New Digital Tools for a New Critical Edition of the Hebrew Bible

Abstract This article describes the digital edition of the Hebrew Bible: A Critical Edition (HBCE), which is being produced as part of a project called Critical Editions for Digital Analysis and Research (CEDAR) at the University of Chicago. We first discuss the goals of the HBCE and its requirements for a digital edition. We then turn to the CEDAR project and the advances it offers, both theoretical and technological. Finally, we present an illustration of how a reader might use the digital HBCE to interact with the biblical text in innovative ways.


Introduction
Critical Editions for Digital Analysis and Research (CEDAR) is a new platform for textual research, currently under development at the University of Chicago, that will host the digital version of The Hebrew Bible: A Critical Edition (HBCE).1 Despite the great expansion of our textual data since the earliest works of biblical text-criticism, there have been almost no conceptual advances in the tools available for studying that data.2 Word processors have made publishing easier, and digital imaging techniques can produce clearer and more detailed photographs, but for the most part scholars are still working with flat, static text and flat, static images.3 The tools offered by CEDAR, made possible by a new approach to encoding texts, thus represent a significant development in the history of textual criticism.
In this article, we first discuss the goals of the HBCE and its requirements for a digital edition. We then turn to the CEDAR project and the advances, both theoretical and technological, that make it ideally suited for the digital HBCE. Finally, we present an illustration of how a reader might use the digital HBCE to interact with the biblical text in innovative ways.

The goals and requirements of the HBCE
The HBCE is the first eclectic edition of the Hebrew Bible to be produced.4 The project aims "to approximate the corrected archetype of each biblical book."5 By "archetype," the editors mean the latest common ancestor of all the variant manuscripts. This ancestor can be inferred, with due caution and epistemological humility, from the textual witnesses of those manuscripts. The archetype -the latest common ancestor -differs from the original in what we can know about it: the archetype can be inferred from the available data; the original cannot. Further, the archetype may represent a different textual state from the original, although we cannot know with certainty because we have no access to the original. The archetype, and not the original, is the target of the HBCE. The HBCE intends to approximate the corrected archetype because the archetype "will contain scribal errors that can sometimes be detected and corrected. For instance, distinctive scribal errors that are in all the manuscripts derive from the archetype. If we can discern these errors, we are obliged as textual critics to correct them."6 Finally, the editors aim to approximate the corrected archetype because they recognize the potential fallibility of their judgments as well as the methodological impossibility of verifying the reconstruction of a hypothetical text.
Each volume of HBCE will be offered in both print and digital versions. The digital version will be an interactive polyglot, comprising the HBCE critical text and commentary as well as the evidence used to construct the critical edition.7 The digital version will thus allow scholars to engage with the textual data in ways not possible in print. Instead of being limited by a critical apparatus showing the variants deemed salient by an editor, scholars will choose which manuscripts they want to compare and which kinds of variants they want to see. They will also have access to full transcriptions and images of those manuscripts so that variants can be understood in context. In addition, scholars will be able to perform their own textcritical comparisons of manuscripts, both within and across languages.
The digital HBCE will be produced as part of a new project called Critical Editions for Digital Analysis and Research, or CEDAR. Based in the Oriental Institute at the University of Chicago, CEDAR is a collaboration between computer scientists and textual scholars. The project has created new ways of encoding manuscripts that enable deeper and more comprehensive study of texts, their histories, and their relationships.8 Although this article focuses on CEDAR's application to biblical studies, the tools we are developing are generic enough to be used for any textual corpus.9 CEDAR is built within a high-performance XML graph database called the Online Cultural and Historical Research Environment, or OCHRE.10 OCHRE was originally developed to record and manage archaeological data, but it now houses over fifty projects across a wide range of disciplines in the humanities, social sciences, and hard sciences. For over a decade, OCHRE has been tested by philologists recording texts with highly complex writing systems. Together, some 25,000 texts have been entered in the database, in languages as varied as Elamite, Assyrian, Egyptian, Greek, Ugaritic, Aramaic, and Demotic. CEDAR is the first text-critical project in OCHRE, but the database has several features that make it well-suited for this kind of work. In particular, OCHRE enables multiple overlapping hierarchies by using an item-based ontology in which representations of hierarchical structures are stored as separate items from the data in those hierarchies. The OCHRE data model is also highly granular, allowing scholars to atomize their data into the most minimal meaningful parts for analysis and manipulation. Finally, implementing CEDAR in OCHRE allows scholars to have a great deal of flexibility both in what data they choose to capture and in how they describe the relationships among their data. These features allow for significant conceptual and computational advances in representing and interacting with digital texts.11

Conceptual advances in CEDAR
CEDAR implements a new approach to encoding text that might be called the database model.12 In the CEDAR project, although we are creating tools to represent and study texts, there are no "texts" per se stored in the database. That is to say, the database does not store linear strings of characters comprising whole compositions. Instead, each verse, each line, each word, each individual character, and even each individual diacritical mark is stored as its own discrete XML document. The various hierarchies in which they can be arranged are stored individually as separate XML documents. When a user asks to view a text, OCHRE accesses the appropriate hierarchies, gathers the database items associated with those hierarchies, assembles these items into words and whatever other structure is defined by the hierarchies (e.g., lines/ columns/pages, verses/chapters/books), and displays an organized, formatted view of the text in the linear strings of characters that readers are accustomed to seeing. There is a distinction, in other words, between how the user reads the data and how the data is organized in the database.
This distinction is not maintained in the dominant approach to encoding texts in the humanities, which might be called the document model.13 The document model both displays and organizes data in the same pre-digital structure of physical documents, in which characters are assigned a fixed position in one or two dimensions. As Schloen and Schloen note, some kinds of scholarly work can be done using this method, but it imposes unnecessary limits if it is used as the primary means of representing a text. It deprives scholars of the power to express in digital form many of the conceptual distinctions they employ routinely in the course of their work. They end up imitating the position-dependent structure of a predigital medium rather than exploiting the potential of the digital medium to represent their information in a more effective way. As a result, they fail to capture in an explicit, searchable form the different ways a given text has been read and annotated, making it difficult to use computational methods to compare and analyze the various interpretations. And it is precisely those texts which are open to many different readings that are of greatest interest to scholars in the humanities, who spend considerable time tracing the history of textual interpretation and the interconnections within and among texts.14 Digitization of texts should offer scholars new and more effective ways of doing their work, but the document model fails to take advantage of that potential because it reproduces the design of print media. The database model, on the other hand, can capture multiple interpretations and versions of a given text, which are precisely what scholars in the humanities are often most interested in studying.
11 Another widely used approach is to encode texts in XML documents marked up according to the standards of the Text Encoding Initiative (TEI); see http://www.tei-c.org/ [accessed 20 December 2018]; Cummings, "The Text Encoding Initiative and the Study of Literature." XML is sometimes mistaken to be synonymous with TEI, but they are in fact distinct. XML is a markup language, while TEI is a set of guidelines for standardizing markup tags within that language. OCHRE uses XML but not TEI. The research questions we want to ask in the CEDAR project require a highly atomized, normalized data model which TEI is not intended to provide. TEI remains a useful way to capture many aspects of transcribed texts, however, and data can be exported from OCHRE in TEI format. 12 The database model and the document model (see below) were first described in Schloen and Schloen, "Beyond Gutenberg." 13 Ibid. 14 Ibid., 4.
The database model is especially powerful for textual criticism because the same database items can be reused in different versions of the same text. In OCHRE, each attested character and diacritical mark is encoded as a separate database item. For example, in bərēʾšît, the first word of Genesis in Hebrew, the bêt is an XML document, the dāgēš is an XML document, the šəwāʾ is an XML document, and so on (see Figure 1). Combinations of these items are then assigned to hierarchies that represent individual manuscripts. When the user asks to view the various manuscripts of Genesis, OCHRE generates them on-the-fly by gathering and arranging the same underlying database items according to each manuscript's hierarchies. The user sees separate texts, but within the database, each "text" is simply a recombination of the same building blocks.

Figure 1: The reuse of database items in multiple contexts
This structure means that instead of existing in one or two dimensions, as in the document model, texts in the database model exist in three dimensions. Rather than being constrained to setting them side by side for comparison, we can stack one on top of the other. Texts become like transparency sheets, and we can compare texts by layering one on top of the other to see where they line up. To add another text to the database, we simply assign the appropriate alreadyexisting characters to another set of hierarchies; the new text is instantly related to other texts already in the database. By contrast, a comparison of multiple manuscripts in the document model would require the creation of many duplicate character strings with links between them. Instead of one single bêt reused in every instance of bərēʾšît in every manuscript, as we have in the database model, we would have as many bêt's as we have manuscripts, with a giant web of links to relate them.

Computational strategies in CEDAR
We employ a number of new computational strategies that result in the conceptual advances discussed above. First, CEDAR makes use of OCHRE's ability to encode overlapping hierarchies. Textual data are organized into two overlapping hierarchies: an epigraphic hierarchy, which is arranged by material and graphical features such as pages, columns, lines, and characters; and a discourse hierarchy, which is arranged by discursive features such as sentences, phrases, and words (or, for a biblical text, books, chapters, verses, and words).15 Each word in the discourse hierarchy is comprised of characters from the related epigraphic hierarchy. Second, texts in CEDAR are highly atomized, as evident in Figure 4. Each character and diacritical mark is a separate database item -more specifically, a separate XML document -with its own universally unique identification number (UUID). For example, in Figure 4, the first character in the epigraphic hierarchy is a bêt (the underlying character) with a dāgēš and a šĕwāʾ (the diacritical dots in the middle of and below the character). In CEDAR, that epigraphic unit is broken down into its three constituent elements, each of which is itself a separate database item. Each database item can be used and marked with scholarly interpretation independent of the others. While this degree of atomization may seem extreme, it is required for the work of text-critical scholars, who attend to differences at this level of detail.
Third, our strategy of atomization makes it possible to compare manuscripts in great detail and with great flexibility. In the example in Figure 5, three different options are attested for the diacritics on the iota in the word ειδεν in Gen 1:4. We represent these options by creating three different epigraphic units, but nested within each is the same iota. We are not forced to normalize spelling, but the different options are still recognized as the same iota because of the reuse of the underlying character. Users can then choose whether they want CEDAR to highlight differences in diacritics or not. Atomization also allows comparison between manuscripts with diacritics and manuscripts without, while still avoiding duplication of character strings. We can use the iota in Figure 5 to encode a Greek manuscript that attests no diacritics at all by adding the iota alone to that manuscript's hierarchy, leaving the diacritics behind. The reuse of the database item creates alignment between that unaccented iota and any of the accented options.
Fourth, whenever possible, we create relationships between texts by reusing items in the database rather than by creating new items and linking them.16 This strategy has a number of benefits: it avoids data redundancy and eliminates the proliferation of links that would otherwise be needed to align content, it reduces the copying errors that inevitably occur whenever another version of a text is produced by human hands, and adding the 100 th version of a text and relating it to all the others is no more difficult than adding the second version.
The principle of reuse is achieved through what we call a content pool. A content pool is the set of all readings attested in physical manuscripts and critical editions of a given text. It represents not any one 16 The one circumstance in which we do not reuse database items is when we want to assign properties to a character that obtain only in one manuscript -for example, when we want to mark that a character is damaged, written above the line, ligatured, partially erased, etc. In these cases, we create a copy of the database item and mark the copy. If we were to assign a property to a reused character, that property would propagate in all manuscripts in which that character is used. Copied characters have no impact on CEDAR's ability to perform comparisons between manuscripts. particular text but rather all of the possibilities of a textual tradition organized into hierarchies. It is thus an abstraction. For example, in Gen 1:6, two Greek manuscripts have omega as the final character in γενηθητω, while another has omicron (γενηθητο).17 The Greek content pool in this verse thus reads γενηθητωο. Note that the content pool is not simply the alphabet for a given language, and it is not the case that every single bêt in the Hebrew text of Genesis is the same bêt: rather, every instance of the initial bêt in the first word of Genesis is the same. Each subsequent bêt is a different database item. After the content pool has been created, characters are then selected from it to build individual manuscripts, which we call local texts.18 Local texts are essentially subsets of data selected from the content pool. The content pool is what enables CEDAR to display local texts as if they were layered on top of each other. If we were to layer the Leningrad Codex, 4Q2, and 4Q8 (see the representations of these manuscripts in Figure 6), CEDAR would display agreement in the word wayyōʾmer because the same database items have been used to build that word in all three local texts: the vav is the same vav; the yod is the same yod; etc. Because the set of manuscripts where these characters occur has been recorded in the database, alignment algorithms are unnecessary for comparing manuscripts. 18 Incidentally, this strategy is very similar to the collation books used by the Göttingen Septuagint project. Along the left margin of a large ledger page are recorded all attested readings, followed by a list of manuscripts which preserve each reading. Because they were doing it on paper, however, the Göttingen project employed this strategy at the word level; CEDAR employs it at the character level. This feature of the database model enables the accurate comparison of an unlimited number of manuscripts and eliminates both the duplication of character strings and the vast network of links that would be required by the document model.
Fifth, once local texts have been constructed, they can be linked to digital images of the manuscripts. We can then demarcate areas of the image to link to specific characters or words in the transcription. When a user clicks on the transcription, the linked character on the image is highlighted. Clicking on the image also highlights the linked character in the transcription. This feature, called "hot-spotting," is particularly useful for studying fragmentary or damaged manuscripts in which reconstructions may be uncertain, because it allows a scholar to communicate precisely which character traces are being interpreted. Furthermore, after a manuscript has been hot-spotted, a user can query the database for all images of a given character. This visual catalog serves as a script chart for evaluating broken characters. One final strategy used by the CEDAR project is that we compare languages by aligning them through the discourse hierarchies of their content pools.19 Multiple items in one content pool can align with one item in another. The example in Figure 9 shows two cases of multiple Greek words aligned with one Hebrew word, as well as a case of one Greek word aligned with multiple Hebrew words. Because each local text is built out of the items in its content pool, aligning the content pools allows us to compare any two or more manuscripts in as many languages without duplicating data. We can compare any Greek local text with any Hebrew local text -or any several Greek texts with any several Hebrew texts -using only the alignment shown in Figure 9. Any new Greek or Hebrew local texts that we add will also automatically be aligned. In cases where the translation is sufficiently loose that it is impossible to establish a one-to-one relationship (or even a one-to-many relationship) between words in the content pools, we have the option to create higher-level structures in the discourse hierarchy. We can create phrases that align with phrases, as in Figure 11, or if need be we can go higher and align entire verses with entire verses, entire paragraphs with entire paragraphs, and so on. The Greek, Aramaic, and Syriac content pools are all aligned with the Hebrew because they are translations of the Hebrew.20 When a user asks to compare a Greek local text with a Syriac local text, that relationship is traced through the Hebrew content pool. Latin, despite also being a translation of the Hebrew, is aligned primarily with the Greek (see Figure  11) because as Indo-European languages Latin and Greek share syntactic and morphological structures which would be obscured if we compared them through the Hebrew. (For example, in the opening words of Genesis, the relationships in/ἐν and principio/ἀρχή would be lost if we aligned both languages with the Hebrew ‫.בראשית‬ Instead the relationship would be recorded as in principio/ἐν ἀρχή. While that level of detail would be sufficient for some kinds of research, textual scholars need their data to be as granular as possible.) Thanks to the flexibility of CEDAR, in cases where the Hebrew and Greek diverge and the Latin follows the Hebrew, we can align the relevant Latin discourse units with the Hebrew.

Using the digital HBCE
The benefit of CEDAR's organized system of reusable content is that rather than reading a list of decontextualized variants in the critical apparatus, users of the HBCE digital edition will have access to all of the primary data used by the scholar in constructing the edition, in addition to the tools to analyze it. Ronald Hendel's edition of Gen 1:9 for the HBCE will serve as a demonstration of CEDAR's potential.21 Let us say that we are reading HBCE Genesis 1, and we come to verse 9: ‫ם֙‬ ‫יִ‬ ֙ ‫מַ‬ ‫ּׁשָ‬ ‫הַ‬ ‫ת‬ ‫חַ‬ ֤ ‫ּתַ‬ ‫מִ‬ ‫ם‬ ‫יִ‬ ֜ ‫ּמַ‬ ‫הַ‬ ‫ּו‬ ‫ו֨‬ ‫ּקָ‬ ‫יִ‬ ‫ים‬ ֗ ‫ֱֹלהִ‬ ‫א‬ ‫ר‬ ‫אמֶ‬ ֹ֣ ‫ַּי‬ ‫ו‬ ‫ה‬ ‫ׁשָ‬ ‫ַּבָ‬ ‫ּי‬ ‫הַ‬ ‫א‬ ‫רָ‬ ‫ַּתֵ‬ ‫ו‬ ‫ם‬ ‫ֵיהֶ‬ ‫ו‬ ‫קְ‬ ‫ל-מִ‬ ‫אֶ‬ ‫ם‬ ‫יִ‬ ‫מַ‬ ‫ּׁשָ‬ ‫הַ‬ ‫ת‬ ‫חַ‬ ‫ּתַ‬ ‫מִ‬ ‫ם‬ ‫יִ‬ ‫ּמַ‬ ‫הַ‬ ‫וּו‬ ‫ּקָ‬ ‫ַּיִ‬ ‫ו‬ ‫ן‬ ֽ ‫י-כֵ‬ ‫הִ‬ ‫ֽֽיְ‬ ‫וַ‬ ‫ה‬ ֑ ‫ׁשָ‬ ‫ַּבָ‬ ‫ּי‬ ‫הַ‬ ‫ה‬ ֖ ‫אֶ‬ ‫רָ‬ ‫תֵ‬ ‫וְ‬ ‫ד‬ ֔ ‫חָ‬ ‫אֶ‬ ‫ֶה‬ ‫ו‬ ‫קְ‬ ‫ל-מִ‬ ‫22.אֶ‬ The accompanying text-critical notes indicate that there are multiple divergences in this verse from the Masoretic Text, and we decide to investigate. We first ask CEDAR to display HBCE as the base text with the Leningrad Codex overlaid in order to see more specifically how HBCE differs from the Masoretic Text. CEDAR uses color-coding to inform the user efficiently about the relationships between the texts.23 Green indicates that the texts agree. Black indicates characters in the base text (in this case, Hendel's HBCE edition) that are not attested in the overlaid text (in this case, the Leningrad Codex). Red indicates disagreement between texts. When we click on the red he, a pop-up window (see Figure 14, bottom left) provides relevant information about that character. The top half of the window shows information related to the texts on display in this particular view. We see that instead of HBCE's he, the Leningrad Codex reads mem, resulting in a different word: ‫מקום‬ instead of ‫.מקוה‬ The bottom half of the pop-up window shows information about texts in the project that are not included in this view. We can see that the he reading is shared by the Dead Sea Scroll 4Q8 and the retroversion of the Septuagint (see left red box). When we click on a black character, say the circled tav, a similar window (see Figure 14, bottom right) informs us that while the Leningrad Codex lacks the black text, the tav at least is attested in 4Q10 and the retroversion of the Septuagint (see right red box).
21 Ronald Hendel's critical edition of Genesis is forthcoming in the HBCE series. 22 "And God said, 'Let the waters below the heavens be gathered to one gathering so that the dry land may appear,' and it was so. The waters below the heavens were gathered to their gatherings, and the dry land appeared." 23 Our use of color-coding is not necessarily a long-term solution given the obstacle it presents for color-blind users. Based on the information from the overlaid view, we now want to look at 4Q8 and 4Q10 along with the Septuagint manuscripts that attest this verse. When comparing across languages, CEDAR displays transcriptions side-by-side rather than overlaid. When we click on a word in any of the transcriptions, CEDAR highlights the equivalent word in all the rest. When we click on ‫ה‬ ‫וֶ‬ ‫קְ‬ ‫מִ‬ in the HBCE edition, we can see the agreement in 4Q8 and the three Greek manuscripts. Nothing is highlighted in the Leningrad Codex because it has a different reading; nothing is highlighted in 4Q10 because it is fragmentary and does not attest this verse. Transcriptions are useful, but we want to see the primary evidence, so we ask CEDAR to display the image of 4Q8. The he is unequivocal. Satisfied with that reading, we move to the Greek manuscripts. A full page of Codex Alexandrinus is timeconsuming to navigate, so in the transcription we click on ϹῩΑΓΩΓΗΝ, the word we want to see, and CEDAR highlights it on the image. We repeat this process for the other two Greek manuscripts and confirm that the reading is sound.
Having worked through the evidence for the first non-Masoretic reading in the HBCE edition of Gen 1:9, we turn to the second. When we click on ‫א‬ ָ ‫ר‬ ‫ּתֵ‬ ‫,וַ‬ we see agreement in 4Q10 and again in the Greek manuscripts. 4Q10 preserves just part of a much longer plus which is attested in its full form in the Greek witnesses.24 24 On principle, the CEDAR team makes as few editorial decisions as possible so as not to hinder future scholarly analysis. At times, however, we are forced to make decisions, as here when we needed to decide how to relate this very small fragment of 4Q10 to the rest of Genesis: does it attest the end of the Septuagint plus, as we have chosen to encode it, or is it instead an apocopated form of ‫ותראה‬ earlier in the verse? While our decision was based on philological research (the wəyiqṭol of ‫ראה‬ is not attested in an apocopated form in MT), another scholar could reach a different conclusion. Because of its ability to encode multiple overlapping hierarchies, CEDAR allows for a future scholar to align the text of 4Q10 differently. This new analysis will not overwrite our initial analysis; instead, both will exist in the database, attributed to the appropriate researcher and available for subsequent use.   While Figure 20 shows a hot-spot only for the word ωφθη, HBCE editors can choose which structures they want to highlight. An editor could, for example, choose to make the entire plus a single hot-spot so that clicking anywhere in the text of the plus would highlight the whole. Because every element of a text is a database item in CEDAR -from biblical verses to lines of text to phrases to words to individual characters and diacritical marks -editors have the freedom to hot-spot any feature of a text at any level of detail.
This example demonstrates the value of the HBCE digital edition. Although there are still advantages to the print edition -reading comfort and the ability to access the resource without an internet connection, to name two -the digital edition can make available a wealth of primary evidence to a degree that print technology cannot match. Furthermore, the evidence is encoded so that it is granular and easily manipulable, creating opportunities for new observations that in turn will lead to new research.
A few distinctive features of the print HBCE which were not demonstrated in this example deserve mention. First, when an editor determines that the textual data represent multiple editions, the editions are printed in parallel columns (this feature is not applicable to the example of Gen 1:9). In the digital HBCE, both editions will be entered into the database as local texts which the user can choose to view in parallel columns or, for a detailed comparison, in an overlaid view (see Figure 14). Second, the print HBCE includes a critical apparatus and commentary that describe the relationships among the variants and, when appropriate, suggest scribal motivations. In the digital HBCE, these notes will be linked to the critical text and will be viewed in a parallel pane. Although this feature has not yet been utilized in CEDAR, it has been used by other OCHRE projects, such as The Ras Shamra Tablet Inventory;25 including it in CEDAR will be relatively straightforward.
Having provided an overview of the digital edition of the HBCE and the CEDAR project, we would like to offer a few methodological thoughts. First, the goal of this project is to support and assist scholarly analysis, not supplant it. We are not teaching computers to do textual criticism. Rather, we want to provide unprecedented amounts of data to researchers in a form that allows them to do their work more efficiently and in greater detail. Those researchers will still need to understand the significance of the data and know how to use it.
Second, we make no claims about the authority or primacy of any given manuscript or textual form. We do not even imply it by the way we organize our data. In CEDAR, all texts are given equal value. This point is important methodologically because we leave it up to scholars to determine, on a case-by-case basis, which reading is original.26 Furthermore, the CEDAR data model allows scholars to formulate their research questions with a great degree of flexibility. The user interface does not insist, for example, that the Leningrad Codex serve as the base text against which all other manuscripts are compared. It is entirely at the user's discretion to decide which of the many manuscripts should serve as the base text and which manuscripts should be added as comparisons.

Conclusion
The digital HBCE will expand the utility of the print edition without losing any of its distinctive features. Scholars will retain full access to the critical text, notes, and commentary of the HBCE while gaining all of the primary evidence that the editors used to construct their editions. The digital edition will thus offer a fuller realization of Hendel's goals for the apparatus of the print edition: "that it be clear, complete, and refutable."27 Even the very best apparatus presents only a sliver of the available evidence, both by design, to focus on what the editor has deemed relevant, and by necessity, because an apparatus must be brief if it is to fit on a printed page along with the text it refers to. The presentation of evidence in the digital HBCE, by contrast, is limited only by the time required for the data entry. Furthermore, CEDAR enables kinds of functionality, such as layering texts and hot-spotting images of manuscripts, that are technologically impossible in print. The digital HBCE will both complement and expand the print edition.28 25 See https://projects.rcc.uchicago.edu/ochre/RSTI/index.html [accessed 1 April 2019]. 26 "In principle, all ancient readings have an equal status, without relation to the text or translation in which they are found. Although there is certainly some statistical validity for the preference of certain textual witnesses over others, this judgment should not influence the evaluation of individual readings. Statistical information is irrelevant when data are evaluated" (Tov,Textual Criticism,272). 27 Hendel,New Edition,37. 28 A version of this paper was presented at the annual meeting of the Society of Biblical Literature on 17 November 2018. A recorded slideshow of that presentation, "A Preliminary Report from the CEDAR Project," can be found at https://cedar. uchicago.edu/test-cases/. Our thanks to Ronald Hendel and Jamie Carr for their comments on the paper. We also wish to acknowledge David Schloen, who conceived of the CEDAR project, and Jeffrey Stackert, who has greatly enriched CEDAR with his engagement, support, and enthusiasm. Finally, our thanks to Simeon Chavel, Joseph Cross, and Doren Snoek for their contributions to CEDAR.