Manuscripts Give Readings, Not Rankings

Article 6 of 27 in the series “When the Canons Collide.”

The phrase manuscript evidence can hide two different things. A codex may directly show that a scribe wrote one form at a passage. It does not directly announce the date of the reading, the quality of its ancestor, its independence from other witnesses, or the amount of authority its testimony deserves. Those are historical conclusions. They may be disciplined and strongly supported, but they are not ink on the page. This article separates the artifact from the rank assigned to it and tests the main scales used in external evidence.

Controlling question: What does a manuscript state directly, and what must a critic infer before that statement can be given weight?

Manuscripts Give Readings, Not Rankings

The phrase “follow the manuscripts” sounds admirably objective. It is also incomplete. Manuscripts disagree, and a manuscript does not state how much authority it deserves. Before its reading can control a decision, the critic must answer at least six questions:

1. What exactly does the manuscript read, including corrections, erasures, abbreviations, punctuation, and uncertain letters?
2. When and where was the physical witness produced?
3. Is its reading independent or copied from an ancestor represented elsewhere?
4. What kind of text does it carry in this series or passage?
5. What habits characterize the relevant scribe and corrector?
6. How does its reading relate causally to the alternatives?

These are learned questions. They require photographs, transcriptions, catalogues, collation, paleography, codicology, versional knowledge, patristics, and historical judgment. The resulting expertise deserves respect. But respect for expertise is not the same as pretending that expertise removes interpretation.

Age: Three Different Dates

“Older is better” compresses three dates that should be distinguished.

The artifact date is when the surviving object was copied. The exemplar date is when its immediate or remote source was copied. The reading date is when a particular wording entered the transmission. Only the first is tied directly to the surviving page, and even it may be a range.

A fourth-century codex can preserve a second-century reading or a fourth-century error. A tenth-century codex can preserve an ancient reading through a careful line of descent. Therefore, the physical age of a witness gives a real but limited reason. It reduces the number of copying generations possible before the artifact; it does not disclose what occurred in those generations.

This is why modern critics do not simply select the oldest manuscript. Metzger qualifies age by the character of the witness, the care of its scribe, its relationships, and the distribution of support. The qualification is correct. It also means that “early” is one weighted sign among others, not an automatic verdict.

Number: Counting Copies or Counting Streams?

The objection to raw majority counting is easy to understand. If nine manuscripts descend directly from one erroneous exemplar and a tenth preserves an independent line, the tally is nine to one while the genealogical evidence is one branch to one branch. Multiplication of copies does not multiply independent origins.

Yet the slogan “manuscripts are to be weighed, not counted” can become an escape from evidence if the weighing is opaque. Numerical prevalence can bear on continuity, diffusion, and ecclesial use. A reading found across a broad, stable transmission is a historical fact requiring explanation. Genealogical dependence can reduce the force of a tally, but it must be shown rather than assumed from a label such as “Byzantine.”

There are therefore two opposite reductionisms:

• arithmetical reductionism: the most copies must preserve the original;
• genealogical reductionism: a widespread tradition counts as one late witness because its members resemble one another.

A sound inquiry asks how a reading became numerous, how early its branches can be traced, whether its support is geographically and linguistically diverse, whether it appears in lectionary and patristic use, and whether competing readings show their own genealogical concentration. Number is neither sovereign nor nothing.

Variety: Independence Must Be Demonstrated

Geographical variety is often treated as evidence of antiquity. If a reading appears in Greek witnesses, early versions, and fathers across regions, independent circulation seems more likely. This inference can be strong. It can also be overstated.

Versions may have been revised toward Greek exemplars. Fathers may quote from memory, adapt wording, or be preserved in manuscripts later conformed to a familiar text. Labels such as “Alexandrian,” “Western,” and “Caesarean” may conceal internal diversity or uncertain borders. A reading's presence in two languages is not automatically two independent testimonies.

The solution is not to discard versions and fathers. It is to state the inferential chain:

1. the citation or version probably reflects this Greek reading;
2. its form is textually secure;
3. it is chronologically relevant;
4. it is not plausibly dependent on the same immediate source as the Greek witness being counted;
5. therefore it adds some independent evidence.

Each step can be investigated. None should be silently assumed.

Textual Character and the Problem of Training Cases

Critics properly judge a manuscript by its behavior across many readings, not by one famous verse. But how is “behavior” measured? One common procedure identifies clear cases, observes which witnesses preserve the reading judged original, and uses the result to assess those witnesses in harder cases.

This can be legitimate calibration if the training cases are genuinely independent and clear. It can also create circular reinforcement:

1. readings are judged original partly because favored witnesses contain them;
2. witnesses are judged superior because they contain those readings;
3. the improved ranking then decides disputed readings in favor of the same witnesses.

The circle is not broken by the size of the dataset alone. Ten thousand mutually dependent judgments can produce a precise feedback loop. The key question is whether some evidence enters independently: securely dated artifacts, demonstrable copying relations, unambiguous mechanical errors, or uncontested readings whose direction is constrained.

The same warning applies to authorial style. A profile of “Paul's style” built from a reconstructed Pauline text can later be used to reject a received reading as un-Pauline. This may be informative where the profile rests overwhelmingly on stable passages. It is circular where disputed features created the profile and the profile then decides those features.

Local Versus Global Quality

No manuscript is uniformly good or bad. A careful scribe can make a local error. A generally erratic witness can preserve an ancient reading. Modern eclecticism rightly judges variation units locally rather than following one codex throughout.

But local judgment creates another tension. If global character never determines a reading, the witness's reputation does little work. If global character often determines a reading, the process risks circular training. Responsible practice must explain when global pattern is allowed to create a prior probability and how strong that prior is relative to local evidence.

This can be expressed in ordinary Bayesian terms without pretending to assign exact numbers. A reliable witness may raise the prior credibility of its reading. A clear local mechanism can defeat that prior. The question is how “reliable” and “clear” were established. The labels cannot be allowed to settle the very evidence from which they were derived.

What This Article Establishes

• The date of an artifact, the date of its exemplar, and the date of a reading are distinct.
• Counting manuscripts and counting independent streams of transmission answer different questions.
• A witness's textual character can become circular when the readings used to train the profile are also chosen by that profile.

Conclusion

External evidence is real evidence, but its weight is never printed in the margin of the artifact. Dating, grouping, independence, and character all require interpretation. The honest critic therefore distinguishes direct report from inferred rank and states how each inference could be tested. That candor does not weaken scholarship. It keeps scholarship from speaking with an authority that belongs only to the evidence or to the Word itself.

Previous: The Canons Have No Umpire: Dabney Vindicated and Refined · Next: The CBGM and the Claim of the ‘Best Manuscripts’