The ECM, the CBGM, and the Reconstruction of an Initial Text

Revision note: Substantially revised July 20, 2026. The argument has been reordered according to the classical rhetorical movement, compressed where necessary, and given one consolidated Notes and Further Reading section.

The Method behind the New Main Text

NA28, UBS6, and the forthcoming NA29 are hand editions. Their compact pages cannot display the full body of evidence or the complete reasoning behind every adopted reading. In the books revised during the Editio Critica Maior project, the hand editions receive their main-text decisions from a much larger scholarly enterprise.

That enterprise is usually named with two acronyms: ECM and CBGM. They are related but not identical. The Editio Critica Maior is a major critical edition. It documents the Greek textual history of the first millennium through selected Greek manuscripts, early translations, and quotations in ancient Christian writers. It defines variation units, presents attestations, evaluates readings, and prints an editorial reconstruction.¹

The Coherence-Based Genealogical Method is a set of procedures and computer-assisted tools used within that work. It analyzes patterns of agreement and the directional relations proposed among readings and witnesses. It helps editors test genealogical hypotheses in a textual tradition too contaminated for a simple manuscript family tree.²

The distinction matters because the CBGM does not push a button and produce the ECM text. Gerd Mink, who developed the method, states explicitly that it does not make textual decisions. Human editors define readings, construct and revise local stemmata, assess internal evidence, decide how variants relate, and finally determine what should be printed.³

Any critique that calls the newest Greek text merely “computer generated” is therefore false. Any defense that treats the method’s graphs as neutral facts independent of editorial judgment is equally false. The Nestle-Aland apparatus is extraordinarily dense, but it is still a hand apparatus. It cannot fully document every relevant witness, variation unit, versional form, patristic citation, and genealogical relation.

The ECM attempts a larger task. According to the Institute for New Testament Textual Research, it documents the Greek textual history of the first millennium using Greek manuscripts especially relevant to the tradition, older translations, and New Testament quotations in ancient Christian literature. Its manuscript selection rests upon prior analysis of the broader primary tradition.⁴

Its purpose is not merely to place more sigla at the bottom of a page. It seeks to reconstruct the text afresh on the basis of evidence assembled with a degree of completeness unavailable to earlier hand editions. The work proceeds book by book or corpus by corpus. Results already published for the Catholic Epistles influenced NA28 and UBS5. Results for Acts, Mark, and Revelation have now entered UBS6 and the intended NA29 main text. Work on other New Testament corpora continues.

The current critical text is therefore layered. In some books it reflects completed ECM work; in others it continues an earlier editorial state. The method applied to “the New Testament” is not yet uniform across all twenty-seven books. Classical stemmatics often imagines a branching family tree: manuscript A is copied from manuscript B, B from C, and so on. If the relationships can be established, readings may be evaluated through the genealogy of the documents.

The New Testament tradition resists so simple a model. Scribes could consult more than one exemplar. A manuscript can contain readings derived from different textual streams. Corrections may introduce readings from another source. The result is contamination—mixture across lines of transmission. Contamination means that one manuscript may preserve an older reading at one variation unit and a younger reading at another. A late manuscript may carry an ancient reading. An early manuscript may contain a secondary change. No manuscript receives a single global rank capable of deciding every local problem.

Mink also emphasizes a second complication: the same reading can arise independently more than once. A short omission, harmonization, word-order change, or grammatical improvement may be produced coincidentally by unrelated scribes. Agreement does not always prove direct genealogical dependence.⁵ The CBGM was designed to analyze coherence under these conditions.

In CBGM terminology, the operative witness is not merely the physical manuscript as an archaeological object. It is the state of text conveyed by that manuscript. The textual state may be older than the parchment or paper on which it survives. This distinction prevents a crude appeal to manuscript date. A twelfth-century manuscript does not thereby contain a twelfth-century reading at every location. Its exemplar history may reach much earlier. Conversely, an early codex does not become uniformly original by age alone.

The method therefore compares the textual profiles transmitted by witnesses. Physical date remains historical evidence, but it is not converted into a universal voting weight. That is a genuine advance over popular descriptions in which “the oldest manuscripts” function as a magic phrase. The ECM’s decisions may prefer or reject readings found in early manuscripts depending upon the entire local and genealogical case.

From Variation Unit to Local Stemma

The method begins locally.

A variation unit identifies a place where the witnesses differ. Each distinct form must be transcribed and classified. Editors then evaluate how the variants may be related. A local stemma represents a hypothesis about which reading could have given rise to another. This step is irreducibly philological. Grammar, scribal habits, context, intrinsic probability, transcriptional probability, and knowledge of the tradition inform the proposed directions. The computer does not discover that reading a generated reading b by raw counting.

Local stemmata may contain:

• a reading judged prior to another;
• multiple possible descendants;
• relations that remain undirected;
• independent emergence of the same form; or
• an open question represented by competing top-line readings.

Because the local stemma supplies direction to the later coherence analysis, an error at this stage can influence the patterns generated afterward. The CBGM responds through iteration: editors compare local decisions with global coherence, revisit implausible relations, and revise the stemmata. This is not a defect accidentally hidden in the method. It is the method’s stated structure.

The Three Coherences

Mink distinguishes three relevant forms of coherence.⁶

Pre-genealogical coherence

Pre-genealogical coherence measures agreement between witnesses before directional genealogy is imposed. If two textual witnesses agree at a high percentage of the sampled variation units, there is a strong coherence between their textual profiles. This information can show where a genealogical relation is plausible and, just as importantly, where it is unlikely. Weak coherence between witnesses supporting the same variant may suggest that the variant arose more than once. Pre-genealogical coherence is comparatively objective because it rests on collated agreement. Yet the result still depends upon the selected witnesses, defined variation units, transcriptions, and treatment of lacunae.

Genealogical coherence

Genealogical coherence adds the directional information supplied by local stemmata. It asks whether the relationships proposed among variants cohere with the broader textual flow between witnesses. The method can then identify potential ancestors for a witness—not necessarily direct physical exemplars, but surviving textual states that stand earlier in the predominant flow and could explain portions of the descendant’s text. These relations are displayed in textual-flow diagrams. The arrows do not mean “this manuscript was copied directly from that manuscript.” They represent predominant genealogical relations between textual states under the method’s assumptions.

Stemmatic coherence

Stemmatic coherence concerns the construction of an optimal substemma: a smallest sufficient combination of potential ancestors capable of explaining a witness, including its mixture from different sources. This is more complex than placing each manuscript on one branch. The method seeks structures that account for contamination without being falsified at the variation units under review.

What the CBGM Actually Contributes

The CBGM contributes several kinds of information that ordinary manuscript counting cannot supply.

The method is evidentially powerful because it makes editors test local judgments against a large network of agreements. It is not self-authenticating because the network is partly structured by those local judgments. Popular argument often reduces textual criticism to arithmetic: if more manuscripts contain one reading, that reading wins; if the oldest two contain another, the older pair wins. The CBGM follows neither rule.

A large attestation can reflect the wide diffusion of a secondary reading. A small attestation can preserve an earlier form. But neither conclusion follows from size alone. The editors ask how the reading could have arisen, how its supporting witnesses cohere, whether the same form may have emerged repeatedly, and how the local relation fits the broader textual flow.

This means that a Byzantine reading can be accepted without adopting “majority rules,” and a reading in an early papyrus can be rejected without denying the value of early evidence. The object of judgment remains the variant at the local unit. For confessional criticism, this is both a warning and an opportunity. It is a warning against refuting the ECM as though it merely counts old Alexandrian manuscripts. It is an opportunity because the method’s rejection of fixed text-type supremacy may expose evidence formerly discounted by older critical theories. The received-text case must engage the actual local stemmata and attestations rather than a nineteenth-century caricature.

Editorial judgment enters before, during, and after the computer-assisted analysis.

Before analysis, scholars select the witnesses, establish transcriptions, define the boundaries of variation units, and decide which differences count as distinct readings. During analysis, they construct local stemmata and assess whether one reading could generate another. After analysis, they interpret coherence, revise relations, decide whether the evidence supports one initial reading, and determine how uncertainty will be printed. This distributed judgment explains why the phrase “the algorithm says” is misleading. The software can calculate agreement and display consequences with a consistency impossible by hand. It cannot decide what counts as Scripture independently of the editorial model and philological judgments supplied to it.

What “Initial Text” Means

The INTF describes the goal as reconstruction of the initial text—the text standing at the beginning of the manuscript tradition.⁷ The online CBGM suite similarly calls it a hypothesis about the text from which the manuscript transmission began. That wording must be preserved.

The initial text is not a surviving manuscript. It is a reconstructed editorial hypothesis. Nor should it be silently equated with the authorial autograph in every book. The beginning of the recoverable manuscript tradition and the moment of original composition may coincide, but the method as such reconstructs the former. Questions of literary composition, authorial revision, early editorial activity, and the gap before surviving evidence require arguments beyond a coherence graph.

This distinction is especially important for theology. Westminster Confession 1.8 speaks of the Old Testament in Hebrew and the New Testament in Greek as immediately inspired and kept pure. “Initial text” is a modern technical target. Whether that target successfully identifies the inspired wording is a theological and historical conclusion, not part of the definition of the method.

The two categories may be argued to coincide. They may not simply be treated as synonyms. A responsible critique must grant at least six strengths. First, the ECM works with a far fuller evidentiary base than a hand edition can display. Second, the CBGM takes contamination seriously rather than forcing the tradition into a simple family tree. Third, it refuses to assign a manuscript one universal value for every variation unit. Fourth, it can reveal when an apparently isolated reading has strong genealogical coherence or when a widely distributed reading may have arisen repeatedly.

Fifth, its data and many of its tools are available for examination through the INTF’s digital platforms. Other scholars can inspect transcriptions, attestations, and diagrams rather than accept a bare committee claim. Sixth, the method preserves open decisions. Split guiding lines and diamond-marked units acknowledge that the evidence does not always yield one confidently preferred reading. These strengths explain why the ECM has changed readings in both directions, sometimes accepting forms associated with the Byzantine tradition and sometimes rejecting long-standing critical readings.

Limitations That Must Also Be Granted

The limitations are equally real.

The surviving evidence is incomplete

Most early textual states are lost. Mink states that detailed historical reconstruction is impossible. The method offers a genealogical hypothesis about available states, not a recovered chain of every copying event.⁸

The local stemmata involve editorial judgment

The direction between readings is not mechanically observed. It is argued. Coherence can confirm or pressure that argument, but it cannot remove its human premise.

The output depends upon the input corpus

Witness selection, transcription, variant segmentation, lacuna handling, and versional interpretation affect the analysis. Better data can change the result.

The ECM is incomplete across the canon

The newest hand text combines ECM-revised books with books still reflecting an older state. One cannot describe the full UBS6/NA29 text as the product of a completed whole-New-Testament CBGM analysis.

Genealogy does not establish canonicity or providence

The method can describe relations in transmission. It cannot tell the church what God promised concerning preservation, how ecclesiastical reception functions, or whether an editor’s reconstructed initial text is the providentially kept text confessed by the Reformed churches.

Iteration is not independence

Because local stemmata inform global coherence and global coherence informs revision of local stemmata, the procedure is iterative rather than composed of wholly independent tests. That does not make it viciously circular. It does mean that “the computer confirmed the reading” is an inadequate description. The confirmation occurs within an editorial model whose assumptions remain open to review.

The Strongest Critical-Text Case

The strongest case says that no edition can responsibly ignore the complexity of the evidence. The New Testament tradition is highly contaminated; many witnesses are lost; and simple manuscript counting or text-type preference cannot reconstruct its history. The ECM provides the most extensive edition yet attempted, while the CBGM enables patterns to be tested at a scale no unaided editor could manage.

The initial text is appropriately called a hypothesis because scholarly humility forbids claiming direct possession of the autographs. That humility does not imply skepticism about the New Testament as a whole. Broad coherence across the tradition remains remarkable, and the method operates within that substantial agreement. This is a serious argument. It deserves a serious confessional reply.

The Confessional Question

The confessional question is not whether computers may be used, whether textual witnesses may be collated, or whether genealogical evidence may inform ministerial judgment. All may be useful. The question is whether the church’s authoritative text is being recognized within a providentially preserved transmission or reconstructed as the provisional output of an open-ended scholarly model. The same tool can serve different doctrines depending upon the rule governing its use.

If Scripture’s identity is allowed to remain indefinitely suspended upon the latest reconstruction of an initial text, the method functions magisterially. If the evidence is used ministerially under a prior scriptural and confessional account of preservation, the method may illuminate relationships without becoming the ground of the text’s authority. This series will therefore examine ECM evidence. It will not allow ECM terminology to answer the theological question in advance.

Rules for Using ECM and CBGM Evidence

1. Cite the exact ECM volume, book, and variation unit.
2. Transcribe every relevant reading and attestation before discussing coherence.
3. Identify whether the ECM prints a single guiding line or a split line.
4. Distinguish pre-genealogical agreement from genealogical direction.
5. State the local-stemma premise on which a flow diagram depends.
6. Do not describe a potential ancestor as a direct exemplar.
7. Do not call the CBGM a computer-generated textual decision.
8. Do not equate “initial text” and “autograph” without argument.
9. Record whether the book has completed ECM coverage.
10. Compare the received-text construction without exempting it from historical evidence.

These rules permit the method to be examined at full strength.

Verdict

It is established that the ECM, not the compact hand edition alone, supplies the fuller evidence and editorial work behind the newest changes in the Catholic Epistles, Mark, Acts, and Revelation. It is established that the CBGM is a computer-assisted, iterative method for testing coherence and genealogical hypotheses within a contaminated tradition. It does not make textual decisions automatically.

It is also established that the method reconstructs an initial text as a scholarly hypothesis about the beginning of the manuscript tradition. The identification of that hypothesis with the immediately inspired and providentially preserved text requires a separate theological argument. No doctrinal derivation grade is assigned because this article defines the method controlling later textual judgments. The operational verdict is adopted: ECM and CBGM evidence will be used, but every human premise, open decision, and corpus limitation must remain visible.

Notes and Further Reading