The catalogue of the Department of Manuscripts and the Institute of Microfilmed Hebrew Manuscripts in the National Library of Israel, lists most of the Hebrew manuscripts in existence in the world, whether held in public or in private collections. The catalogue provides rich data about these manuscripts, including references to censorship (http://web.nli.org.il/sites/NLI/English/collections/manuscripts/Pages/default.aspx).

The manuscript catalogue was exported as an XML file on January 31, 2016, and contained a total of 163,248 records. First of all, we excluded 30,418 Genizah manuscript records (18.6% of all the records). The Genizah manuscripts are fragments of codicological units that were separated from their original manuscripts (due in part to damage or deterioration of codices) and were preserved in designated storage units called Genizah (from Hebrew “גניזה”, the verbal noun signifying the act of storing/archiving). Most of them are from the famous Cairo Genizah and some are from the European Genizah. These fragments are not relevant to our study because they have different characteristics than codex manuscripts and since Genizah research is a separate field of study [Richler 1990].

Next, in order to create a dedicated censored manuscript collection, we automatically searched the catalogue's XML file for records with the word stem “censor” (both in English and in Hebrew). In almost all of the cases, a proper “censor” stem was found in a dedicated MARC field, leaving only the cases where some morphological variations of the stem “censor” were found for manual inspection. The obtained dataset was further narrowed by filtering out some tens of manuscripts that were not censored in Italy and/or were censored later than 1775. The final dataset contains 5,185 censored manuscripts.

The structure of the catalogue follows a fundamental observation about the formation of medieval codices. The word “manuscript” has multiple meanings including: scrolls and codices, entire volumes (these can include several works copied by different scribes either at the same time or in different periods), a fragment of a few folios or a single folio. In our research we analyzed only codices. In every codex we found three fundamental layers: (1) the entire manuscript – the Bibliographic Unit (BU) that refers to the physical entity comprising the entire codex with its shelf-mark in the collection (generally corresponding to microfilm numbers). (2) Codicological Units (CU) – Some of the bibliographic units include a number of codicological units, namely a number of different parts of the manuscripts that were produced in different locations and/or at different time periods, either with or without a common theme. (3) Paleographical Units – The codicological units can also contain a number of paleographical units, and fragments written by different hands [Sirat 2002] [Beit-Arié 2018].

The catalogue database's basic unit, the record, describes a single CU. When the entire manuscript comprises a single CU (i.e. the codex is a single manuscript produced in one place and time), then the BU and the CU overlap and we obtain a unified codex [Beit-Arié 2018]. The record's data, in this case, refers to this manuscript as a whole (the majority, more than 90% of the manuscripts in the catalogue are of such a unified nature). The catalogue does not have, however, an independent notion (e.g. a special record type) of BU. Rather, when a codex comprises several CUs (termed as an assembled codex - multi-CU), the catalogue contains a special record, an index record that describes and lists the CUs of this codex BU. This set of CUs (excluding the index record) comprises a multi-CU BU with the shelf-number as its unique ID. In contrast, when there is a series of records that share a shelf number, but there is no index record with the same shelf number, we did not construct a BU for them and treated each record as an independent BU that contains a single CU. The index records themselves were not counted while calculating the various statistics of this study as in most cases they only summarize the data already contained in the individual CU records. However, we did use the data contained in index records when we built the study dataset, mainly as a means to verify the completeness of the data gathered from the BUs individual records (e.g. in some cases, the cataloguers specified the normalized full censor name only in the index record, while in the individual records they only gave a partial name).

In many cases it is difficult to decide when exactly in time a multi-CU BU (a codex) was bound. For example, if two censor signatures appear in a single BU, one dated to 1590 and the other dated to 1600, then one possibility is that this codex was bound before censorship (i.e. before 1590 (and censored together, while another possibility is that different CUs in this codex were censored in different settings/occasions only to be later bound together. Also, in some cases when the censor signed the manuscript, he also wrote the name of the place where the censorship took place (e.g. Mantua, Lugo). Following the previous example, by comparing the 1590's place of censorship to the 1600's place of censorship one may learn about manuscript transitions between people's hands and places. We manually checked all the BUs that contain two CUs or more. For each such BU, we used several inference rules to determine whether the signatures that the codex contains refer to all the manuscripts (CUs) in this BU or not. For instance, when the signature(s) appears at the end of the codex we assumed that the signature(s) refer(s) to the entire BU. If different types of script were used in different CUs of the BU, this can be an indication for distinct manuscripts that were censored separately.

Based on the above observations, BUs can be logically divided into two types of virtual entities: Whole Censored Unit and Non-Censored Unit. A Whole Censored Unit (WCU) groups CUs (of the given BU( that underwent censorship as a whole textual unit. The rest of the CUs in a BU, which are non-censored CUs are grouped under a Non-Censored Unit (NCU). We, thus, split partially or wholly censored BUs to one or multiple WCUs and a single NCU (see Figure 1).

We found that out of the 586 examined BUs, 428 were censored as a single WCU. Out of the remaining 158 BUs, 144 BUs have undergone censorship as a single WCU and were bound together with a non-censored unit at a later time. In 13 cases, we identified two separate WCU units, and in only one case did we split a BU into three distinct WCUs as we could identify that each one of them had been censored at different occasions and only later bound together into a single physical codex.

In this study we adopted the event-based modelling approach presented in Zhitomirsky-Geffet and Prebor 2016 and defined three types of events relevant to the goals of this research:

According to the attributes of these types of events (represented as rectangles in Figure 2), the manuscript dataset was divided into subsets, such as censored (WCUs) vs. non-censored (NCUs) manuscripts. These subsets’ data were analysed separately in different dimensions, then the findings were compared and differences between the subsets were analysed.

First, the corpus of manuscripts underwent a process of data cleaning and normalization. This process included decomposition of catalogue fields with mixed information into its defined atomic entities (e.g. people names, events, dates and places), standardization of a variety of date formats in the catalogue (e.g. specific Hebrew or Gregorian dates, ranges of years, centuries), and names of places (which appear in different languages and without the name of the country or geographic coordinates). Missing information was completed as well, using existing authority files and ontologies, such as VIAF (https://viaf.org/) and Geonames (http://www.geonames.org/). As part of this place normalization process we also obtained the coordinates of all the mentioned places in the collection which were extracted from the abovementioned authority files. The data normalization process was performed semi-automatically. It was based on identifying lexical patterns for each ontological entity, and further manual correction and completion by two experts in manuscript studies. For instance, for extraction of the censorship event’s location from the following string in the catalogue: censor signature: “LUGO 1611 CAMILLO JAGHEL”, the matching pattern was <CENSOR * [PLACE] [DATE] [NAME]>, thus LUGO was retrieved and then looked up in the Geonames’ “1000cities” list.

Next, we performed a systematic event-based analysis of the manuscript corpus. In addition to the descriptive statistical analysis we propose a new multi-dimensional methodology for a large-scale quantitative analysis of the historical manuscript data. The following types of variables were investigated for each of the dimensions:

Based on the above dimensions and variables a three-phase analysis has been performed:

Only 5,185 CUs have censorship information recorded for them in the catalogue. These 5,185 CUs are grouped in 2,466 WCUs, while the other 118,921 non-censored CUs are grouped in 85,562 NCUs (as shown in Table 1). The total of 124,106 CUs are associated with 87,848 BUs. As explained before, there are cases when a single BU was (manually) “split” into a WCU and an NCU or cases when a BU was “split” into more than a single WCU (a total of 158 of both cases). Note however, that the vast majority of cases (87,690) are indeed cases where a BU consists of either a single WCU or a single NCU. Moreover, note that 79,415 BUs (90% of all the BUs in the study corpus) are unified codices, that is, BUs that comprise a single CU.

The 2,466 WCUs are associated with 3,914 instances of Censorship Events (CEs) (see Table 2). The number of CEs is greater than the number of WCUs since in many cases, a WCU has been censored more than once, thus having more than one CE associated with it as discussed below. From Table 2 it can be noticed that in addition to the 3,190 CE instances where the censor responsible for the censorship was identified, in 724 cases, there is a clear notion of censorship, but no censor was documented in the manuscript's record.

Out of the 3,914 CEs, we filtered out 64 CEs: 56 CEs did not occur in Italy (and also didn't occur in the study period), and 8 Italian CEs were excluded since they occurred after 1775. The resulted dataset of Italian censorship events in the period of 1550-1775 consists, therefore, of 3,850 CEs (in 2,408 WCUs), 677 (17.6%) of which are un-named cases (see Tables 2 and 3). The earliest identified evidence of censorship in a manuscript in Italy is from 1554 and only two censorship events of Italian identified censorship have a recorded date after 1775 (in 1802 and in 1833), which fall outside our study period of 1550-1775.

From Table 3 we observe that out of the 677 CEs in Italy that have no censor name associated with them, in 513 cases a censorship signature was blurry and thus could not be identified. In 151 cases a clear sign of censorship text erasures is present in the manuscript, but without a signature, and in 13 cases the cataloguers identified self-censorship actions.

A rather interesting finding from analyzing the censorship events of the 2,408 WCUs in our corpus is the number of re-occurring censorship actions, that is, censorship of manuscripts that have been already censored. In other words, these are the cases where the same WCU (a single or a bounded set of CUs) has been submitted for approval to the same or a different censor at various times and locations. Of the 2,408 WCUs, 992 WCUs (41%) were censored more than once (see the distribution in Table 4). For example, the WCU - PARIS BN Velins 908 (identified by the shelf number F 73304) has been censored on four different occasions by four different censors, according to this manuscript’s record in the catalogue. It bears Luigi da Bologna’s signature from 1599, Camillo Jagel’s signature from 1623, Renato de Modena’s from 1626 and Girolamo da Durazzano’s from 1640. WCU - Moscow - Russian State Library, Ms. Guenzburg 147 (F 6831), on the other hand, records Girolamo da Durazzano’s signature, twice: once from 1640 and then from 1641, which suggests that this manuscript was censored in two consecutive years by the same censor (this manuscript has an earlier third signature by Renato de Modena from 1626).

The fact that Hebrew printed books have undergone repeated censorship is well known as is mentioned, for example, in a study by Baruchson-Arbib and Prebor (2007). What we are showing in our study is that a) the repeated censorship occurs in handwritten books as well, and more importantly b) that its scope is quite substantial: in total, for the 2,408 censored manuscripts (the WCUs) in our study, compared to an expected 2,408 censorship events (CEs) (one for each), 3,850 distinct CEs are registered for these manuscripts, i.e. 1,442 additional, repeated CEs (see Table 4). This finding may indicate that the repeated censorship phenomenon cannot be explained only by sporadic actions by local censors, and that much more weight should be attributed to other, more general, explanations that were given to that phenomenon.

Several directions were suggested to explain this phenomenon. One explanation for this phenomenon is possibly that Church authorities feared that the Jews “corrected” their manuscripts after the censorship and hence re-checking is required. However, the substantial amount of renewed censorship must also be attributed to factors relating to the Church’s conduct on one hand, and the conduct of local inquisition administrations, on the other hand. Firstly, the act of censorship of Hebrew texts did not receive an official and formal authorization. As a direct consequence, manuscripts which were approved by an official censor did not gain the official recognition of the Church. Also, the Church’s policy regarding the censorship of Hebrew books was seldom changed as new Popes came into office and the rules by which censors should examine the text did also change and a stricter expurgation policy was adopted in the course of time [Raz-Krakotzkin 2007]. On a local inquisition basis, censors apparently doubted their predecessors’ work for negligence or even deliberate eye-closing facing the monetary benefits from the Jews who paid for their job.

The findings in our study show that re-occurring censorship is a phenomenon that spread across time and across places. There are cases where repetitive censorship took place in the same place, while in other cases, the repeating censorship occurred in different place(s), and usually by different censor(s). Table 4 summarizes the repetitive censorship statistics. More detailed investigation, however, which will be dedicated to this phenomenon, could produce more insight. It may identify the censors who are responsible for repeated censorship more than others, it may find whether the repeated censorship happened on specific years or places, it might supply some characteristic as to the time elapsed between CEs of the same manuscript (e.g. is re-occurring censorship more common after a long time since previous censorship or maybe shorter elapsed time is more common).

By lexical processing the censorship recording in the catalogue, we were able to automatically identify the cases where more than one censor was responsible for a censorship mission. Part of these cases were identified by having the censors’ signatures on the same page with a single mention of the year of the censorship (e.g. Alessandro Scipione 1597 (275b); Dominico Irosolomitano (275b)) like in the example of the campaign of expurgation by a censorship committee that was in Mantua in 1595-1597 (see below). Other cases were identified by their unique form of the censor’s approval text which explicitly mentions the multiple censors as in the following example: “Subsignavi ego Guidus menotius ... Jac Geraldini ... 28 Marti 1556”. [4]

Another aspect of spatial analysis can be learned from the script types of the censored manuscripts. Hebrew manuscripts have been written in many scripts that varied from region to region, each with its typical script type. The Sephardic script, for example, is the typical script used by Sephardic scribes who worked within the Jewish community in the Iberian Peninsula. Since censored manuscripts were in the possession of the Jewish community in Italy, analyzing the distribution of their script types may provide new (quantitatively) perspectives of the extent of the transition of manuscripts from outside regions into Italy as well as on the immigration of non-Italian Jewish scribes who continued to write with the script type of their homeland and education [Beit-Arié 2018].

Five main branches of script types were proposed in the literature: Eastern script (which includes the Yemenite script and Persian script subtypes); Sephardic script; Ashkenazic script; Italian script and Byzantine script [Beit-Arié 2018][Richler 1990]. The catalogue named more than 70 distinct Hebrew script types. As a first step of the analysis, we have compiled a list of 17 script type classes with an additional Miscellaneous category for the rest of the scripts: Ashkenazic Script, Italian Script, Eastern Script, Sephardic Script, Yemenite Script, Western Script, Byzantine Script, Karaite Script, Persian Script, Provencal Script, Samaritan Script, Ashkenazic-Italian Script, Sephardic-Eastern Script, Sephardic-Provencal Script, Sephardic-Italian Script, Sephardic-Western Script, Typewriter and Miscellaneous. Next, we classified all the manuscripts into seven unified script classes (ustypes for short). For instance, the Sephardic unified script class was assigned to all manuscripts with variants of the Sephardic Script (e.g. Sephardic-Eastern script, Sephardic-Western script and Sephardic- Provencal script).

We then examined the distribution of the seven ustypes amongst 5,015 censored manuscripts and compared it to their distribution among a set of 72,233 non-censored manuscripts (overall, 77,248 manuscripts were created before 1700 as almost no censorship took place afterwards). The result of this comparison is depicted in Figure 6.[5]

It can be immediately observed that the Italian script type is the dominant one amongst the censored manuscripts (59.9%) compared to this script occurrence percentage (only 19.9%) amongst the non-censored manuscript created during the same time period. There was a significant difference in script type distribution between the censored and non-censored corpora, χ2(6)=118.21, p<0.0001. This implies that a major part of the manuscripts in the possession of Italian Jews at that time were probably written by this community, namely, written in Hebrew with an Italian script type. This finding also strengthens the results of Beit-Arié that the Italian script was the most frequent one (55%) used in the manuscripts written in Italy before 1540 [Beit-Arié 2018]. These results can be explained by Italy being the centre of the Catholic Church on the one hand, as well as being a prominent centre of rich Jewish life and culture, on the other hand.

The presence of manuscripts written in other script types can be explained by manuscripts brought/bought by Italian Jews at different times as well by the use of these writing scripts by newcomer Jews. We show that 17.4% of the censored manuscripts were written in the Sephardic script which is an indication of the (mainly forced) immigration of Sephardic Jews into this region. This script type is found in only 14.0% of the non-censored scripts which strengthen the indication of a major transition of Sephardic Jews into Italy, compared to all other regions. Another well recognized fact confirmed from this comparative analysis is regarding Yemenite manuscripts. From Figure 6, one can infer that almost no manuscript written in Yemen arrived in Italy before 1700, as there are no censored manuscripts found that were written in the Yemenite script type. Thus, through analysis of script types we can indirectly identify migrations (or the lack of migrations) of Jews (with scribes among them) and partly answer research question number 1c (other parts of this paper discuss other types of geographic migrations, those of the manuscripts themselves, this time) see also in: Prebor, Zhitomirsky-Geffet & Miller (2019).

Another indication of the migration of manuscripts (albeit, through a later time span) is their current location. The original Hebrew manuscripts that are documented in the catalogue of the Department of Manuscripts and the Institute of Microfilmed Hebrew Manuscripts in the National Library of Israel belong primarily to public collections held by many public institutions (mainly academic or municipal/governmental libraries) or to private collections all over the world. As part of our research we wanted to shed light on the distribution of censored manuscripts in these institutions. We found that the Hebrew manuscript corpus is spread across 1,133 different institutions around the globe (see Figure 7). 37.6% of them are stored in Israel's libraries (the majority are in the National Library of Israel in Jerusalem), 16.3% - in the United States and 15.7% are located in Russia.

However, the picture for the censored manuscripts is quite different as can be viewed in Figure 8.

The largest portion of the censored manuscripts is still stored in Italy and the Vatican City: 35.6%. This portion is much higher than the portion of Hebrew manuscripts in general found in Italy and Vatican City out of all Hebrew manuscripts (5%). This is another indication of the censored manuscripts being primarily written in Italy, as well as an indication of a relatively high retention rate (i.e. archiving) of material of this nature in a close proximity to where it was originally written/possessed.

Yet, the majority of censored manuscripts were transferred over the centuries to institutions outside Italy as shown in Figure 8. Following Italy are the United Kingdom (20.5%) and France (10.3%). Israel is only sixth with a 5.5% share.

Catalogue records include subjects from the Library of Congress Subject Headings (LCSH) (http://id.loc.gov/authorities/subjects.html). Each manuscript record was assigned multiple subjects which were sometimes too specific.

In order to determine what were the main subjects of the censored manuscripts and whether there is a difference in the distribution of censored vs. non-censored manuscripts among subjects, we created a list of 13 general topics. The list of 13 subjects that generalize and group together some hundreds of subjects from the catalogue are: Popular Beliefs and Magic, Ancient Rabbinic Biblical Exegesis, Non-Legalistic Exegesis and Homilies, Science, Philosophy and Ethics, History, Mishna and Talmud Commentaries, Jewish Legal Texts, Liturgy, Bible and Commentaries, Kabbalah and Mysticism, Letters and Belles Lettres, Grammar. The Miscellaneous category includes Karaite literature, Jewish apocrypha, list of names, polemic, Gematria. For each censored manuscript we asked an expert researcher to assign subjects from the above list (thus, a single manuscript may have been assigned more than one subject).

In order to compare the distribution of the subjects in the censored manuscript corpus to that in the non-censored manuscript corpus, a random sample of 857 non-censored manuscripts written before 1700 was subject-annotated by our expert as well (with subjects from the above list only). For this analysis, we used a subset of 5,046 censored CUs that were created before 1700. The results are depicted in Figure 9.

We found significant differences in subject distribution for censored vs. non-censored manuscripts, χ2(7)=14.525, p<0.05. The three major differences are in the Kabbalah and Mysticism, Philosophy and Ethics, and Liturgy that were censored significantly more than other subjects. Subjects with percentages of 5% and lower are included in Miscellaneous in Figure 9.

The subject analysis of the censored manuscripts might shed light on the goals of the censorship activity. We found that the most censored subject was Kabbalah and Mysticism, 20% of the censored manuscripts belong to this subject (compared to only 14% of the non-censored manuscripts). This is not surprising since in the 16th century Kabbalah study became more widespread in Jewish culture and a debate arose over the printing of Kabbalistic literature. In this period, Kabbalah also attracted the attention of Christian scholars. Several Kabbalistic works were translated into Latin, and the interest and preoccupation with Kabbalah became wider. The main reason why Christian scholars expressed interest in Kabbalah is the similarity they saw in some of the foundations of the Kabbalah and the principles of the Christian faith. It is impossible to ignore the historical context of the debate over Kabbalah literature in the Jewish world in the period of the highest Catholic censorship activity when each body acted out of its own motivations and goals [Raz-Krakotzkin 2007] [Weiss 2016]. The second most censored subject is Philosophy and Ethics (17% of the censored corpus compared to only 11.4% in the non-censored corpus). The books of philosophy that deal with the core of faith and content that can harm Christian faith, such as Rabbi Joseph Albo's Sefer ha-Ikkarim (The Book of Principles), which has the most comprehensive censorship requirements in Sefer Ha-Ziquq [Prebor 2003]. The third subject with a significant difference between the percentage of censored and non-censored manuscripts is Liturgy. The ecclesiastical supervision of liturgical books, especially in the Ashkenazi mahzorim (festival prayer books), was massive because they contained anti-Christian piyyutim (liturgical poems) and many expressions condemning the nations and the Christian faith [Benayahu 1971] [Raz-Krakotzkin 2007]. The prayer books and the mahzorim were the most popular category in the libraries of the Jews of Mantua [Baruchson-Arbib 1993]. The percentages of the top-censored subjects partially correlate with the numbers of the required corrections for various subjects specified in Sefer Ha-Ziquq. The highest number of corrections (24 on average) in Sefer Ha-Ziquq is required for the Liturgy subject and Kabballah is ranked third (with 16 corrections required on average) following the ancient rabbinic biblical exegesis and non-legalistic exegesis subject (Prebor, 2003). In contrast, we found more “neutral” subjects that did not catch the censors' attention, subjects with “non-problematic” content. This includes Karaite literature, apocrypha, lists of names, polemics, gematria, letters and belles-lettres, science, popular beliefs and magic, and history.

The National Library cataloguers have identified 65 censors that worked in Italy between 1554 and 1775. Some of the censors were very active with hundreds of signed censorship actions, while others only appeared in 1, 2 or 3 CEs (censorship events). Figure 10 displays the counts of CEs for the top-20 censors (out of the 65) that were involved in at least 15 CEs.

As shown in Figure 10 the most active censor was Domenico Irosolimitano. Figure 11 displays his (and three additional most dominant censors’) activity on a timeline. Domenico Irosolimitano acted mainly from 1595 to 1602 (the pick year is 1597) and then again in 1612. His signature is found in 671 manuscripts (about 19% of all CEs) although only 378 (56.3%) of them has a date of censorship. Domenico Irosolimitano was not the first censor to act in Italy. He was preceded by several censors (Caesar Belliosus, Jacob Geraldini, Guido Menotius, Hieronymus Caralus and Laurentius Franguellus) that worked between 1554 and 1589, but each of these censors is responsible to only few dozens of CEs.

The censors’ frames in Figure 11 are presented in chronological order of their activity. Thus, the next censor (chronologically) is Luigi da Bologna who acted mainly between 1597 and 1602. Unlike Domenico Irosolimitano who has quite an exceptional pick year, the CEs of Luigi da Bologna are more evenly spread along its active years and are virtually all dated (326 out of 351). The third censor, Camillo Jagel, acted mainly in three years: 1611, 1613 (the pick year) and 1619. Camillo Jagel was responsible for 371 CEs (356 of them are dated). The fourth censor in Figure 11, Giovanni Domenico Carretto, acted mainly in 1617-1619 (the pick year - 1618). He is the second most active censor after Domenico Irosolimitano with 572 CEs (552 of them are dated).

The last frame in Figure 11 shows the overall censorship activity distribution over time. As can be observed, the CEs are not evenly distributed along the timeline. There were four main micro-waves of censorship activity in 1595-1602, 1608-1613, 1617-1619 and around 1626. These micro-waves generally match the activity periods of the four most dominant censors discussed above.

To complete the censor-related CEs analysis we performed a detailed breakdown of the CEs over time for the top eight places (including “the Papal States”) that have at least 30 CEs. The result is shown in Figure 12 (these places refer to 96.7% of the CEs with known or inferred locations: 2,009 out of 2,077). The figure also demonstrates the relative part of CEs that are associated with each of the most prominent censors (Domenico Irosolimitano, Camillo Jagel, Luigi da Bologna and Giovanni Domenico Carretto, represented with different colors).

From Figure 12 it can be observed that in some places the censorship took place during a very short period (sometimes of one or two years), e.g. in Lugo, Ancona, Urbino and the Papal States. Censors’ movements in time and geographic locations can be traced as well, e.g. Domenico Irosolimitano (denoted by the red color in the chart circles) moved from Mantua to Ferrara and then to Rome, and Camillo Jagel (marked by the green color) moved from Modena-Reggio to Lugo and later to Ancona. Also, it can be noticed that Rome's CEs are mainly attributed to Domenico Irosolimitano, who was also a dominant censor in Ferrara and Mantua. Camillo Jagel is responsible for virtually all censorship in Lugo and Ancona. Modena-Reggio CEs are mainly executed by Luigi da Bologna with some noticeable contribution of Camillo Jagel (the late CEs in the 1640's are Girolamo da Durazzano's). Finally, Mantua's censorship was performed by Domenico Irosolimitano (in the early period until 1603), while more than half of the CEs in 1617-1619 are attributed to Giovanni Domenico Carretto. The CEs in the Papal States were almost entirely signed by Caesar Belliosus and in Urbino all CEs were signed by Antonio Francesco Enriques (not in the list of the top four censors).