also involves a difference in meaning, though not in theology. The verse is Jude 15, and the question is whether Jude wrote “convict every soul” or “convict all the ungodly” or some other construction with “ungodly.” For all the details I direct the reader to Mink’s chapter.[70]

This is one of those “Oh-no” passages for a traditional textual critic, one who would now be described as a user of “reasoned eclecticism” (like it or not), because he first notices that the Alexandrian witnesses 03 and 01 are divided: 03 has “all the ungodly,” while 01, together with the third/fourth century papyrus P⁷⁵, reads “every soul.” The fact that 01 is joined by the early papyrus is very significant since we now know that the former is characterized by singular (unique) readings. Not only is this not a singular reading; the papyrus establishes it as an old one by traditional dating.

When we consider transcriptional issues, the Greek spellings of the two phrases are completely different, so there was no simple “slip of the pen” or oversight that would explain the differences. Whatever led to the change was a deliberate decision by the scribe or scribes responsible for the original change (“scribes,” that is, if the reading arose more than once coincidentally). There are a number of other complications because Jude is quoting from the non-canonical Book of Enoch, and his citation does not match the extant Enoch text; in comparison to it he does some paraphrasing. As a result, while the background of the Enoch text provides possible clues, at the end it leaves just as many puzzling questions. So in attempting to back-trace Jude’s thinking, it is very difficult, if not impossible, to determine whether “every soul” was Jude’s original reading or he instead wrote “all the ungodly” (or one of several versions of it), which was changed to “every soul” by later scribes. Internal criteria seem indecisive, and I would have to confess that the external evidence is of no help in view of the split between the Alexandrian witnesses.

As it turns out, this is not one of the 34 variation units where NA28 differs from NA27; however, for those who follow the history of the NA text, NA27/26 did part company here with NA25 (1963), for which the editors preferred the reading “all the ungodly” found in 03. Most translators have since opted for that reading as well, taking 03 as superior to 01 when forced to choose between the two. From the traditional viewpoint that is sound, though certainly not infallible, reasoning. I note that the latest version of the NIV has merely “all of them,” revealing that the translators actually refused to make a choice. Really?

So what can the CBGM contribute to this conundrum? For reasons that will become clear as you read on, anyone who uses it should first compare the computer output of version 2 with that of version 1. Version 2 is the latest published “iteration” of the method, and it incorporates many new and modified decisions in the calculations. Note that Mink’s publications were based on v. 1.

The overall options and conclusions for any variation unit are found in the “Local Stemmata” module, and you go there to get your bearings. This is what we find for our reading according to version 1 at intf.uni-muenster.de/cbgm/LocStem1.php:

1. a)  πασαν ψυχην
2. b)  παντας τους ασεβεις
3. c)  παντας τους ασεβεις αυτων
4. d)  παντας ασεβεις
5. e)  τους ασεβεις

First, a note of explanation: Jude has only one chapter, but in the application a chapter number is mandatory. Thus, we want “1:15.” You will probably find “14-16” confusing because it looks like it should refer to the words in order, but clearly, there is a mismatch. Following the practice of the ECM, the authors of the application numbered all the words using even numbers, and number the spaces using odd numbers. Odd numbers essentially represent places where alternative readings are treated as insertions. What can be confusing is finding the correct combination to designate the reading you want. The variation you have in mind may not line up with the variation unit recognized by the editors of the ECM2. In fact, Mink has said that one needs to have the ECM2 in order to find the variation units. Now, the ECM2 is supposed to be available online, but as of this point in time, it is not. I have found that with some trial and error, you can determine the correct combination to retrieve the information you want. Consult NA28 and be careful to follow the markings in the text, then count off the words by two’s for substitute alternative readings, or spaces by odd numbers for alternatives that are insertions.

So to continue, I retrieved the stemma for version 2 and got the same results as version 1. Whatever changes were made to the database leading to version 2 did not affect the choice for the initial text here. I could ignore all the other data for this unit from v. 1, but it will be useful in revealing problems or issues in the CBGM.

As we look at the diagram and accompanying data, we see that reading a is “every soul,” and reading b is “all the ungodly.” The other three readings are variations that can be explained as being derived from reading b, and this is evident from the diagram. What this stemma tells us, then, is that the editors who have constructed the “initial text” A (which is always represented as default reading a) had concluded as best they could for version 1 that “every soul” most likely accounted for all the other readings, beginning with b. Again, none of the many modifications or additions to the database leading to version 2 affected their initial decision.

Keep in mind that what we hope to obtain from the CBGM is a genealogy of this reading, because it is another way of identifying the original text. To find the genealogy, we are looking for what Mink and Klaus Wachtel call “connectivity” between readings. It is of two types: 1) the reading in question is found in two manuscripts that are closely related to each other, or 2) the manuscripts may not be closely related, but the nature or quality of the reading in one manuscript is such that it seems to depend on its counterpart in the other manuscript. The best case of the first type is that of all witnesses of a reading being connected to their highest-ranking (rank number one) potential ancestors; this is considered perfect genealogical coherence. It is the primary focus of the computer application.

The online module “Coherence in Attestations” shows us the genealogical connections between witnesses (and the manuscripts that contain them) for any reading we choose, with the additional option of choosing any reading as the initial text. For Jude 15, Mink observes (and I have verified this and all other statements) that reading b (“all the ungodly”) shows perfect coherence when b is assumed as the initial text. This sounds like a ringing endorsement for b, and we also know that it is supported by manuscript 03 and is the preferred translation in most Bibles. “Perfect coherence” means in this case that we can see the same reading repeated in a chain of manuscripts all related to their closest potential ancestors (i.e. according to the CBGM) almost without exception, going back to the original text. The problem, in this case, is the golden rule of textual criticism, i.e. that the original reading is the one the best explains or accounts for all the others. If we take b to be the original and query the CBGM for the attestation of reading a (“every soul”), this is what we obtain:

And it is not a pretty sight. The three numbers inside the box are the three manuscripts with reading a, “every soul”–note especially 01 and P⁷², while 1852 is a thirteenth-century manuscript. The dotted arrows point to the three manuscripts as descendants from witnesses that are outside the box because, as we have seen previously, they all have a different reading (as listed above in the preceding diagram). Now, notice the slashes followed by numbers. The numbers indicate how far distant each descendant witness is from its ancestor. For example, 945/10 means that 045 is tenth among potential ancestors of manuscript 1852; not closely related at all by comparison with most other witnesses. How high the maximum number and consequently how distant the relationship allowed in the results is something that you can set in the application, and the default is 10 (as indicated by “Con=10”). The fact that “b:A” has no slash means that the hypothetical initial text, which we have temporarily given b as its reading, in this case, is actually the closest potential ancestor of the three witnesses/manuscripts in the box. What we want to see primarily are close connections between real witnesses that have the same reading. Mink has pointed out that when the most closely related ancestor of a witness has a different reading, it is usually the result of contamination. Simply put, the scribe making the copy “corrected” it to a different reading not found in his original but in some other manuscript.

In evaluating the genealogical relationships, however, we also need to consider the percentage of agreements in variation units between witnesses. In the case of 945 and 1852, for example, while the latter has nine other witnesses more closely related, it agrees with 945 at about 87%, which is just about the average. P⁷² may have 945 ranked sixth as its ancestor (papyri like this are fragmentary and hard to compare), but the agreement rate is only about 82% by comparison. The application supplies convenient tables to provide this information. A low percentage of agreement indicates that there were many manuscripts in the genealogy that did not survive.

If we were to take our hypothetical b:A out of the picture, then we would have the scribes of our three extant witnesses copying manuscripts that are distant descendants of text contained either in 1739 (tenth cent.) or in 945 (eleventh cent.). Again we have to bear in mind the CBGM provision that late manuscripts can contain (very) ancient witnesses–or portions thereof. In this case, we need not consider the possibility that one manuscript might have been used by the scribes to “correct” the other because neither manuscript has reading a. Since 01 without the initial text would be left with 1739 as its only source, and 1739 also does slightly better as an ancestor to the other two witnesses, we could simplify the findings and ignore 945. Then we could say that our three witnesses all have 1739 as their only real source and a somewhat distant one at that.

This is a little misleading, however, due to the structure of the application. Lines between boxes only flow from the nearest ancestor to its nearest descendant. In the present case we should especially take note of the fact that 03 actually supports reading b, so rather than simply eliminate b:A, to make this relevant for the traditional viewpoint we should substitute b:03 for b:A, because it will do better as an ancestor for 01 and P⁷². For 01, 03 rates as second ancestor at a little over 87% agreement, and second ancestor for P⁷² at about 85%. Both percentages are significantly better than those for 1739. It is true that 03 does not relate as well to 1852, agreeing with it at about 87.8% as opposed to a solid 88% for 1739, but it is obviously close. Even if we replace 1739 with 03 as ancestor to our three witnesses, however, we still have only average agreement overall, and no agreement with them on this particular reading. This is not good genealogical coherence. What it does have in its favor is a much more palatable genealogy for traditional textual critics, who would have no problem seeing the text/witness of 03 as the ancestor of 01 and plausibly of P⁷², as well as distant ancestor of 1852.

Before we continue, I have been following Mink in stating the overall percentages for the witnesses. For example, the diagram above shows 945 as tenth ancestor to 1852 (87% agreement) for the General Epistles. For Jude, 945 becomes the twelfth ancestor at about the same level of agreement. However, while 1739 ranked as fourth ancestor at 88% agreement, for Jude alone, it ceases to be genealogically related (‘0’ rank) at 87% agreement because the two witnesses have an equal number of prior variants. By comparison, P⁷² was shown as having 945 ranked sixth as ancestor (82% agreement), and 1739 ranked fourth (82%) for the epistles overall. For Jude alone, 945 is ranked only thirty-fourth (78% agreement), and 1739 a negligible sixty-third (77%)!

As we discussed earlier, changing this one setting can drastically alter the textual flow diagrams, since they are based on the potential ancestor’s rankings. Mink has observed that the mixture of Byzantine and “non-Byzantine” elements throughout the Catholic Letters (General Epistles) is not homogenous,[71] but he does not appear to factor this into decisions about variants; and more importantly, for the present at least, the computer application only processes the overall rankings for the diagrams. I would guess that the reason for this is the mathematical benefit of having the largest possible sampling of data to determine genealogical priorities. It is the same reason why fragmentary texts are excluded by default, yet the user does have the option to include them.

To return to our Jude passage, what makes the poor coherence in the attestation even worse than it may already appear is that there are no lines connecting the three witnesses in the box. This is seen even if we set a as the initial reading, in which case we get:

This means that all three witnesses were copied independently from the initial text. It is not too difficult to imagine this for 01 and P⁷², due to their age. For a traditionalist, this might be a leap of faith for manuscript 1852 (thirteenth cent.). In favor of the CBGM analysis, one can say that the fact that we do have a few ancient papyri and manuscripts like 03 and 01 extant even today at least allows the hypothesis that a witness like 1852 was copied from another ancient manuscript that was since lost. For percentages, 1852 agrees with A at about 90.4% overall, slightly higher than 01 agrees with A and about 89% for Jude. On the other hand, if we want to postulate that the scribe of the text of 1852 copied that of 01, the application casts doubt on this by dismissing 01 even as a distant ancestor, with less than 84.6% agreement overall and less than 80.3% agreement in Jude, possibly much less for both.

The lack of genealogical connections between the three witnesses means, in the real world, that the scribes who produced them had no access to each other’s work, nor to any other shared witnesses with the a-reading. Yet they all came up with the same variant reading (“every soul”) to replace what they found in the manuscripts that they were copying if we assume b to be the initial reading. Every reading except e (“the ungodly”) gave them “all,” but how did all three of them decide on “soul”? Evaluating the results of the CBGM attestation analyses, Mink concludes that if reading b was the initial (or original), then the agreement was coincidental. There seems to have been no genealogical connectivity, nor any other form such as a word in the Book of Enoch that might have led to the change.

A coincidental agreement of three independent witnesses is difficult to accept, and with that Mink concludes that the only credible scenario to explain reading a is that it was the initial text, copied into the three witnesses. The editors of NA26/27 decided for reading a for other reasons, I suspect, perhaps because they liked the combination 01/P⁷² better than the combination for reading b despite its inclusion of 03. Given the information that we have, I think Mink makes a strong case.

Before rushing to agree with him, however, one should ask what the CBGM attestation is for reading b with a as the initial text, a question that Mink did not cover in his discussion. I show the resulting CBGM diagram below in two parts, the full diagram first, then a partial zooming in on the relevant information:

Manuscript 2774 to the far right in the first view is negligible for our purposes. What you will notice is that if reading a (“every soul”) is the initial (original) text, then virtually all other witnesses depend on 03 as the actual ancestor for their reading. The scribe who produced 03 changed reading a to reading b for his own reasons, just as any one of the three scribes would have to have done in the previous scenario with b as the initial reading, and from that point on the genealogical connectivity is not significantly better with the real potential ancestors. Witness 04 (manuscript C) is third in line to 03, which is not too bad, but then it has only 1243 as its descendant, which in turn is third in line to it. The rest of the manuscripts depend on 468, which is a fifth descendant to 03, and also only in 87.7% agreement with it (about average).

How then does this compare with the previous scenario featuring b as the initial reading? Recall that our best-case scenario has manuscript 03 as the witness with the original reading, and this, in fact, is consistent with the appraisals of traditional textual criticism. For our purposes, we have also boiled the variation unit down into two scenarios, in which either reading a replaced b or vice-versa. According to the CBGM analysis, if a replaced b, then the three scribes who produced our three witnesses (01, 1852, P⁷²) independently and deliberately chose the same replacement reading a.

On the other hand, what we see from the diagrams above is that if the initial (original) text was a, then one scribe, the producer of 03, deliberately changed the reading from a to b, and with one exception all other extant manuscripts were copied with the reading here of 03. Mink may very well have done this additional analysis and omitted it in his chapter for brevity. The crucial difference between the scenarios is that this one does not have the occurrence of three identical readings produced independently, the only explanation for which appears to be a coincidence.

A traditional textual critic arriving at the same conclusion would, as I implied above, most likely have a slight preference for the combination 01/P⁷², and say that this is a case in which an error crept into the copying process “very early” in the history of the text. This usually is a tactful way of saying that 03 has a mistake. The split between 03 and 01 is evident, leaving the traditionalist with a difficult choice, but I think the CBGM analysis is helpful when substitutions of witnesses are made for the traditional viewpoint.

I will present three more examples, which should go more quickly now that we have covered most of the details found in the CBGM. The first is another discussed elsewhere by Mink, this time in his online presentation of the method.[72] I have already mentioned the verse: 1 Pet. 4:16. From the traditional perspective the reading “in this name” has excellent support: P⁷², 01, 03, and a number of other manuscripts, but the agreement of the first three witnesses is impressive. Ordinarily one would need to ponder nothing further. The problem, as I noted previously, is that the alternate reading of “part” for “name” is unquestionably more difficult, and therefore by standard internal criteria, more likely the original. It is also is problematic because it is the Byzantine/Majority Text reading, though of course, this is good news for those who favor that text. “Name” has been the preferred reading in the NA text until now, with NA28 reading “part.”

Mink begins the analysis with the assumption of “part” (reading a) as the initial text since it is the more difficult reading. The textual flow for reading a shows virtually perfect coherence. Mink then retrieves the CBGM v.1 diagram showing the textual flow from “part” to “name” (reading b). Version 2, below, is somewhat different, but not in any significant detail:

The layout proves to be more complex than what we found for Jude 15, but as Mink notes, we find a lack of genealogical coherence, in this case for reading b at average connectivity. There are eight separate genealogical “branches” (Mink’s word is “set”) in the box–version 1 had 10–and their most closely related ancestors are four witnesses that are all outside the box (above it) because they have the other reading. Notice that 03 inherits directly from A, and 01 and P⁷² both inherit indirectly from 03, all of which is consistent with the traditional viewpoint if one leaves the original reading open to either b or a.

At the top level inside the box, the scenario is like that of reading a of Jude 15 if b were the initial text: independent creation of the same variant, only not just three times but eight! Following Mink’s reasoning for Jude 15,[73] it seems that the unlikelihood of coincidental agreement on a larger scale would strongly favor reading b in this case. There is one important difference, however. If we look two verses earlier at 1 Pet. 4, we find “name of Christ” in v. 14, and in v. 16 we see “suffers as a Christian,” which might remind a scribe of “name” as a concept. So unlike the Jude passage, we find here in the immediate context wording that could suggest to a scribe to substitute “name” for “part” if he found the latter confusing, as he very well might.

If b is taken as the initial reading, the diagram for reading b shows something else interesting:

The textual flow of witnesses with reading b is nearly perfect in the center of the box. Even with b as the initial reading, however, the three witnesses reading a remain as sources for a coincidental creation of b because they are the nearest potential ancestors of the witnesses inside the box connected to them. This would mean that if b was original, it was surprisingly changed by one or more scribes to a. Then while most scribes copied b directly from a primary exemplar (as represented by the CBGM), there were some who copied from manuscripts with reading a, but corrected the reading to b coincidentally. So we still encounter a lack of coherence, and the analysis is also problematic because it requires us to accept another case where scribes changed the easier reading to a harder one; not impossible, but still unlikely.

Mink goes on to show that the textual flow for reading a when taking it as the initial text has nearly perfect convergence. Since it is the harder reading, this is to be expected, he notes. I am inclined to agree: it is difficult to explain how or why several scribes would change the easier reading to the (same) harder one.

Granting that, however, we have seen significant coherence for b as the initial text as well, and from the traditional viewpoint, it has external evidence on its side. What we would also like to see from the CBGM, then, is if there is a plausible textual flow from b to a, which of course we can query. This is the output from the CBGM application (v. 2) with b as the initial text:

Once again, I have zoomed in on the relevant part in the second diagram, which is a copy of the first. If Mink queried this attestation in version 1 of the application, the main difference in the result was that 468 (thirteenth cent.) was not recognized as a descendant of 025 (ninth cent.) and the arrow to 468 came directly from b:A (outside the box). Another arrow pointed from b: A to 025, which had no descendants. Here is the relevant part of the diagram:

What led to the change in version 2 is that version 1 showed 468 and 025 as having an equal number of prior and posterior readings, with a large number that were at that time “unclear” to the editors of the initial text. After one or more “iterations” (i.e. reviews of all the evidence by internal criteria), 15 of these unclear readings were determined to be “prior” for 025 relative to 468, and only four for 468, making 025 a new ancestor of 468 by CBGM standards, second only to the initial text A.

Let me add as a side comment that this comparison of the two diagrams also illustrates further how the CBGM application works, and what it is designed to establish. From a real-world, somewhat skeptical viewpoint, one might ask why version 1 above could not be the way 468 and 025 were actually produced (ignoring any dating issues), i.e. independently of each other, regardless of what the CBGM editors decide about prior and posterior readings? The CBGM answer is that when any two witnesses show a relatively high level of agreement (in this case 91%), it follows that the agreement and the relationship between differing readings are not results of coincidence. A preponderance of prior readings in one witness indicates that it is the ancestor. So rather than treat 468 and 025 as two separate branches (according to v. 1), the computer application is written to automatically link them (v. 2) when it is determined that one is the potential ancestor of the other.

So if we accept the conclusions about reading a indicated by the diagram from version 2 with b selected as the initial reading, it is telling us that the witness (text) found in ninth-century manuscript 025 is the first extant witness in the genealogy whose scribe either changed “name” to “part” or copied another manuscript that already had this change. The scribe who produced 468 faithfully preserved “part,” then we have two other manuscripts (307 and 617) in which the same was done, and a significant number of others produced from them. This does not encounter the same problem found in Jude 15, that of three different scribes seemingly choosing the same alternative reading independently and without a discernible reason. It is much easier to believe that one scribe can choose a different reading, for whatever his reasons. From that point on, we find nearly perfect coherence in the textual flow for reading a, so the answer to our question about the plausibility of the flow in this direction seems to be affirmative.

This analysis also clears up potential confusion over the diagram of the attestation of b with b as the initial reading earlier above, with three different a witnesses (424, 2423, and 642) as the sources for some of the b readings. If it appeared then that these three witnesses might have acquired reading a independently, which would have been like the problem in Jude 15, we now see that in fact (according to the CBGM) they are merely descendants ultimately of 025; all three witnesses appear as such (through 617) in the full diagram of the attestation of reading a. We still have the scribes of the three witnesses independently changing to reading b, but the initial change from b to the harder reading a can be attributed to witness 025.

I am not sure that the CGBM analyses of the variation unit make the editor’s or translator’s decision much easier, but I think they do clarify the issues. We know that the ECM editors chose reading a (“part”) as seen in NA28, and there is no doubt that it is the harder reading. Either reading displays good coherency if one can account for the anomalies in the reading b.[74] For those, there are the possibilities that the agreement was coincidental (inferred from the context), or with a bit of a twist, one can argue that there is too much agreement to be coincidental and therefore the anomalies actually support b as the initial reading.

Moreover, I think this example brings into as sharp relief as any could the fact that the CBGM application functions as an eclectic method, and in the process, it also focuses our attention upon the main point of contention between thoroughgoing eclecticism and reasoned eclecticism. The former, you will recall, uses only internal criteria, while the latter aims at a balance between internal and external criteria. Thoroughgoing (sometimes called “radical”) eclectics complain that those using “reasoned” eclecticism inconsistently allow external criteria to overrule internal. I will say this about the inconsistency, apparent or real: there is little else that one can do with external criteria. If they always supported a decision made from internal criteria, then there would be no need for external criteria at all, other than as additional evidence for decisions.

In the example from 1 Peter, there is no division within the Alexandrian manuscripts, or what many traditionalists would call the “best” manuscripts. Not only that, it appears that the witnesses supporting reading a are all late in date. This is the reading of the Byzantine text (to use the traditional label). The authors and editors of the CBGM and its initial text “A” have a very high regard for 03, the traditional favorite, and a comparatively low regard for either the Byzantine or the Majority Text (sometimes treated as the same), despite comments to the contrary. But what we have seen in the CBGM application is a complete disregard for any of these viewpoints. In the diagram showing the attestation of reading b from reading a, in particular, we see 03 represented as incorporating a change of the original reading as found in late manuscripts, including Byzantine witnesses. In contrast, we see that scribes of the Byzantine witnesses, in which easier readings usually prevail, steadfastly preferred the harder reading in this case. This is exactly the kind of result that advocates of the Byzantine text may desire as support for the text, given that the CBGM is built on an infrastructure of traditional internal criteria, rather than those favoring the Byzantine text.[75]

If we ignore the acknowledged superiority of these Alexandrian manuscripts and can accept that the reading “part” found in late manuscripts actually was the original reading many, many copies removed, then the “harder reading” criterion clearly favors it. The reading cannot be explained away as a transcriptional (scribal) error. Yet traditional external criteria are firmly on the side of “name,” and reasoned eclecticism can claim that this overrules internal criteria, even that of the harder reading. So the difference between the two systems is priority or ranking of criteria at various levels. A practitioner of reasoned eclecticism might say that external criteria overrule most internal criteria, but not the harder reading, with some exceptions. In the case of 1 Pet. 4:16, some may insist on “name” due to the manuscript support, and their greatest challenge (according to the CBGM) will be to account for the change to “part” in one decisive witness. Others, taking the view that internal criteria ultimately overrule external, will say that the harder reading, in this case, should prevail and that the manuscript evidence reveals to us that the change to “name” was made unusually early in the textual tradition. I think it is fair to say that the CBGM is effective in mapping out these possibilities for us.

In the next example, I offer, there is no significant difference in meaning among the variants, and it was most likely a transcriptional error that led to an alternate reading: James 1:20. It is material to our discussion because it is another case where the ECM editors have preferred a reading different from that favored in previous NA editions and supported for the most part by late manuscripts. This and the other two CBGM a-readings that we have discussed appear in NA28.

The example found here in James 1:20 is not discussed by Mink, but its textual flow in CBGM v. 2 bears some resemblance to those of the previous two examples. We will have to look carefully at the Greek spelling; fortunately, in this case, it can easily be represented in English. The problem is the phrase “does not accomplish.” The alternate reading supported by the Alexandrian manuscripts has a basic form of the Greek verb, while that chosen by the ECM editors has a compound form that is slightly intensified, but practically synonymous. There is one other reading that has the same compound form, but without “not.”

I have to distinguish CBGM v. 2 from v. 1 in this case because when the first version was done, the editors could not decide what the initial text was. As a result, the stemma looked like this, with question marks:

1. a)  ουκ εργαζεται
2. b)  ου κατεργαζεται
3. c)  κατεργαζεται

As you can see from the Greek terms in the list of readings, the first word in a, which is “not,” has an additional letter at the end (kappa) because the next word begins with a vowel (epsilon). This is like our practice of adding ‘n’ to ‘a’ (“an”) before vowels. The problem is that in Greek, there were no spaces between words, and the compound version of the verb begins with kappa as in readings b and c. So what we actually find for readings a and b in the corresponding English letters are “OUKERGAZETAI” AND “OUKATERGAZETAI” (the Greek had the equivalent of all capitals in English). The question becomes whether the original reading was a (chosen for previous editions of NA) or b, in which case a was the result of the omission of ‘AT’ from “OUKATERGAZETAI.” As I discuss elsewhere, one of the traditional guidelines for textual criticism was a preference for the shorter reading, which in this case would favor a. But to the contrary, modern scholarship leads us to consider that the omission of ‘AT’ might have been more likely. In version 2 of the CBGM, the stemma for this reading was resolved to this:

1. a)  ου κατεργαζεται
2. b)  ουκ εργαζεται
3. c)  κατεργαζεται

It shows that after further iterations, the editors came to the conclusion that b (v. 1) was the initial reading, so it became the new a reading. As for c, which lacks “not” (i.e. “the anger of man accomplishes the righteousness of God”), the CBGM now represents it as derived from a by a simple omission of the negative. We will, of course, look at its attestation, but you might wonder why c does not win out by virtue of being the hardest of the readings found here. The reason, as we discussed in the chapter on criteria, is that the harder reading principle is tempered by the caveat that it should not be too hard to be possible for the context. I am a great believer in the harder reading, so I usually require a reasonable explanation for its rejection if it is to be rejected. In this case, we have one: just before the letters ‘OU’ for “not” we have the same combination ‘OU’ as the ending in the Greek for “God” (the English word order differs from the Greek). So again, without spaces between words, what a scribe saw for readings a and b was actually ‘OUOUK’. I think it would be easy enough for a scribe to omit one ‘OU’ combination by mistake, without realizing that he had thereby omitted “not.” This would eliminate reading c from our consideration if it does not commend itself otherwise as the original reading.

Now then, what does a traditional textual critic using reasoned eclecticism do with this variation unit? It is a relatively simple decision if one strictly adheres to the “shorter reading” guideline and to external criteria: reading b above (CBGM v. 2) is to be preferred as the shorter reading (i.e. non-compound verb, and we exclude c as a blunder) and as the one with the best manuscript support, i.e. 03 and 01 in agreement, while a has the additional disadvantage (in the eyes of most) of Byzantine support. The one thing I find troubling about this is that taking b as original requires that I explain a as the result of a scribe’s deduction somehow that James must have intended to use the compound KATERGAZETAI, and that ‘AT’ was omitted in the scribe’s exemplar by accident. A quick check of James’ vocabulary reveals that the compound form is in fact used elsewhere one other time, earlier in 1:3; but he also uses the simple form once, later in 2:9, which would seem to negate a decisive preference for the compound. In any case, I consider the book too short to establish patterns like this in James’ usage (others might disagree). It seems simpler to infer that the scribe instead found the compound form in his exemplar and mistakenly copied it as the simple “OUK ERGAZETAI” (i.e. “OUKERGAZETAI” instead of “OUKATERGAZETAI”). All this said, however, the combination of traditional internal and external criteria have resulted in the choice of b in most Greek texts until now.

If we query the CBGM for the attestation of reading a taking it as the initial reading, we are given the following diagram:

No doubt it is too small for the reader to make out the details, but I am sure that Mink would call it another case of perfect coherence. Here is a portion of the left half in greater detail:

Nearly every witness is a direct descendant of its ancestor. I count three in the second position, and only one in third and one in fourth. If we check the attestation of reading b taking it as the initial reading, the result is the following:

This is quite similar to what we saw for 1 Pet. 4:16. Note that there are three extant manuscripts reading b whose scribes coincidentally changed the a reading they found in their exemplars to b. We can assume that they either omitted ‘AT’ from their exemplars by mistake, or correctly guessed what the original reading was (if it was b) and deliberately changed to that. Of these three, only one (321) has descendants that are somewhat closely related percentage-wise (89-90% range overall), and they all have reading a.

Another feature we can determine from careful observation is that coherence is good at the first generation and most of the second if we exclude the three manuscripts just discussed and manuscript 93, which is a fifth-ranking descendant of A. Then, a look at the potential ancestors of 93 reveals that it tends to agree with the Majority (Byzantine) Text, and agreement with the MT increases with 93’s descendants, most of which consequently are more distantly related to their ancestors and thereby to 93 in this attestation. That is arguably what we would expect, from a traditional viewpoint, of the relationship between the early (Alexandrian) manuscripts and the Byzantine text.

If we run the attestation for a with b as the initial reading, this is the output:

For greater legibility, I again delete the right portion:

Since we already know that coherence is very good with a as the initial reading, it is no surprise that it remains good once the initial change to a is made from b: A. Moreover, the diagram shows manuscript 1739 (tenth cent.) as being the leading ancestor of the a reading and making the change directly from the initial text (b:A). If we look at the potential ancestry of 1739, we see A (the hypothetical initial text) as its highest-ranking ancestor, and 03 is next, its first real ancestor.

Let me reiterate a note of clarification here about the CBGM potential ancestors analyses. While we have the flexibility to designate any reading as the initial reading in an attestation, just as we have done above with reading b, the computer application never allows any modification to the initial text A in the ranking of potential ancestors and descendants. Therefore when we query the application for the ancestry of a manuscript, e.g. that of 1739, the A-text always has the reading a that is found in any attestation. This may be an aspect of an overall flaw in the application, that of the user’s being precluded from changing any decisions of the editors. One might think that changing a single decision would have no effect on the ancestral relationship between two witnesses, but this is not the case, as we have already seen for witnesses that have zero genealogical direction.

To return to the present attestation taking b as the initial reading, we know that 03 also has the b reading, so the scenario that we can construct, if we wish, is that of the scribe of 1739 copying 03 or a lost manuscript containing the same witness but changing the reading from b to a (for whatever reason), which is subsequently copied by the scribes of all descendant witnesses/manuscripts. This is another example of the computer application’s revealing that external criteria are ignored. A proponent of reasoned eclecticism might very well prefer to give 03 preference, as opposed to designating b as a hypothetical initial reading that is also temporary, since the application defaults to the A-text.

It is also noteworthy, for traditional textual critics and translators, that the attestation diagram above shows C (04), a fifth-century manuscript, inheriting reading a from 1739, its second potential ancestor. This is a clear instance of thoroughgoing eclecticism taking precedence over the traditional criterion that the older reading is to be preferred. I have already conceded, of course, that the older reading may be found in a later manuscript. Nevertheless, for more natural or traditional dating some might prefer to interpret C as being copied from 03 or another old but lost exemplar, and if from 03, to understand C’s scribe as changing the reading, just as 1739’s scribe (or the scribe of its witness) is assumed to have done in this analysis. In that case, in the CBGM C would inherit directly from b:A just as 1739 does because A is its highest-ranking ancestor (1739 is second and 03 third). In the diagram we could put a dotted line from b:A outside the box to 04 (C). However, one of the basic assumptions of the CBGM is that scribes tended to follow the exemplars they had rather than introduce changes, so if dating is not an issue, then by this assumption 1739 is preferable as the ancestor of C here simply because C agrees with its reading.

If we query the attestation of reading b with a as the initial reading, this is the result:

We see five independent and presumably coincidental branches (four from extant manuscripts), and the relationships between a number of the witnesses branching off further from manuscript 93 (fourth from the upper right in the box) are rather distant, as previously, accounting for the Byzantine text. Mink probably would say that the coherence is poor. In this scenario, the change from “OU KATERGAZETAI” to “OUK ERGAZETAI” was relatively frequent and random compared to the reverse, where b is taken as the initial text from which a is derived. The latter attestation, you will recall, has only one change from b to reading a. So if b was the original reading, at some early point, a scribe changed it, and the new reading proved to be connective, or persistent. We could even hypothesize that some scribe copied 03 and changed 03’s reading for this verse. There is no question but that 03 as a witness represents many, or even most, original readings in the New Testament by standard internal criteria.

So, did the ECM editors make the best decision when they changed the reading of James 1:20 in NA28 to KATERGAZETAI? I have little doubt that it won out because its genealogy is much more coherent than reading b, and because traditional external criteria were ignored. What the genealogical evidence indicates is that once reading a was established–whatever the initial reading–scribes almost uniformly maintained it in their copies.

By comparison, any attestation of reading b is more or less messy. If it is set as the initial reading and the Byzantine branch (93 with descendants) is ignored, as it probably should be in this case, then there is good coherence with eleven manuscripts and only coincidental agreement for three. If a is set as the initial reading, however, there is practically no coherence, only what we see above in the case of 01 and 03 (secondary).

We could leave the decision process at this point, accepting NA28 if we think coherence should prevail or retaining the reading of previous additions if we think external criteria should prevail. A true student or scholar of textual criticism does not leave the process now, however, but recognizes that the discipline is as much, or more so, an art as a science. So we should ask ourselves about the psychology of the scribes from what we know of the evidence.

Given that the witnesses for reading a are so connective, or as I said, persistent, I infer that the reading looked correct to them. If so, they immediately saw OU (Greek “not”) in the combination OUKATERGAZETAI as one word because they immediately recognized KATERGAZETAI as a word, which they had also seen previously in what is, to us, v. 3. Therefore it did not occur to them to change it.

On the other hand, suppose there were some who instead saw OUK (the normal spelling of “not” before a vowel) first, followed by ATERGAZETAI. There is no such Greek word as the latter, so I think that would have left them puzzled and looking for a solution. The prudent course of action, and incidentally the one consistent with CBGM assumptions, would have been to reevaluate OUK as OU and immediately solve the problem. By contrast, I think it would have been rather pig-headed to stick with OUK and solve the problem by omitting ‘AT.’

So I do not think it credible that those scribes who chose reading b did so by this process, which we would call an error of the eye. What about an error of the ear? If someone was reading the exemplar to them, or if they were reading it to themselves (which would be common enough), is it possible that they missed the ‘AT’ or unconsciously ignored it for the simpler form of the Greek verb? It is a syllable that sounds nothing like ‘ER’, but it is, after all, a single syllable, so we probably cannot rule out the possibility. Perhaps the somewhat sporadic attestation of b with a as the initial reading reflects this, supporting a as original.

If b (OUKERGAZETAI) was the original reading, what would have led a scribe to change it to OUKATERGAZETAI as seen in the CBGM attestation? The uncompounded ERGAZETAI would have appeared simple, and OUK would have been expected with it. The only thing I can imagine as motivation is a desire to echo the wording of v. 3, otherwise, changing the verb to the compound form would seem an unnecessary complication. However, once the change was made, it most likely would have appeared original to other scribes as it was passed down through manuscripts. Also, if some whose exemplars had reading b had access to other manuscripts with reading a, we could expect them to adopt reading a in their copies. The CBGM attestations seem to indicate that this is what happened.

In the final analysis (if I may use the term loosely), I think that those who prefer to follow the “best” manuscripts have good reason in this case, despite the difference in coherence between the two readings. As I see it, the CBGM analyses do not provide us sufficient help in deciding what the original reading was; they only supply a reasonable history of the textual flow that could result from either reading.[76] This is not to say that we necessarily have anything better with which to work from the external evidence; the only thing it gives us for certain is a documented early date for the reading found there (but C, also relatively early, supports reading a). Ultimately, we find ourselves attempting to retrace the thinking of the scribes, which is in the category of internal criteria.

The fourth and final example that I offer probably is unique in that the choice of reading for the initial text A (and for NA28) is not supported by any Greek witness. It is found in 2 Pet. 3:10. The Apostle states that the elements will be destroyed by being burned, then he prophesies that the earth and its works will be…what? The “best” manuscripts have “found”; P⁷², the papyrus to which I have referred previously, has “found destroyed,” and manuscript C (04), also mentioned above, has “disappear.” The Byzantine reading is “burned up.” As far as the translations go, the NIV has “laid bare” while the ESV has “exposed.” Both are paraphrases of “found.” Ordinarily the NASB would not take the Byzantine reading, but in this case, it does, with a note that “[t]wo early mss read discovered,” referring to 03 and 01 which normally are preferred.

When we query the CBGM, we find that the initial reading, also found in NA28, is “not be found.” No Greek manuscript in existence has the word “not” here. Assuming that this was the original reading; however, the stemma is relatively easy to construct, granting that the word “found” is part of the original. The CBGM gives us the following:

1. a)  ουχ ευρεθησεται
2. b)  ευρεθησεται
3. c)  ευρεθησονται
4. d)  ευρεθησεται λυομενα
5. e)  αφανισθησονται
6. f)  κατακαησεται
7. g)  κατακαησονται
8. h)  καησονται

If you do not read Greek, allow me to assist by pointing out that the first word in reading a in the list below the diagram is the Greek “not” (another variation of OU, like OUK) followed by the word for “found”. Reading d has “found destroyed,” e has “disappear,” and the last three have some form of “burned up.” The arrows in the diagram suggest the following: b and c were the result of the accidental omission of “not.” Reading e probably was a deliberate change by a scribe from the negative “not be found” to the positive “disappear.” The scribe responsible for reading f found the original odd and chose to repeat the idea of burning found earlier in the verse instead. Reading d was the deliberate addition of “destroyed” to b’s “found” to make sense of the reading. Reading g is a change from singular to plural for the context, and h is just a simplification of g by eliminating the compound prefix.

Now that we have the stemma, we should query for the attestation of reading a, taking it as the initial reading. What we are given may be unique in the CBGM:

You will immediately notice that there are no arrows and no other witnesses. The initial text A is hypothetical by nature, so the hypothetical existence of this reading is not a problem. Instead, the difficulty is the fact that there are no descendants in existence with reading a, unless we decide to include non-Greek witnesses, namely the Sahidic, the Coptic dialect V, and some Syriac witnesses. Thus there is no Greek witness available to go in the box with A.

If, however, we change the parameters and ask for the attestation of one of the other readings–say the Alexandrian reading b–with a still as the initial reading, then we obtain a diagram that looks normal:

In this analysis the dotted arrow from a:A going into the box tells us that the source of witness/manuscript 03 was the initial text A, but the scribe of the witness of 03 found reading a and changed it to reading b. The change from “not found” (Mink’s choice based on ancient versions) to “found” is illogical, and we are to assume that “not,” if original, was omitted by error. However, the change for witness 018 is a problem that we will look at below in more detail.

If we so desire, we can obtain a rather more complicated diagram showing the textual flow for all the readings and their witnesses. These diagrams are obtained from the “Coherence at Variant Passages” module. No doubt for the sake of relative brevity, the diagrams do not include all the witnesses supporting the readings, just those that are genealogically connective to witnesses supporting other readings. The attestation diagrams that we have already examined are obtained from the “Coherence in Attestations” module, and they provide more information for each reading. Compare the stemma for the reading above with the following diagram at average connectivity (= 10):

This diagram shows some contradictions with the stemma. For example, the latter shows a to b to d, while this diagram has direct, independent lines from a to both b and d. The online diagram is color-coded: blue lines are direct ancestors/descendants, green represent lower-ranking relationships, with rank numbers shown for convenience. So the diagram indicates that the scribe of reading d (“will be found destroyed”) more likely was influenced by the hypothetical reading “not found” than by the real b reading “found” from witness 03, the ancestor to P⁷² that is second only to the initial text A. My guess would be that since the stemmata were done for each variant unit at the beginning of the CBGM process and later revised as more data and analysis were acquired, the stemma we have in this case has not yet been revised to reflect the textual flow. According to the diagram, we also see a as a direct contributor to g along with f. On the other hand, a is not shown as a contributing ancestor to c at any level, in contradiction to the stemma. As it happens, manuscript 398, the only witness listed for reading c, shows A as an ancestor at rank 15 and agreement at 89%. However, 424, with reading f, is the first ancestor to 398 with about 92% agreement. So according to the diagram, it is more likely that the scribe of reading c (“will be found” plural) was influenced by f (“will be burned up”), the Byzantine reading. I find that very curious, indeed, counterintuitive.

Moreover, we also find 424 identified as the source for 018, a witness for the Alexandrian reading b (“will be found” singular). The only explanation I can imagine at this point for either of these phenomena is that our scribes were aware of the Alexandrian reading, recognized it as the harder reading, and deliberately replaced the easier Byzantine reading with it as the more likely original. If so, they clearly did not agree with Mink in considering the Alexandrian reading so difficult as to be impossible. It is also interesting that both the singular (b) and plural (c) forms were supplied, either of which is justifiable. This shows that they had a minor problem with the grammatical construction, but not with the concept–though of course, other scribes did, as exhibited in the differing readings.

Before we accept this puzzling analysis, we need to consider the fact that it is based on the default level of connectivity (10), and we can set it for higher numbers (i.e. lower levels), or even for the “absolute” level of 499 that will recognize virtually any connection between two witnesses no matter how remote. I did this and obtained the following result:

At this most remote level, notice that there is now no connection at all between witnesses 424 and 018. For the explanation, we need to consult the attestation diagram for 018 at a high number of connectivity, and we learn that it inherits reading b from witness 1448 as its eighteenth ancestor. This is not shown in the diagram immediately above because it is not relevant to the connections between readings. If we check the table of potential ancestors for 018, it confirms for us that 1448 is the eighteenth-ranked ancestor at 91% agreement. This does not change the relationship between 018 and 424; 424 is still ranked first as ancestor to 018 at nearly 97% agreement. So the diagram above reveals that inheritance of the same reading always takes precedence in probability over changing the reading of a close ancestor (in this case 424), provided that the number for the level of connectivity is set high enough to identify an ancestor with the same reading.

As a matter of fact, now that we know that a connectivity level of 18 links 018 to 1448, we can if we wish to check the attestation of reading b at that level to see all the relevant connections in that attestation:

The connections begin with 03 (theoretically altering the initial text A exhibiting Mink’s preferred “not found”), descend to 1739, from there to 025, and from there to 1448, with a total of twelve generations removed between 1448 and 03. The table of potential ancestors for 018, on the other hand, has 025 as twentieth ancestor, 03 as twenty-eighth, and 1739 as thirtieth, so of the witnesses listed here, it is connected to 1448 as its closest ancestor of the four.

Now we need to keep in mind that what all the missing generations represent is differing readings. In the case of 1448, for example, there are seven other witnesses/manuscripts that are more closely related to it than 025 (including 424 and A), and in this case, with the exception of A, they all have reading f (the Byzantine). So coherence appears to be poor. It is tempered by the fact that 025 agrees with 1448 in almost 89% of their variation units, which is better than average. Coherence appears to be much worse for the relationship between 018 and 1448 with 1448 ranking only as eighteenth ancestor, but it agrees with 018 at 91% (its first ancestor agrees at almost 97%).

Something very interesting happens if we query the potential ancestors module for 1448 in 2 Peter only. Its first ancestor is irrelevant because it, unfortunately, has a large lacuna encompassing our passage. The second ancestor, with reading f in this case, shows affinity for the Alexandrian manuscripts; the third is the initial text A, and the fourth is 03. So if the application provided attestations by individual books, 1448 would rise to the top as a third descendant of A.[77] Witness 018 advances only to sixteenth descendant to 1448, but its percentage of agreement increases to almost 94%. I should note that 424 continues to be first ancestor to 018 at a very high percentage of agreement, almost 98%. To be more specific, in 2 Peter there are 413 passages with variations, and 018 agrees with 424 in 404, leaving only nine passages where it disagrees. In one of these cases the CBGM editors consider the variations unrelated, so at this point only eight are relevant.

In real-world terms, it appears that the scribe of 018 copied from 424, or a lost manuscript almost identical to it, as his exemplar. The only practical difference is that in the latter case we assume that 018 is a virtually perfect copy of a manuscript that differed from 424 in the same ways that 018 does. The evidence indicates that some scribe (or scribes) made only eight significant changes to the witness (text) found in 424 out of more than 400 variation units that are known to exist among extant 2 Peter witnesses. Fortunately for us, we do not need to know who was responsible, only what changes were made. With such high fidelity to the witness of 424 (98% in 2 Peter) and the indisputable, qualitative difference between the two readings in question, it is hard to escape the conclusion that the scribe of the witness of 018 did, in fact, reject the reading in the witness of 424.

We are then faced with two provocative questions. The first is a consequence of the CBGM feature allowing the user to select the connectivity number in coherence diagrams. When we used the default, “average” number 10, we were informed only that witness 018 changed the reading found in witness 424, not how the choice of reading was made. The application left it entirely to our imagination to determine why and how the scribe of witness 018 came up with the reading found in the Alexandrian manuscripts. When we increased the connectivity to “absolute” (499), the connection between 018 and 424 was severed, and we saw from the attestation for b at connectivity 18 that the reading then came from witness 1448, which was a distant ancestor compared to 424. So do we infer that the scribe coincidentally invented a change that agreed with the Alexandrian witnesses, or did he somehow have access to the witness of 1448? Given that there seems to be nothing in the text to suggest the Alexandrian reading, together with the fact that 018 is a highly faithful copy of 424 overall–and thus the scribe of 018’s witness was not inventive–he most likely became aware of the Alexandrian reading.

If the scribe was indeed aware of the reading in 1448, he probably was not searching for it as an alternative to that in 424 since the latter is the easier reading. It seems far more plausible that he was routinely comparing the two witnesses. Between them, 1448 has 17 fewer agreements (still a high number) and is less firmly established as an ancestor, so 424 is the slightly better choice as the exemplar. The second question, then, is, did the scribe actually choose the harder reading over the easier? It appears that he did, at least according to the CBGM results. So in this particular case, we have a scenario in which a Byzantine witness (018) displays a hard reading supported by early Alexandrian witnesses, one (or more) of which most likely was available to the scribe of the Byzantine witness.

Before moving on, I suggest that we consider one more scenario, that of b as the initial reading, which the traditional criteria favor. With “average” connectivity (= 10) this is the diagram that is returned:

In this option (or any choice with initial reading other than a), reading a disappears because it does not occur in any Greek manuscript, and if I may say so, making this selection allows the CBGM to operate as it was so designed, even if Mink is the designer. Let me clarify: for the programmer, reading a was easy to input, but it represents either a pure conjecture or a back-translation from one or more of the ancient versions in which it exists. Arguments can be made in favor of including either, however, unless they are strictly optional, consistency demands that they be included for every variation unit in the ECM whenever relevant, and this is not done. Aside from the philosophical and philological issues, adding either would add a very large, additional dimension to the CBGM that it might not be able to manage. And I think a compelling argument can be made that including conjectures or back-translations (or original readings from versions) should be optional in any case. In the present case, reading a was included as part of the initial text, which cannot be altered–at least not off-campus–by CBGM users.

If by now you are becoming comfortable with the CBGM, you will notice that like the earlier diagram with a as the initial reading, there appear to be multiple connections between readings f and g. In Mink’s terminology, the same change from f to g emerges multiple times, indicating coincidental coherence. If we compare the readings, there is a simple philological or linguistic reason. The two readings are identical except that one (f) is singular while the other (g) is plural. The singular, which would not be acceptable from an English viewpoint, can be used in Greek as going mainly with the singular “earth.” Also, “works” is a neuter plural in Greek, and neuter plurals take singular verbs, so the singular number would not necessarily be rejected as a mistake. One could think of it as the harder reading, however, because the combination of the earth and its works would probably appear to be better represented by a plural verb. So it would follow that a scribe would be inclined to change from reading f to g, and this, of course, is what we find in both diagrams, witnesses with the latter reading having witnesses 2200, 252, and 424 as their first ancestors and sources (424 rates as first ancestor to two g-reading witnesses). Witness 623 in the g attestation actually inherits from witness 5 (its first potential ancestor), while the best source for witness 5 is A (always hypothetical) or 468.

If we look at the attestation of reading g, we learn that the genealogical coherence among the g witnesses themselves is very poor, with few exceptions. Connectivity has to be set to 21 before all the witnesses will connect to a common ancestor (2186), so genealogical connectivity is low. Yet, this is an example where the other kind of connectivity, which has a linguistic or philological basis, is high. In this case, it is a matter of grammatical number, leading the scribes of the g witnesses to make the same change coincidentally.

Reading f deserves our attention as well. Its primary source genealogically is b, which makes little sense in this case with “found” (from witness 1739) being changed in 2298 to “burned up.” In terms of meaning, both the hypothetical reading a (“not be found”) and e “disappear” are closer. The first would be third ancestor to 2298 and the second (04) fourth, with about 92% and 90% agreement respectively, so they are both reasonably close. From a text-critical viewpoint, however, they are arguably more problematic than reading b because one must ask why change either to different terms with roughly equivalent meanings? If the scribe of the witness of 2298 encountered b, however, he might very well understand it as a clear mistake (just as Mink does), and look for something in the immediate context with which to replace it. The participle “with intense heat” (NASB) in the preceding clause would suggest the verb that was chosen. As to the problematic connection between 424 and 018, the connection again disappears as previously if we change to a high number of connectivity.

Continuing with the rest of the readings, at the lower left of the diagrams we see d (P⁷²) which has A as its most closely-related source. If the scribe of its witness encountered the hypothetical reading a (“not found”), then for some reason he chose to replace the negative with “destroyed.” We could theorize that since 03 ranks as second ancestor to P⁷², he perhaps compared the two readings and inferred that there must be some kind of problem with the negative. In that case, however, the simper solution would seem to be the replacement of “not,” rather than the addition of the more inventive “destroyed” (cf. Mink’s second assumption earlier). If, on the other hand, b was the initial reading, it would follow that the scribe did not want to do something as drastic as to negate the verb, and chose instead to add what was essentially a gloss explaining the reading. He undoubtedly conceived the idea from the preceding “will be destroyed” in the text. His choice of putting the Greek participle “destroyed” in the neuter plural was a grammatical match with the main verb in the singular, as I explained above.

Consistent with both diagrams, the first ancestor for P⁷² is A and the second is 03, so if we dispose of Mink’s reading a, then the scribe had the actual reading “will be found” before him. The percentages of agreement overall are not very high (03 is about 85%), but are significantly better for 2 Peter (03 at 91%).

If we look at the right sides of the diagrams, nothing much appears to change except for assigning the b reading to A. In accounting for e (04), however, reading a would seem to have a clear advantage in sense. The e reading “will disappear” is a positive way of saying “will not be found.” However, as I noted for reading f, the advantage may actually go to b from a text-critical viewpoint. Adding “not” is a much simpler and less creative change than replacing “found” with “disappear,” which does not even occur elsewhere in the book. In any event, the relative uniqueness of reading e here makes it of less interest than the other readings.

We are left with c and h. Witness h (2464) is not difficult to explain as deriving its reading from witness 623, its primary ancestor in either diagram. The reading of g (623) is a compound verb, exactly like that discussed above for James 1:20, and the compound prefix was deleted in reading h. This may have been a mental omission on the scribe’s part.

Reading c places before us the same problem that we encountered between witnesses 424 and 018, however. It is not surprising since this is merely the grammatical plural of reading b. I already noted the counterintuitive effect of the scribe of the witness of 018 changing the Byzantine reading f found in 424 to reading b.

We discovered that in the case of the relationship between 424 and 018, selecting remote connectivity eliminates the connection between them in the CBGM application by allowing 018 to inherit its reading from remote ancestors, and checking the potential ancestors for 2 Peter only increases their levels of agreement. The same cannot be said for 424 and 398 (reading c), however. If we check the attestation of reading c, we discover that it is unique even at the most remote level of connectivity, i.e. reading c has only 398 as an extant witness; 398 has no ancestors or descendants with the same reading.

It is quite implausible that the scribe of this witness would invent the plural “found” from the reading f; indeed, a change from the singular reading b would seem the most likely explanation by far, and short of that, some other reading including “found.” If we query the attestation of c at “absolute” (the most remote) connectivity, we see witnesses with b, g, a, and e as other ancestors for 398 along with f, and among these four witness 018 with b ranks highest overall with about 91% agreement. It ranks only as eleventh ancestor to 398; however, 424, the closest ancestor, is actually less than one percent higher than 018 in agreement with 398. The agreement is much lower for all witnesses if we query the potential ancestors module for 2 Peter alone, so 398 seems to have a significantly different character from 018.

Having examined the attestation of reading c and 398 closely, we can form some useful conclusions. As far as the CBGM application is concerned, the most likely source of the reading c is reading f, the Byzantine “burned up” (singular). Yet it is implausible, perhaps even incredible, that a scribe would independently change an easy reading to one so difficult as to be considered virtually impossible, at least by modern textual critics. Moreover, this is not a simple mistake, but a major lexical change that had to be a conscious choice.

The reading, as we noted, has no descendants as does its singular counterpart b. Of course, there is always the possibility that the descendants were in manuscripts that were lost. However, we saw that the corresponding change from f to g happened (or could have happened) several times coincidentally, and thus we have reason to expect the same for this reading, suggesting that there should be more extant copies, given that it is a tenth-century manuscript. The presence of multiple extant witnesses indicates that scribes had no problem with the singular form. More importantly, witness 398 stands as evidence that, if we assign reading b as the initial reading, then the scribe of the witness of 398 had no difficulty with the concept, only the grammatical form. If we take reading a to be the initial reading, then again the scribe focused on the grammatical form, and it appears that he must have omitted “not” by mistake, just as the scribe who originated the b reading must have done, by coincidence.

I think one more diagram, with the reader’s patience, will be interesting for illustrative purposes. Mink has made the claim that the CBGM is useful regardless of one’s position on textual criticism, i.e. regardless of the methodology or type of text (“cluster” may be the preferred term) favored. I have expressed my doubt that this will prove true since the decisions about variant readings are made on the basis of internal criteria, and not everyone (especially not those favoring the Byzantine/Majority Text) will accept what have become the standard criteria. The basis for Mink’s claim is the fact that the user can select any reading as the initial reading. I have done this a number of times above, but in no case have I assigned a Byzantine variant as the initial reading. Since reading f, the Byzantine, has been problematic in this last example, let me exhibit what the application returns when that is selected as the initial reading:

I venture to say that this is no worse in coherence than any of the other diagrams with connectivity at 10; indeed, one thing that we do not see is the complication of an arrow pointing upwards from 424 to 018. Instead, it joins A as a source for b. So it might appear that Mink is correct in claiming that even advocates of the Byzantine/Majority Text can make use of the CBGM. But the representation in this diagram is deceptive–not that this was an intended outcome on Mink’s part. We correctly think of the f box as the Byzantine reading, and it contains A as a witness due to the simple fact that we have temporarily assigned a Byzantine reading to the initial text. It certainly is not impossible for some readings formerly classified as Byzantine and secondary to be reevaluated as original if such criteria as the harder reading point in that direction, as in the case of “part” in 1 Pet. 4:16.

The case before us now, however, is the typical situation of the Byzantine reading’s being the easier of the options f, b, and c, and we know that in fact, this is contrary to the character of the initial text A. The other witnesses included for reading f really do have the reading and are in character with such a reading. So in the diagram above, we have the contradiction that A reads f, yet it remains the closest real ancestor to 03, P⁷², and 5 (a very mixed text) because the potential ancestors module of the CBGM application cannot be manipulated with different initial readings as the attestation and variant readings modules can. Mink’s conjecture, I would maintain, actually creates the same situation since the intent is to provide a similar meaning. As in the diagram above, A serves as the source for b, d, e, and g. The only difference is that in Mink’s hypothesis a source is needed for f, and 1739 with reading b is supplied as closest ancestor to 2298. This actually proves to be a problem in Mink’s theory, as I see it, because it appears that the accidental omission of “not” with “found” was the advantage and purpose of the theory. Even with the hypothesis, however, the CBGM application with a as the initial reading reveals that the Byzantine reading f (“burned up”) came from b (“found”), and this is the result regardless of the level of connectivity. So according to the CBGM itself, the scribe of the witness of 2298 made a deliberate change from the most difficult reading by using a different verb suggested by the context, and the change stuck. It does not prove that Mink is wrong, but one wonders why scribes did not think of it as the most likely explanation. Indeed, the closest extant reading to Mink’s conjecture is the elaborate reading d (P⁷²).

I would contend that the initial text A actually has no relevant role in CBGM analyses when a Byzantine (or any other) reading out of character with the A-text is selected as the initial reading. But I think that, fortunately, there is a simple workaround for the user. It begins with the conspicuous visual mismatch of any lowercase letter other than ‘a’ with ‘A,’ as in the diagram above where we find ‘A’ with ‘f.’ Mink and his colleagues might have been able to make life a little easier for the rest of us if they had identified such readings, for example by reserving b readings for “Byzantine.” Of course, I realize that it is more complicated than that, not the least because of the highly mixed character of many or most witnesses. So for now at least, it is up to the user to figure out whether a reading is incompatible with A. As noted earlier, information is available in the potential ancestors tables, showing the percentage of agreement with the Majority Text at the top, and of course there is also the “Byz” designation in NA and ECM2. As far as text-types go (to use the old term), we have only the Byzantine/non-Byzantine distinction in the CBGM for now as in NA, and one can only guess what the future holds. I personally doubt that the situation will change when analysis of other books is finished because the identification of types has been contrary to the spirit of the CBGM. Textual critics and other scholars will be able to perform analyses and form their own conclusions, however.

Being able at least to distinguish Byzantine readings as much as possible, I think the user just needs to ignore A and arrows connected to it when a Byzantine reading is selected as the initial. We are then left only with actual witnesses whose connections to other witnesses are valid (according to the CBGM). In the diagram above, for example, the source for b would then be–as we have discussed seemingly ad nauseam–witness 424. It is also the source for g, and logically so. To replace the remaining A connections, we have to investigate the next most closely-related ancestors. In the case of e, 1739 does well as second ancestor to 04, and 03 serves as second ancestor to P⁷².

Now we are in the position to come to final conclusions about the readings for this variant unit. In evaluating the default diagram that has the conjecture “not be found” as the initial reading a, we can consider it an appealing candidate as the source of reading b on the ground that it requires only the accidental omission of “not.” At the same time, however, we need to bear in mind that this initial reading is roughly equivalent in meaning to the easier Byzantine reading, making it unlikely as a source for an Alexandrian witness such as 03. If we check the potential ancestors for 03, in fact, we are given only A, the initial text, which normally would not feature the easier reading. The change from a to b can only be justified as a blunder.

Even ignoring this issue, the problems still mount up for reading a. I will address the readings alphabetically, beginning with reading c in 398; we know that the scribe of the witness of c made a conscious choice in his change of reading because changing the number from singular (as in b) to plural cannot be done by accident, let alone choosing to use a different word (i.e. rejecting f). Traditional textual critics and those who favor the Byzantine text alike will agree that a scribe who had reading f before him in his exemplar ordinarily would have kept it. Furthermore, it is hard to believe that a scribe who paid enough attention to change the grammatical number would make the blunder of omitting “not” if it were in the original. So I think that unless one simply ignores manuscript 398, it stands as evidence against the existence of reading a. Reading b commends itself both as the most likely source transcriptionally and the best alternative to reading a according to the CBGM.

Other than A, the diagrams show witness 03 as the most likely source for d, and I have already discussed why it is a more likely source than the conjectured reading a. For e, we are faced with the problem of the scribe’s operating against assumption one and choosing a roughly equivalent meaning to a, if “not found” was in his exemplar. If we set aside A, the only remaining source is b through witness 1739. It is however closely rated as second ancestor to 04 at only about one percent lower agreement (91%) than A. Clearly, the scribe who authored reading e made a conscious choice, not only rejecting the simple negative reading a if he had it but ignoring other words suggested by the context as well. In fact, the word chosen does not appear elsewhere in Peter. I consider it more likely that the scribe had only witness 1739 and invented a different reading.

I think it is fair to say that f is also problematic for reading a from a text-critical viewpoint. Its closest ancestor through 2298 is 1739 with the b reading, and A is ranked as third ancestor, with witness 04 fourth after that. It also happens that the second-ranked ancestor to 2298 is witness 323, which also has the b reading, and for that reason is not shown in the diagram of variant passages (since 1739 outranks it). What this indicates to us in real-world terms is that the scribe of the witness of 2298–i.e. the scribe who authored the text found therein–probably had before him witness 1739 as his exemplar, with the b reading. As so many ancients and moderns alike have, he judged the reading to be a mistake. Taking Mink’s assumption four as our guide, the scribe looked for alternatives in other manuscripts, and allowing him access to A and 04 would not violate assumption three or four, in my estimation.

Assumption one, the primary guideline, argues in favor of some variation with “found” since that most likely was in the scribe’s exemplar, so “not found” (reading a) should have been the best choice, and if Mink’s hypothesis is correct, it was also the scribe’s most immediate alternative according to CBGM analyses. That is, we are to assume that scribes had as the norm fewer texts, which were closely related. In this case, if our scribe had three texts, he essentially had 1739, 323 (both with reading b), and a witness containing the initial text with reading a. If he had four texts, he also had 04. So he rejected reading a if he had it. He also rejected reading e, probably viewing it as an invention by another scribe and one that did not commend itself from the wording in the context. Furthermore, an analysis of attestation for witness 2298 even at the most remote level of connectivity (= 499) does not allow 2298 to inherit reading f, so the logical conclusion seems to be that its scribe did not have reading a as an option.

We could consider witness 2298 especially important in the history of this reading because according to CBGM analysis it is the ancestor of the same f reading in all other witnesses, which is to say that its witness is the original ancestor of this Byzantine reading. Of course, one of the descendants is the problematic witness 424.

Reading f best accounts for the two remaining variants, g, and h. We have already noted that several f witnesses account for several g witnesses, a coincidental agreement that is easily explained as scribal preference for the plural (reading g) over the singular (reading f). For one g witness (5) we find A as the closest ancestor, and an f witness as second ancestor (468) which argues against the existence (or availability) of reading a yet again, since the choice of wording of a lower-ranking ancestor (“burned up” vs. “not found”) was preferred to it.

Reading h is the same as g with the compounding prefix deleted, and its primary source by the CBGM attestation is a g witness that ultimately (though remotely) inherits its reading from f witnesses. Its third-ranking ancestor is an f witness, which is the compounded verb in the singular. Initial text A comes in as fourth ancestor, but it seems irrelevant given the far greater similarity to the other two readings. If we have to decide, we are left only to determine whether the scribe of reading h (witness 2464) just deleted the prefix, or also changed the grammatical number when he deleted the prefix.

This variation unit is an excellent example of the caveat to the harder reading principle requiring (or suggesting) that it not be so difficult as to be impossible. As far as Mink is concerned, “will be found” is impossible as the original reading, and I think all textual critics would agree that it is the hardest reading of all the options. A good number of ancient scribes appear to have agreed that it is impossible. As we have seen, Mink’s solution is a conjecture supported by several ancient versions (i.e. translations) that essentially makes the reading compatible with the Byzantine reading, and it offers a plausible explanation for the reading found in 03 which is so troubling in view of the high value in which 03 is held. Unfortunately, I think the evidence from CBGM analyses makes Mink’s hypothesis doubtful at best. Unfortunately, that is, because the hypothesis has been hard-wired into the application, and even if it were not, it is appealing in its simplicity.

However, if the CBGM is truly viable, it should be able to retain its integrity and reveal strengths and weaknesses of any hypotheses fed into it. I think it has done that in this special case, though the application is in need of more options and fine-tuning. For all our misgivings about the reading “found,” the CBGM as I interpret it keeps pointing to that as the original. It also seems to be telling us that if we prefer reading f instead, as some do, then we will need to explain why “burned up” was changed to “found” in the b and c witnesses. This appears to be an impossible task, a theory even more difficult to accept than the hypothesis of reading b as the initial text. We also find in b, if it is not impossible, a reading that happens to meet the criterion of being the shortest reading, which can still have validity under such circumstances as these. The word “not” (Mink and versions) or “destroyed” (P⁷²) added to it serves to make more sense out of it.

Optimal Substemmata

There is one more aspect of the CBGM that we need to discuss prior to a final evaluation of it: the creation of optimal substemmata for every witness. Whether this will prove to be helpful to the user, or a necessary evil, or just an evil, is difficult to predict. Let me begin by explaining what it is.

We have already seen how stemmata for readings and witnesses are represented in the CBGM. In the case of witnesses, after stemmata are constructed for all variant units in a particular witness/manuscript, all the witnesses that have contributed readings in one way or another can then be listed as a complete collection of sources for this particular witness/manuscript, which has thus “descended” from these sources. That is, the list will be comprised of a number of source manuscripts, each of which will have more variant readings that were changed to posterior readings in the descendant manuscript than the reverse, and every variant reading in the descendant manuscript will be covered or accounted for by one or more of the sources. It will probably also help if we keep in mind the distinction between the tasks that are human and those that are (or can be) done by machine. Deciding which reading is most likely prior (or original) is a job for humans; they decide and input their decisions into the computer by classifying the readings as prior or posterior, and–I assume–they construct stemmata of readings. The computer then collates all the information, and among other things produces the potential Ancestors/Descendants databases that are so important to the CBGM.

There are of course many extant Greek manuscripts to consider now that later manuscripts which have largely been ignored in the past are included, together with new discoveries; and as you might imagine there can be a great deal of overlap among sources for a witness/manuscript. If you wonder about that, you only need to look at the apparatus for NA28 and see some of the listings of manuscripts for any given reading. These are abbreviated listings, at that; consult the ECM2 for more information.

The CBGM application has been written to produce a database of possible combinations of ancestors necessary to account for a witness, ultimately narrowing all ancestors to those absolutely necessary to account for all the variant readings. Even so, the list of necessary ancestors can be long, and Mink has been concerned that it will require too much computing power to process all the data. For example, ten potential ancestors do not pose a problem, but 20 results in over one million possible combinations of ancestors to account for all readings, and 30, in over one billion. He notes, too, that with the increase in combinations comes an increase in the number of witnesses included.[78] These are imposing numbers, certainly, and if the CBGM remains confined to the INTF mainframe, it appears to be a legitimate concern as, and if, access to the application increases substantially.

For many users, optimal substemmata could simplify the application because the sheer volume of data may make it appear daunting, and these substemmata promise to greatly reduce the numbers of potential ancestors to be considered while still covering every variant reading in a witness/manuscript one way or another. Also, if less computing power is required, results will, of course, come more quickly. I do not know how Mink can make the task of producing optimal substemmata practical, however, and can only assume that he has a plan for it. The approach that he describes in both his Contamination article (pp. 189-199) and the online Presentation (485-560) appears to be very labor-intensive, for only 10 potential ancestors and eight respectively in the two examples. Even he concedes that the process is complex, with frequent interaction between computer processing and philological (i.e. human) assessments. I would think countless hours of committee work would be required.

I also discovered what seems to be potentially a serious obstacle to completing optimal substemmata. The example Mink gave in his Contamination article (pp. 190-99) was an optimal substemma for witness 323, based on data from version 1 of the CBGM. After the painstaking evaluation of problem passages, four witnesses were chosen as the sources: 1739, 617, 307, and 03. At the time, all of these witnesses had a significant number of passages for which the relationship of priority/posteriority to 323 was unclear. After many of these relationships were subsequently decided following further review (iteration) for version 2, however, witnesses 617 and 307 ceased to be rated as ancestors and became relatively distant descendants instead. This means that a new optimal substemma will have to be done for 323. The same may be true for the example in the Presentation as well (witness 35). A former descendant from version 1 graduated to become a higher-ranking witness in version 2 than one of those included in the optimal substemma. This does not guarantee that it will be preferred for the substemma, but it must at least be considered now. And yet another witness with 13 unclear passages is poised to outrank all other witnesses if even one decision is made giving it an additional prior reading. These are optimal substemmata for just two witnesses out of hundreds; they may not all be so volatile, but many undoubtedly are, and it seems impossible to remove the human element from the evaluation processes.  It appears to me that the only practical alternative is to postpone the determination of an optimal substemma for any witness until the mathematical uncertainty of its status as a potential ancestor or descendant is essentially eliminated.

Another problem I see with the creation of optimal substemmata is that of unintended consequences. It is clear from Mink’s descriptions that the process is an intelligent culling of overlap; they do not intend, for example, to simply delete all potential ancestors ranked 11 or lower. Yet I am concerned that even intelligent culling can result in the elimination of ancestors that may prove to be important for connections in cases that are difficult to imagine or anticipate. As we saw in the examples above, there were instances where potential ancestors that were remotely related to their descendants might nevertheless be important. I am also concerned about relationships that depend on a connected change of reading that such a connection may be eliminated by oversight because the witness in question was not included in an optimal substemma. This would be an evil.

Then again, if optimal substemmata were to prove to be an evil in analyses of readings, I wonder if they would really be a necessary evil, in the sense of the drain on computing power that concerns Mink. Thus far it does not appear that optimal substemmata have been implemented; that is, I do not see any missing witnesses in the tables or diagrams of CBGM v.2. Witness 044, for example, is reported to have 103 potential ancestors (and 19 descendants), obviously too many for an optimal substemma. If the application is operating without the benefit of these substemmata, it is certainly running well enough for the current demand. Moreover, computing history suggests that computing technology increases to equal, or more often, outpace demand for processing power. Aside from that, another solution to computing power might be to make the application and the sources available to personal computers so that the application is being accessed one user at a time. I doubt that storage space for witnesses would be a problem (though of course I could be mistaken), and undoubtedly personal computers are fast enough to do the processing.

If computing power is eliminated as an argument for optimal substemmata, I think Mink and his colleagues have already built the necessary options into the application to satisfy anyone who might be intimidated by the volume of witnesses. The tables of potential ancestors cannot be abbreviated, but the user can follow the rankings as desired. For the two coherence queries, the user is free to choose the connectivity level and thereby limit it to as little as single digits if so desired. So then, unless I am missing something, optimal substemmata do not really seem necessary in the long run and are problematic in the short run, as they obviously will require a great amount of time to construct. Human resources could better be used; I would think, doing further iterations with the goal of eliminating uncertainties about readings.

Evaluation of the CBGM and Its Usefulness

This brings us to an overall evaluation of the CBGM. Since I just mentioned uncertainties about readings, let me begin with the evaluation of variant readings, which was the first or second step in constructing the database. I have already noted that this was done supposedly with consideration to both external and internal criteria initially, but my own experience with the CBGM has shown that readings were actually chosen that best met only internal criteria, despite the high regard in which 03 and a few other manuscripts are held. This could be taken as a negative criticism. If anyone argues to the contrary that 03 does very well in the CBGM, it is only to the extent that it agrees with the hypothetical initial text. Experience proves that nothing in the application gives priority to 03 or to any other witness as being older, “better,” or preferable by any other external criterion.

The high agreement between 03 and initial text A (96%) means that conflicts between the two will be uncommon, however. This is less so for 01, the other major Alexandrian manuscript with agreement between it and A at only 90%, but then again agreement between 01 and 03 is only slightly more than 91%, accounting for those split decisions over various readings that have been so problematic for traditionalists like myself. When conflicts do occur, there are workarounds. The most obvious is simply to reject the CBGM preference and choose the traditionally superior manuscripts as having the best reading, especially if it is the harder reading, as is usually the case. This is the normal way of doing “reasoned” eclecticism; however, distasteful thoroughgoing eclectic critics may find it. Another workaround is to select the reading of the preferred manuscripts as the initial reading in the queries, as I did above for 2 Pet. 3:10, and analyze the results. They should be useful since there is general agreement on the internal criteria.

The main criticism that I think can be leveled at the process of evaluating readings, both initially and at iterations, can also be leveled at any other modern system to a greater or lesser extent: the lack of transparency as to how the decisions were reached. We have in the General (Catholic) Epistles of NA28/ECM2, and in the online initial text A found within the CBGM tools, all of the textual decisions made thus far by the editors, i.e. Mink et al. Unfortunately, the only accounting we have for these decisions is the information that Mink has provided in discussing a few examples to illustrate the operation of the CBGM. It may be that the INTF views this information as proprietary and for in-house use only. I would certainly hope not, given our need to know; and if privacy is the issue, then the discussions behind decisions could be reported anonymously. If, on the other hand, the discussions were never documented, then in hindsight the solution is documentation, together with a simple rule that often is ignored: when in doubt, document. Whether working alone or in committees, my experience is that translators, and I am sure text-critical scholars as well, do not like to make notes of how they reached their decisions on readings in the text. Aside from future accountability, it is a chore, a particularly tedious one at that, and no one wants to draw the short straw and be the secretary. It is also easy to justify skipping documentation because 1) standard criteria are used, and therefore theoretically any scholar should be able to deduce how the decisions were made, and 2) doing documentation would seemingly slow the process too much, when time usually is precious.[79] Yet it is frustrating when one really needs to know later how an undocumented decision was made, and it reminds me of one of Murphy’s many corollaries: there is never enough time to do it right, but always time to do it over.

So, if text-critical decisions in the CBGM are undocumented to this point, I would strongly urge a change of practice. At least it is early enough in the process of compiling the initial text that most of the decisions are yet to be made. The UBS and Metzger made laudable efforts toward this goal with their rating system and the Textual Commentary respectively. I would put the ‘A’ to ‘D’ system of the UBS Greek New Testament in the category of “much better than nothing.” Metzger’s Commentary is immensely helpful for all the passages that it covers; one could only have hoped for an exhaustive version of the same. Perhaps something like the Commentary could be done for CBGM passages, i.e. summaries of discussions for difficult decisions, leaving the easier ones to be reverse-engineered by internal criteria.

Taking the evaluation of readings a step further, it obviously would be desirable for the user to have the option of changing decisions in the CBGM, since there are cases in which (as we will note again below) a change or two can affect tables of potential ancestors/descendants. This is analogous to number-crunching in a spreadsheet. I can well believe that such a feature might overwhelm a mainframe serving many users at once, and perhaps it would only be possible if the application and database were installed on personal computers. Along with this, to make the application truly adaptable to all text-critical positions, it might be desirable to have the decisions on variants keyed numerically to criteria, giving the user the option to weight the criteria differently for any or all decisions. For example, if desired, the user could choose to have the easier reading preferred. Or now that the shorter reading guideline has come under intense scrutiny, it could be very helpful to switch preference between the shorter and the longer. Within the option could even be the sub-option to select exceptions.

Of course, these are wish-list items. I think the strongest criticism that can be brought against the CBGM is the arbitrary and volatile quality of the parameters for potential ancestors, which are crucial for the application. It makes sense that a high level of agreement for variant units in any two witnesses indicates a close relationship between the witnesses, while at the same time they have to be different enough so that the differing readings of one will commend themselves as being prior to those of the other. If I may again use extremes to illustrate: if two witnesses disagreed on every variant reading, we would conclude that the scribe of one found absolutely nothing he trusted in the other if that was his exemplar. Instead, he changed everything; very unlikely, bordering on impossible. On the other hand, if two witnesses agreed on every variant reading, one would provide us no more information about their exemplars than the other, and so only one of them would have any value. What we ideally need, then, is an absolute percentage of agreement confirming for practical purposes that one manuscript was copied from another.

It is a little ironic that the tables produced by the Potential Ancestors module do not include the percentage of prior readings among all variant units for witnesses. I would be astounded if I were to learn that the question of a minimum percentage was not debated in CBGM discussions. It would not surprise me at all if there was no agreement for quite some time, however. We know that in the end, the barest possible margin was chosen: just one reading. Moreover, it bears repeating that no weighting was given to the rankings based on the margin of prior readings; it was entirely based on the levels of agreement, and the priority of readings essentially was evaluated as pass/fail parameters. Larger margins serve only to stabilize the ancestral relationship.

Eldon Epp called attention to this issue when he pointed out that the margin of priority of witness 468 (the default witness for the module) over witness 617 amounted to a microscopic 0.13% of their common variation units.[80] As it happens, his figures came from CBGM v.1; in version 2, where two passages moved out of the unclear category and decreased 468’s lead to just two, the difference was an even smaller 0.07%. For all that, in version 2, 15 passages still remain in the “UNCL” category, seemingly guaranteeing that the genealogical relationship between these two witnesses will be volatile for a considerable time to come. In their case, the term “potential” is especially appropriate, and they also happen to be rather conspicuous witnesses in textual flow diagrams.

It is hard to speculate as to whether there is a solution for this problem. It surely strains credibility to posit that a scribe copied from a manuscript rather than to it because it displays just one more prior reading than it does posterior readings relative to its counterpart. If two manuscripts were in complete agreement except for a single variation unit that clearly was prior in one of them, it could more easily be accepted as the ancestor. But the typical coexistence of both kinds of readings in the same witness introduces an undeniable element of uncertainty, which in turn is only increased by the fact that the ECM editors are not infallible in their decisions about priority. So I doubt that many textual critics would disagree with a preference to find at least 20 percent more prior readings as a minimum for “ancestor” witnesses, and perhaps the CBGM position is that beggars can’t be choosers. We all have to work with what we have, not with what we prefer to have. At the least, however, I think it would be desirable to have options in the CBGM both to exclude those witnesses whose number of “UNCL” passages could change their status as ancestor or descendant and to switch their status at will. It would be even better if the option were available to modify the ranking number and direction of descent for any witness and be able to toggle the results of queries of the other modules to reflect the differences. Indeed, if that were possible, it might not be necessary to change any decisions in variation units for the user to obtain results compatible with his or her own methodology.

Let me use an example with which we are already familiar: the variation unit in Jude 15. We saw above in the attestation diagram for reading b that 468 inherited its reading from 03, and was, in turn, the ancestor of almost all the other witnesses. One of the things we did not note is that 468 is also the principle ancestor for reading c by virtue of being the closest relative to 617, which has that reading. It is identical to b except for the addition of a personal pronoun at the end, making it the longer reading of the two and raising the question of whether the pronoun is a scribal addition or was omitted in reading b. Taking the CBGM results for granted with 468 as first ancestor to 617, it seems clear that reading c was a modification of b. Here is the diagram from the Coherence at Variant Passages module:

Now, however, we know (as Epp observed) that there may ultimately be no genealogical relationship between 468 and 617 by CBGM analysis, or that it may even be reversed, which actually seems more likely if there is a change. In either case, A would become the closest ancestor to 617 (A is now second). The arrow from 468 to 617 would either disappear or reverse; if the former, then the diagram would indicate that both readings b and c arose independently from a, as is already the case for b and d/1739. This would create a situation similar to the lack of coherence that led Mink to prefer reading a, and would conceivably call for reconsideration of one of the other readings as the initial. If 617 became the ancestor to 468 and the arrow reversed, then the new diagram and attestation of reading b would indicate that the bulk of b witnesses had 617 as their source. Also, whereas I had reason to conclude previously that c was a derivation of b, I could now conclude that in fact, the analysis provides no such evidence, due to the uncertainty of the genealogical direction between 468 and 617.

You may wonder whether there is any workaround for this problem. I can think of nothing elegant nor simple. For anything that I could not easily visualize, I probably would have to print out the relevant diagram or table and crudely draw or write in changes. We can hope that the necessary options are easy to program and will be considered for the CBGM in the near or foreseeable future. Exclusion is already done for witnesses that have an equal number of prior and posterior readings to the witness being compared. The relevant values seem to apply only to the rankings, so I can see no reason why the flexibility cannot be written into the application spreadsheet-style, other than the extra demand on mainframe processing power.

For now, we can still work with volatile witnesses because future decisions might have no effect on their status; but “what if” scenarios could be very helpful in textual decisions. And I hardly need to repeat–but will nonetheless–that more information would be welcomed about all the textual decisions upon which the CBGM database is founded.

Earlier I remarked that the “A-text” might seem most puzzling to a novice learning about the CBGM. The creation and employment of a hypothetical initial text (the “A-text”) may be objectionable to some. In NA28 it is the text for the General Epistles, and eventually will be the text for the entire New Testament if the current methodology and oversight by the INTF continues. For Greek NT students and scholars who have used the UBS or NA texts for many years, it will simply replace what is already an ultimately hypothetical text, as the base text relative to which all other manuscripts are collated. As a translator, I do not find this troubling because I have never taken the NA (or UBS) text to be the original. It has always represented nothing more or less than the choices of the editors, and in point of fact translators generally are free to choose readings differing from the text. In recent years this has become more evident with the appearance of The Reader’s Greek New Testament, and the SBL’s Greek New Testament (SBLGNT).

It must be granted, of course, that none of these Greek New Testaments exists in complete, continuous form in any extant ancient manuscript, and advocates of the Byzantine text point this out as a fatal deficiency. Yet even Byzantine advocates admit with other textual scholars that many ancient manuscripts have been lost, and we have also seen that the Byzantine text itself does not exist continuously in the prime Byzantine manuscripts. If one were to require that any proposed text be found in an extant manuscript to be viable, then advocates of the Alexandrian text theoretically would just resort to 03 as a witness of the original text, together with other reliable witnesses necessary to fill in its lacunae. There are some advocates of both reasoned eclecticism (including Holmes), and thoroughgoing eclecticism (e.g. Elliott), who require any reading to be in some extant Greek manuscript to be viable; in other words, they reject conjectures. I have always taken the position that we do not have any of the autographs or fragments thereof, or if we do, that we have no way of verifying them as such. But I also take it as a matter of faith–and perhaps more importantly, of practice–that we have the entire New Testament extant piece-by-piece in Greek manuscripts. I never accept a conjecture as the original reading.[81]

I should briefly summarize the CBGM’s handling of text clusters or “types.” There is, above all, no recognition or treatment of traditional text-types or families. Of course the initial text A is recognized, which has more in common with 03 than any other extant manuscript, at 96% agreement. Moreover, we have already noted that a distinction clearly is made between Byzantine and non-Byzantine witnesses. The percentage of agreement with MT (Byzantine) readings is provided in the table of potential ancestors for every witness. Some textual critics have expressed eagerness to see how the concept of text clusters will fare once larger portions of the New Testament are analyzed. We can expect contamination to continue to be a factor, I am sure, but we do not whether the statistics for the rest of the NT will be comparable to what we now have for the General Epistles.

One could question the fairness of distinguishing the Byzantine text from all possible clusters in the database. Given the large number of extant Byzantine manuscripts, the text in the database was taken from a sampling that appears to be reliable and objective. The evidence thus far indicates that Byzantine witnesses show a very high level of agreement. The high percentages of agreement inevitably group them together and separate them from earlier witnesses, and in the CBGM programming, they are not penalized for their late dating or any negative appraisals of their quality. The only prejudice I can find against them, as I have already noted, is the internal criteria for textual decisions that operate mainly to their detriment and cannot be altered by users. Of course, this is not a problem for traditionalists who use standard internal criteria.

As you undoubtedly noticed, in this final evaluation of the CBGM I have thus far focused on the negatives. Yet for a more detailed negative review of the system, I recommend Bengt Alexanderson’s Problems in the New Testament: Old Manuscripts and Papyri, the New Genealogical Method (CBGM) and the Editio Cricia Maior (ECM).[82] A number of his criticisms, some of which I have echoed, are sound; others are debatable. In particular, he takes issue with Mink’s position that minor agreements in variants are just as important as other readings in connecting witnesses.[83] We can all agree that the choice of a significant variant found in a particular witness is connective, at least as a case of contamination if not that of the witness as an exemplar. However, I think it is easy to underestimate the significance of minor differences that create variants, perhaps in part because collations have so neatly sorted out all the variants. Scribes were mostly meticulous in the copying process, so to find the very same variation in the same place, knowing that the scribe was copying word for word, is in my viewpoint significant, favoring Mink’s position.

Turning now to the positive aspects of the CBGM, as I said before, it has brought a new dimension to the process of textual criticism: a true genealogical history of individual readings insofar as is possible to determine from the extant witnesses by internal criteria. Through the use of massive computing power, the application has provided organization, objectivity, and accountability, making textual criticism a little less of an art and a little more of a science. By the latter, I am referring to the fact that the meaning of art is relative to the beholder, while scientific conclusions are verifiable and falsifiable.

We can and should be very grateful for the collations we have had for so many years, but with collations, most of us were mentally channeled into textual decisions (if we made any) based on favorite manuscripts listed in the apparatuses. This is the problem to which I alluded in regard to splits in the Alexandrian manuscripts. Most majuscule manuscripts beyond ‘D’ in the alphabet, and virtually all minuscules were likely to be ignored as too obscure. If we did research them, we would probably learn only their dates and where they were housed.

The CBGM has put all of these manuscripts in their proper places to the extent that this is possible, provided that one agrees with Mink’s four assumptions for scribes. I think the assumptions make sense, and as already noted, the statistics on agreement in readings are objective. We can now see theoretically how the manuscripts relate to one another internally, giving meaning to the collations. I dare say that unless one resorts to traditional external criteria, typically preferring the leading Alexandrian manuscripts and rejecting the Byzantine text (or vice-versa), other collations will now become obsolete next to the ECM, used in conjunction with the CBGM. As a matter of fact, even if collations were able to provide textual flow diagrams or stemmata, they could not begin to include a significant number of manuscripts for readings in comparison with the CBGM. We are probably approaching a time when students and scholars of the Greek NT will no longer carry printed texts, but only electronic note pads (or perhaps just smart phones) either with CBGM online access, or the application and database built in.

With such information comes the accountability I mentioned. It once was enough just to cite the age and superior quality of certain manuscripts, and to point to one or two internal criteria as an explanation of how the preferred reading explains the creation of the others. Whether we realized it or not, when we were doing the latter, we were opening the door to the construction of a genealogy based on internal criteria. It results just from making the same explanation over and over again from one manuscript to the next. As to the age and quality of certain manuscripts, I think these can still be used as overruling criteria, as I’ve said before. However, the CBGM forces us to look at how these manuscripts fit into the genealogies of individual readings and evaluate whether our assumptions and arguments are reasonable. Sometimes, as we have seen, it may be the CBGM that is at fault.

Hopefully, the examples I have provided give the reader some useful exposure with the CBGM, and I strongly encourage the interested reader to go online and use the tools, even if only to experiment with them. As a tool for the traditionalist, either in textual criticism alone or in translation, I can recommend the CBGM as an effective way to trace and evaluate a reading by internal criteria. It might seem that the omission of external criteria is a fatal flaw, but bear in mind that these criteria tend to be significant only when they are used to overrule a decision favoring internal factors, so in that case, one simply rejects the preference of ECM2 and the CBGM. For a thoroughgoing eclectic, the CBGM should be a godsend.

I would conclude with one more criticism, i.e. that the work appears to be going much too slowly to finish the New Testament at an acceptable date (one acceptable to me, at least). The problem seems to be the lack of sufficient human brainpower for such a demanding project, and more people undoubtedly need to be involved in speeding the process. Of course, it may be that there are few at present who are equal to the scholarly demands. Those of us who want to see the project finished can only hope that the situation will improve.

SCROLL THROUGH DIFFERENT CATEGORIES BELOW

BIBLE TRANSLATION AND TEXTUAL CRITICISM

BIBLICAL STUDIES / INTERPRETATION

EARLY CHRISTIANITY

CHRISTIAN APOLOGETIC EVANGELISM

TECHNOLOGY

CHRISTIAN THEOLOGY

CHRISTIAN FICTION

[1] As noted previously, some worthy students and scholars have complained about the subjectivity of the ratings and suggested that they be ignored. I (DW) have always considered them of value, not as objective truth but as a window into the minds of the editors.

[2] Bart D. Ehrman and Michael W. Holmes, eds., “Thoroughgoing Eclecticism in NT Textual Criticism; by J. Keith Elliott,” in The Text of the New Testament in Contemporary Research: Essays on the Status Quaestionis, 2nd ed., Studies and Documents, vol. 46 (Leiden; Boston: Brill, 2013), 752.

[3] Institut für Neutestamentliche Textforschung, ECM2.

[4] The four assumptions are stated in Gerd Mink, “Contamination, Coherence and Coincidence” in Klaus Wachtel and Michael W. Holmes, eds., The Textual History of the Greek New Testament: Changing Views in Contemporary Research, Society of Biblical Literature Text-Critical Studies, no. 8 (Atlanta: Society of Biblical Literature, 2011), 151–55.

[5] Ibid., 153.

[6] Ibid., 189-202.

[7] Available for download at egora.uni-muenster.de/intf/service/downloads_en.shtml.

[8] “Presentation,” 17.

[9] He says the same in his article, “The Coherence-Based Genealogical Method – What is it about?” at egora.uni-muenster.de/intf/projekte/gsm_aus_en.shtml.

[10] This is my own judgment call, based on readings preferred in the General Epistles of NA28 and various examples discussed by Mink, because nothing like Metzger’s Textual Commentary has yet been published documenting the decision processes.

[11] Wachtel and Holmes, Textual History, 202.

[12] Ibid., 142-143.

[13]Ibid., 161.

[14] Metzger, Textual Commentary, 657.

[15] Metzger, 657.

            [16] Textual Variation: Theological and Social Tendencies?: Papers from the Fifth Birmingham Colloquium on the Textual Criticism of the New Testament, ed. D.C. Parker and H.A.G. Houghton, Texts and Studies, vol. 6 (Piscataway, NJ: Gorgias Press, 2008), 123.

[17] Ibid., 124.

[18] While there is logic in the choices of the boundaries of variant readings that span more than one word, the choices are not infallible and would never be identical from one critical NT Greek text to another. For this reason alone it is impossible to state with certainty what the percentage of variant readings is in the NT.

[19] Ibid., 126.

[20] I.e. the hypothetical A-text of the ECM editors, not 02.

[21] The relationship there is even more fragile than it appears, with four variation units “unclear” about the direction of descent.

[22] Mink has a process to weed out any nonessential potential ancestors, with the goal of obtaining an “optimal substemma,” i.e. the bare minimum of ancestors needed to account for all variant passages in a witness. The process involves both a computer application not (yet) available online, and text-critical decision-making that can only be done by the human critic.

[23] I should add that my own experience with CBGM/ECM2 readings thus far has been a greater tendency to reject difficult readings that I might find acceptable.

[24] See above 288.

[25] In a “negative” apparatus, marked by an ellipsis in the a-reading, any witness not mentioned in one of the other readings is to be understood as included in the a-reading; so in these eight prior readings 03 is listed somewhere not with the a-reading, while 01 and 02 are not listed but assumed with “a”.

[26] The same is basically true for yet another witness, 436 (first ancestor A, second 81), whose relationship is neutral because it has an equal number of posterior readings.

[27] In two-word combinations like this Hebrew has an article associated with the first word attached instead to the second word.

[28] Metzger, Textual Commentary, 660-661.

[29] More often the opposite probably is the case, i.e. the choice of the easier reading, typically favoring the author’s style.

[30] Two variant readings in Jude 15 are shared with 623, which ranks as third potential ancestor to 03 in Jude.

            [31] Gerd Mink, “Problems of a highly contaminated tradition: the New Testament

Stemmata of variants as a soιιrce of a genealogy for witnesses,” in Studies in Stemmatology II, ed. Pieter van Reenen, August den Hollander, and Margot van Mulken (Amsterdam; Philadelphia: John Benjamins Pub., 2004), 59.  I suspect that he is correct.

[32] Ibid., 60.

[33] Ibid., 38.

[34] Mink comments that the node “helps to solve a problem which is the result of conflicting data…and ultimately of contamination and loss of intermediary witnesses” (Ibid., 60). I am suggesting that “helps to solve” implies only a partial solution.

[35] Wachtel and Holmes, Textual History, 153.

[36] I use “logically” in the sense that the outcomes in the witnesses can reasonably be interpreted as such, whatever the actual (historical) processes might be.

[37] For all of the General (Catholic) Epistles 81 has 15 potential ancestors, including 03 (rank 12) which shows how different the appraisals can be with different parameters. One must be very careful to make the correct query for the situation at hand.

[38] That is, 03 agrees with A in two more variants than it does with 81, so there had to be some merging of the two witnesses.

          [39] Wachtel and Holmes, Textual History, 155.

[40] Only eight passages are rated as “unclear,” so even with some changes it is doubtful that the direction would change according to the CBGM. 2 Peter has a difference of 12 prior passages favoring witness 81 and 11 unclear, so a change is possible but unlikely. By comparison 1 John has 81 leading by only six passages with an additional six unclear, so a change is plausible though probably unlikely.

            [41] In fact, I suspect that this happened in most cases where there were substantial numbers of ancestors and descendants. 81 itself is an example: 2423 (rank 9) and 020 (rank 10) were two ancestors in v. 1 that became descendants (ranks 17 and 21 respectively) in v. 2. To take a random example, the default witness for the module in both versions is 468. It has four full ancestors in v. 1 and seven in v. 2; one descendant at rank 78 and another at rank 82 in v. 1 became ancestors respectively at ranks 3 and 4 in v. 2!

[42] This is indicated by a dash in the “D” (for “direction”) column of potential ancestors/descendants tables; it also is given a rank of “0” (number rank column).

[43] It would be helpful to have the option to search all Potential Ancestors lists for witnesses in the first line (highest percentage) or so at zero rank or “–” direction; but no doubt there are a number of other needs with higher priority.

[44] Such a routine, as I imagine it, would have a value quantifying the connectivity of the prior reading (based on internal criteria) and substituting it in a reasonable way for the zero genealogical ranking.

[45] Problems, 57. The formula is this difference x 100 / the number of passages where comparison is possible.

[46] As seen earlier, for any witness ranking two or lower (i.e. a higher number), the ranking is included following a slash after the manuscript number in CBGM diagrams. I am assuming that a suffix “/0” is meaningless in the program.

[47] ECM2 Part 1, 22* (Introduction).

[48] Ibid., 21*-22*.

[49] To mention a few verses where this occurs: James 1:1, 3; 2:14; 4:8; 1 pet. 1:4, 7; 2:18; 3:14; 5:2; 1 John 1:5; 2:7; 3:1; 4:14; 2 John 5, 9; 3 John 4, 10; Jude 2, 9.

[50] ECM2, Part 1, 22*; italics are mine.

[51] See the discussion of “optimal stubstemmata” below, 360 ff.

[52] ECM2, Part 2, Supplementary Material, 10, repeated on following pages for each epistle.

[53] For James, the editors tolerate up to two of the manuscripts disagreeing with the group, allowing the remainder to represent the “undivided” Byzantine witness, whose reading is then noted as Byz in the apparatus. For remaining epistles, only one manuscript opposing the group is tolerated for an undivided witness.

[54] ECM2, Part 2, 10 ff.

[55] ECM2, Part 2, 10.

[56] In all fairness I note that the Byz text as a subset of the MT is my own characterization, but logically it seems undeniable given that the MT is described as readings supported by the majority of manuscripts “whether or not they agree with the established text,” while those that differ further comprise the Byz text in the ECM.

[57] ECM2, Part 1, 22*.

[58] This reminds me of the CBGM practice of deciding Potential Ancestors by a simple majority of primary readings.

[59] ECM2, Part 1, 21*.

[60] I.e. The New Testament in the Original Greek: Byzantine Textform, 2005, compiled and arranged by Maurice A. Robinson and William G. Pierpont (Southborough, Mass.: Chilton Book Pub., 2005); hereafter cited as “RP” (Robinson-Pierpont).

[61] Ibid.

[62] See http://bibletranslation.ws/down/Robinson-Pierpont-nocopy.pdf to download RP, and http://www.bgnt.net for BGNT.

[63] RP, ix.

[64] BGNT, vii.

[65] Ibid., vi-vii.

[66] It is noteworthy that the a-reading in this case is the Byzantine, and that 01 and 03 have the anarthrous form which was preferred in NA26/27. What may have tilted the decision toward the Greek article in NA28 is its additional support from P¹⁰⁰.

[67] See, e.g., RP, 536 ff., etc.

[68] See Ibid., 546-7. Robinson states first that such a reading must be found in the majority of witnesses, and then verified to be the more difficult and the original reading. In saying the latter, he hints that one should or can be suspicious as to whether the reading really is the more difficult in the context.

[69] Ehrman and Holmes, “Thoroughgoing Eclecticismm; by J. Keith Elliott,” 746 f.

[70] Wachtel and Holmes, Textual History, 181-89.

[71] Ibid., 200-201.

[72] “Presentation,” 566-71.

[73] See above p. 46.

[74] I.e. five witnesses that changed from the a reading found in ancestors 424, 2423, and 642 instead of inheriting b from other ancestors. Such anomalies can usually be eliminated (as in this case) by setting connectivity to “absolute” (i.e. 499). When that is done, however, one typically finds untenable dependencies on distant ancestors. In this case, the closest ancestor for one of the witnesses is 22 generations removed, and the most distant, 40. “Average” connectivity, it should be recalled, is up to 10 generations.

[75] Readings like this certainly get the attention of traditionalists as arguments for an “early” Byzantine text. On the other hand, finding the harder reading in Byzantine witnesses is the exception rather than the rule, so what would really be needed to support the text is proof of early “easier” readings.

[76] I should also emphasize that this result is consistent with what Mink has promised for the CBGM. Among his final remarks in his CBGM Presentation (p. 576), he explicitly states that the CBGM does not make textual decisions; it only finds structures within the tradition and enables users to see the consequences of what they are doing. On the other hand, I read this mainly as a disclaimer; CBGM analyses certainly can lead users to decisions, just as the analyses from version 1 led Mink and his colleagues to the changes seen in CBGM v. 2.

[77] A check of the rankings of relationships in the attestation modules shows that they are based exclusively on the potential ancestors for all the epistles (collectively), and cannot be set for rankings based on individual books.

[78] See Wachtel and Holmes, Textual History, 206 n. 106.

[79] I can almost hear someone suggesting that the solution is to hire a secretary with the skills of a court reporter. However, someone who thoroughly understands the issues and the dynamics of the decision process is also needed to make useful notes; in short, a committee member, like Metzger for the previous UBS (Textual Commentary).

[80] Ehrman and Holmes, “Textual Clusters; by Epp,” 551.

[81] This includes the NA28/GNT5 reading “not found” in 2 Pet. 3:10, but I believe I provided adequate text-critical reasons for rejecting it earlier. Otherwise, I would give it more credit than most conjectures because of the support it finds in important versions.

          [82] Bengt Alexanderson, Problems in the New Testament: Old Manuscripts and Papyri, the New Genealogical Method (CBGM) and the Editio Critica Maior (ECM), Acta Regiae Societatis Scientiarum et Litterarum Gothoburgensis. Humaniora (2014).

[83] Ibid., 63-64.