University of Virginia Library

VI

Relatively few same-face fonts exhibit the plethora of discriminants seen in mixed fonts such as White-M and Stafford-EF. Generally, font analysis must resort to the kind of tedious search that is necessary for detecting low-frequency foul-case letters. Foul-case pica roman letters constitute a category of discriminants between same-face fonts and are, quite simply, distinguished from the normal letters by the wrong-face and size differentiae discussed earlier. It can be generally assumed that the source of such letters is another pica roman font previously or currently used in the shop; hence, fouling of an S-font by Y-face letters usually implies the possibility of a two-font shop, although this inference must be verified by a survey of books by the printer. However, anomalies occur where the source of a foul-case pica roman letter cannot be explained in this manner; for example, Simmes-S contains a very few Y-face letters although he never used a Y-font. These include Y-face 'k' (Mal Q1, F4v:30; Q2, F1v:33), 'M' (Q2, G3:24; Q3, C1:29, G4:26) and 'P' (Q3, G4:11); further appearances, if any, have been overlooked.[41] Similarly, Felix Kingston's Y-font contains a few Guyot 'G' (The Works STC12312 [1599], C3v:34, C4:1, C5v:11, D2:17 etc.), and White-M uses the unique 'th' and 'ch' ligatures. The value of these letters as discriminants depends upon their rarity (either in a quantitative or stylistic sense) or longevity in the context of other same-face fonts. Windet-S is fouled with the short second-stem 'w' and B-face 'a', both fairly rare variants which also appear in White-M and Danter-M but not as a result of fouling. Both Eld-S and Windet-S show a few wrong-face medium 'z' which do not correspond in either instance to a second font in their shops and do not appear in the other S-fonts discussed here. Letters from the small capitals and shorter roman fonts (68mm, 76mm) constitute another common class of foul-case roman letters. In addition, the damaged 'ſſ' ligature distributed into the 'ſl' ligature sort and foul-case punctuation provide two kinds of valuable discriminants. It seems that compositors were little concerned with stylistic integrity in the latter area, and promiscuously set punctuation from wrong-size roman, italic and black letter sorts when shortages developed in the roman text-font. Hence, quite a few combinations resulted from mixtures of: S- and Y-face roman, black letter, and short and tall italic variants of the query and exclamation point; large Guyot query; S- and Y-face, black letter and italic (two degrees of lean) colon and semicolon; black letter period; and three useful sizes of comma. As a general rule, it seems logical to assume that foul-case italic and black letter punctuation marks are by definition a result of fouling rather than replenishment.

The complementary italic font in a shop was a major contributor of foul-case letters in the strict sense (i.e., an italic letter set in a roman text), but as a practical matter, the concept should be extended to include the use of italic for emphasis in italicised titles and names, speech prefixes and

stage directions.[42] Both kinds of appearances of italic letters serve the same identification function in font analysis. Generally, foul-case italic letters serve as simple wrong-face discriminants. However, italic typefaces also exhibit some major stylistic variations that can contribute complex discriminants, that is, stylistically variant (Guyot vs Granjon), wrong-face (italic) letters. An analysis of a printer's complementary pica italic font is necessary to identify such complex discriminants for use in a composite. It seems that many printers replenished the upper-case and ligatures of their italic fonts without regard for stylistic integrity and the process produced a rich lore of potential discriminants. These italic stylistic variants are usually easily detected in the upper-case and ligatures for the same reasons noted in regard to pica roman fonts. Simmes's italic font uses capitals in two typefaces and two sizes. For example, his italic 'C' appears both in a normal-height version positioned on the baseline, and a tall version that equals the height of the typeface (i.e., extends from top of ascenders to bottom of descenders), and the italic 'S' and other capitals come in normal and small sizes.[43] Since the italic was used primarily as an emphasis font, it is frequently necessary to survey several books by a printer in order to accumulate a sample of italic letters sufficient for determining whether or not he used a particular stylistic variant. Given a typical dramatic text with the italic font used only for emphasis and speech prefixes, it is unreasonable to reach such a conclusion exclusively from the letters in a single text.[44] For example, between Q2 and Q3 of The Malcontent, Eld purged the italic 'M' with the swash terminating in a curl and replenished the sort with a simple swash 'M', and also purged the italic 'w' with the stem-strokes extending above the x-line and replenished with the variant whose stems begin at the x-line.[45] Unless books before and after Q3 were surveyed, it would seem that Eld's italic appears either in Q1-2 or in Q3, with another printer represented by the variant italic 'M' and 'w'. In some fortunate instances, the extent of the survey required for examining an adequate sample is lessened by the appearance of long italic passages in dedications, prologues and epilogues (see Sejanus, ¶3-A1v:6; All Fools, A3-A3v:14; Romish Spider, A3-A4v), in the peculiar reversal of text and emphasis fonts in Romeo & Juliet Q2 (B4-4v) and Volpone STC14783 [1607], and in rare instances, in an entire text (Vertumnus STC12555 [1607], [2]B-H). As the Eld example implies, a composite of italic letters based upon an extensive survey of a printer's books can be valuable for dating a particular book.

As noted earlier, the distinction between fouling and replenishment is quantitative, with the former implying a second font in a shop. The distinction is also important as regards the function of foul-case evidence in font identification, and beyond that, in the analysis of compositors, method of setting and distribution, standing type and sequence of printing, since the two processes have a different effect upon the manner in which variant letters reappear. A wrong-face population created by replenishment in significant proportions remains stable throughout the lifetime of the font, since it results from a deliberate choice to introduce wrong-face letters, and unless the standards about font integrity were reversed, the letters would not be purged.

The average ratio of normal and wrong-face letters remains stable across a sequence of gatherings and texts; both variants of a letter will reappear consistently, although the ratios on a given page, pages or gatherings may vary because of the varying demands on a sort by a text; or the ratios may vary in a sequence of texts because of a shift in proportions produced by subsequent replenishment. In general, replenishment does not produce the kind of evidence needed in applications other than font discrimination unless, of course, just one of two cases were replenished with a variant letter. Low-level replenishment represents a special case that will be discussed below. In contrast, the very low density of foul-case letters precludes any consistency in their appearance, although in practice they usually do reappear across a sequence of gatherings or texts. However, their recurrence is both random and unpredictable in the context of predefined units such as the page, gathering, or even two texts. The patterns of recurrence will emerge in the following discussion. In general, a few foul-case letters in a sort will recur with no more regularity than identified types, which often skip gatherings and texts between appearances. But a fundamental difference between foul-case types and identified types must be recognized. The non-specific nature of foul-case evidence is a problem in itself in as much as the very concept of "evidence" is inseparable from the notion of specifically identifiable facts. Ideally, reappearing foul-case letters could be identified and their recurrence thus proven, and this is possible in some instances. For example, Braddock-Y1 is fouled with S-face 'A' (among other letters) that occurs 28 times (an average of three per gathering) in MND Q1. S-face type A1 accounts for six of these appearances (B4v:2, C2:21, D3v:4, F1:22, G4v:26, H1v:13). Three damaged 'ſſ' in the 'ſl' sort recur as well: types ſſ7 (B3:18, C3:11), ſſ8 (D4v:16, F2v:4), and ſſ9 (D4v:6, E1:18, H3:20). Damaged type ſſ2 recurs at least twice in the 'ſl' sort in Short-Y in R3 (L4:4, M3:27). Nonetheless, the use of foul-case evidence is complicated by the fact that most foul-case types are unidentifiable due to the lack of unique damage, and hence must be treated as representatives of a class of letter, that is, as members of a sort (i.e., "three appearances of S-face 'A'" as opposed to "type A1 appears in three locations").

However, the obstacles presented by the nature of foul-case evidence can be circumvented by a method of analysis based upon two concepts that render foul-case types valuable despite their non-specific identities both in font identification and more specific applications related to compositorial and presswork analysis. First, the critical distinction between transient and resident fouling must be recognized in dealing with this kind of evidence. Transient fouling, which endures for a few gatherings but not beyond a text, is of little value in font identification, but can be useful in presswork analysis, since it signifies compositorial behavior in response to conditions such as shortages in a sort. Consequently, short-term transient fouling can provide strong circumstantial evidence of the sequence of printing. The fact that only A-C of MND Q1 are fouled with black letter colons suggests that the book was printed sequentially beginning with A, and not B-H then A, as has been suggested.[46] In contrast, resident fouling endures across several texts (and frequently

throughout the lifetime of a font) and is of primary importance in font identification and presswork analysis. Due to the random recurrence patterns exhibited by foul-case letters, it is necessary to establish whether fouling is transient or resident before applying foul-case evidence to bibliographical analysis. Otherwise, fundamental errors in interpretation can result. Treating the recurrences of small capitals 'I' and 'A' throughout MND Q1 as repeated responses to newly developed shortages in these sorts (transient fouling) totally distorts the picture of presswork that emerges, if, in fact, these foul-case letters are resident in the font, i.e., already in the respective sort compartments before setting began. Verification of resident fouling requires surveying the font through the target text, and then other approximately contemporary texts. "Contemporary" need not be interpreted too strictly: resident fouling endures in Eld-Y1, Eld-S, Creede-4, Windet-S, Purfoot-Y, Short-Y and other fonts during a significant portion of their lifetimes. The necessity of surveying several books in distinguishing resident from transient fouling is illustrated by the dramatic drop in foul-case 68mm roman 'Y' in Short's section (H-M3v) of R3 Q1 (1597). Eight appearances occur in H, but only one in I-L and two in M, although textual pressure on the 'Y' sort shows no similar decrease, a pattern that suggests transient fouling. However, the 68mm 'Y' fouling is resident in Short-Y at least from 1595-99 (3 Henry VI STC21006 [1595]; i Henry IV Q1 STC22280 [1598]; and The English Secretary STC6404 [1599]).

Second, an understanding of the manner in which fouling occurs is essential to the interpretation of foul-case evidence. The non-identity of recurrent foul-case letters is compensated for by the fact that resident fouling of a font almost inevitably occurs in two or more sorts and thus produces a cluster of fouled sorts that reappears in sequences of gatherings or texts. Treating a font's foul-case letters as a "cluster" is simply an application of the statistical concept of a randomly selected sample population that is to be measured against a predefined distribution of variables in order to determine the probabilities that the sample conforms. In this instance, the variables consist of the cluster of foul-case letters in the populations of the affected sorts as a whole. A comparison of the total populations of affected sorts in two fonts reveals the order of statistical probability that the fonts are the same or not. Since resident fouling usually exhibits a fairly consistent cluster of affected sorts, it is possible to rely upon a quantification consisting of the density of the members of the cluster in a sample population whose limits are set in response to criteria suggested by the text or texts under examination. The quantification need not be overly rigorous when attempting to distinguish two same-face fonts. Each font can be treated as a whole without respect to the distinction between two sets of cases. Hence, a definition of a cluster in units of member density per gathering is entirely adequate; in practice, the cluster compositions alone will frequently distinguish two same-face fonts without reference to member density. Hence, a simple application of this approach sans statistical analysis can show that the cluster remained consistently fouled through the sample texts, thereby providing strong circumstantial evidence of the identity of a font.

The random recurrence patterns exhibited by low-density fouling are determined by the dynamic relationships among at least four factors: the continuous process of fouling and purging as affected (1) by the ease of purging and (2) by compositorial skills and intention; (3) the variation in the demand upon fouled sorts by sections of a text or by different texts; and (4) the distribution of foul-case letters deeper into sort compartments or into a different case, thereby delaying later reappearances.[47] Although the first two factors seem too obvious to belabor, the standard method of interpreting foul-case evidence in compositorial analysis suggests the need for a closer examination of the interaction of the two as they affect the permanence of fouling in a font. First, the ease with which fouling types could be detected during distribution seems important. The fact that compositors distributed on the basis of a visual recognition of each type seems well established since Charlton Hinman first made the point, and is confirmed by the pervasiveness of one kind of fouling in roman pica fonts. The damaged 'ſſ' ligature which lacked the right shoulder approximated the appearance of the 'ſl' ligature and was commonly distributed into that sort. These damaged ligatures are obvious in proof because the right letter retains the spur which distinguishes it from a straight-edge 'l', but the spur would be virtually invisible during distribution. Due to the damage, it is possible to establish many of these foul-case ligatures as separate types, such as three in Windet-S (Sophonisba, B2:34, D2:1, F4:37) and others as recurrent types, as in Windet-S (Fawne Q1, E1:29, I2v:19), Eld-Y1 (Fools, A4v:14, H4v:2), Braddock-Y1 and Short-Y (noted above). Eld-Y1 presents an interesting instance (unless my eyes deceive me) of visual confusion during distribution: ſſ4 first appears in the 'ſl' sort (Mal Q1, C2:33, E3v:7) and then the 'ſt' sort (Mal Q3, H3:20). On the other hand, foul-case italic capitals, for example, would be obvious on the stone and easily purged, but their permanence varies among fonts. Four appearances of italic 'S' occur in Simmes's section of The Malcontent Q1 (F2v:23, F3: 10,24, H4v:1), but none in the same sections of text in Q2, nor in Simmes's section of Q3 (B-G). Obviously, textual demands on the 'S' sort were exactly the same in Q1, Q2, and F of Q3; hence the non-reappearance of italic 'S' fouling suggests routine purging of the sort at the natural break-point between editions. Similarly, the five variants of the query (S-face, Y-face, italic, black letter, large Guyot), the S-face and italic exclamation point, and the black letter colon and period could probably be recognised during distribution and purged, but on the whole they seem to remain resident once introduced into most fonts. But unless a compositor distributed against proofs, which seems extremely unlikely, the clash presented by oversized capitals such as the S-face 'G S E' in a Y-font (usually quite noticeable in proof) and by less obvious smaller capitals such as Y-face 'G S' in an S-font (fairly difficult to detect in proof) would go unnoticed during distribution and remain resident in a font. Thus the smaller Y-face 'S A T' which recur in Eld-S in sequential gatherings of Sejanus also are found in later texts, even when a prose text, such as The Romish Spider (STC5704, new STC5693.7 [1606]), requires relatively few capitals per gathering.[48] Windet-F 'A D G H M O T' fouling in

Windet-S is seen in Fawne Q2 and Sophonisba. As noted of Eld-Y1 below, even the larger S-face letters seem to remain in Y-fonts. Small capitals presumably could be purged if a compositor desired, but small 'Y' fouling is resident in Short-Y from 1595-1599; the residence of small capital 'T' in Danter-M is probably attributable to a lack of concern about fouling in this mixed font. Finally, the stylistic clash created by wrong-face lower-case roman letters could hardly be detected without proofs whether or not a compositor was intent upon sorting them out. Thus the Y-face 'g k p' in Sejanus still reside in Eld-S in Spider despite seven(?) intervening texts which provided ample opportunity for purging.[49] Likewise, the Windet-F 'g w' reside in Windet-S at least from 1604-06. The S-face expanded x2 (and x3) seems the only likely lower-case letter that could be distinguished without proofs; nonetheless, it remains in Eld-Y1 after its introduction in Mal Q3.

Not surprisingly, the compositor is the second factor responsible for the difference between resident and transient fouling. Compositors seem to vary considerably in their purging skills and attitudes. Although the intention and skill of a compositor cannot be taken as the sole factor influencing the permanence of fouling in a font, the contrasting treatment of transient vs resident fouling reveals the importance of compositorial intention.[50] That is implicit in the frequent instances where foul-case letters in previously fouled sorts are tolerated while new foul-case letters are purged. It seems that the newly fouled area of text or the kind of newly introduced foul-case letters are fresh in mind and a deliberate decision made to purge. Heavy textual pressure on the query sort in R3 K4 produced fouling with the black letter query, followed by a complete purging. The previous pages of K required six italic, four Y-face and one black letter queries, but in K4 the one italic and four Y-face queries were supplemented with 13 black letter queries. However, the ratio of black letter to Y-face and italic queries remains in LM at the previous levels, suggesting that the K4 foul-case marks were immediately purged, but the resident black letter queries earlier in the gathering left undisturbed. Since this compositor could purge 13 marks from K4, he obviously could recognize and purge the remaining black letter queries if he intended to. Furthermore, the density of black letter query fouling in Short-Y changes radically across three texts, confirming the purging ability of Short's compositors. In 3H6 (1595), black letter query appears in proportions similar to that seen in 1H4 (1598) and English Secretary (1599).[51] The persistence of resident fouling with italic capitals in many fonts suggests a decision to tolerate certain foul-case letters while purging others. Italic 'T' fouling in Short-Y varies in a manner similar to the black letter query, appearing at moderate density in 3H6 (1-5 per 8 pages or gathering) and H of R3. One compositor then fouled his halves of I(16 'T'), K(11) and L(10) extensively with italic 'T'. Both cases were incompletely purged, so that no foul-case 'T' appears in 1H4 until I3v(1), K2v(2) and K3(1), and once in Secretary (1599) (Ll4v:29). Sort-pressure is an unlikely explanation for these appearances.[52] The types were probably buried in the sort compartment when the 'T' was purged as part of the replenishment of the sort after R3 with S-face 'T' in about equal proportions, a factor that

probably accounts for the much-delayed recurrence.[53] But the italic 'S' in R3 (L1:21, L2:4) does not seem to reappear. In the meantime, other foul-case italic capitals are tolerated through the four texts. Fairly heavy italic 'I' fouling occurs in all four. Other low-density italic capitals 'E F K P' recur sporadically, skipping gatherings and texts in the typical manner. Next, the italic 'B C L S' remain in Purfoot-Y1 after Dutch Courtesan STC17475 and Romish Spider to reappear in Fawne while the low-density 'swash-B F G swash-P V' do not; the transient heavy fouling in DC A-E with 'I' (76) and 'T' (27) was purged after DC. [54] One Windet-S case (Windet-S1, see below) is fouled with 'S' in both Fawne Q2 (H4v:6) and Sophonisba (Bv1:24, C2:29,38, D4:37, E4v:31,33, F1v:6,25), but the italic 'A B I L T' cluster in Windet-S2 does not recur in the latter. The difference probably is attributable to the attitudes of the S1 and S2 compositors. However, the difficulty of purging even obvious foul-case letters despite the compositor's intention is revealed in instances where these letters are eventually purged, but not entirely until two or more distributions. Fouling of MND Q1 with black letter colon begins with one in A2, then B is heavily fouled with 20, most of which were immediately purged, leaving two in C1 to be purged at the next distribution; only the roman and italic colon occur in the remainder of the text. The large number of other foul-case letters in Braddock-Y1 were simultaneously ignored. Likewise, Simmes-S was fouled with italic 'I' in A of Mal Q3 which were eventually purged by the distribution of F (A3-4v: 12, B: 5, C: 5, D: 4, E: 4, F: 3, G: 0). Compositorial intention to purge local large-scale fouling is evident in both instances. The kind of oversight implicit in the sequence of distributions of Simmes-S suggests the probable manner in which low-density obvious foul-case letters became resident in a font. It seems likely that fouling of portions of a text by one of two alternating compositors of different competence would increase the possibility of foul-case italic letters remaining in a font, especially if the second compositor distributed the fouled sections without being alerted to the need for purging them.

The varying dynamics of the continuous process of fouling-purging due to differing textual demands upon vulnerable sorts, ease of purging, and compositorial intention are exemplified in Creede-4, where heavy italic fouling with 'I' and moderate fouling with 'S T' were tolerated as resident while other italic letters were routinely introduced and then incompletely purged. In an early book, the text was fouled extensively with italic 'A swash-A' (26) and 'Guyot A' (1) in Creede's section of Antichrist STC7120 (1603, [2]H-M, see [2]H2v,3,4v,6v,8,8v), and was then purged, since it does not reappear in later gatherings or books. The resident cluster begins to develop in I (two C), then in K with 'G' (23), 'swash-G' (2), 'I' (23), and 'P' (2); the 'I' remains in L (9), 'swash-P' emerges (1) and 'P' (6) increases; only the 'P' reappears in M (10). Compositorial intention explains the localization of fouling, since sort-pressure remains constant ("God", "Priest") throughout. By Honest Whore CD (1604), the cluster is in the font. High-frequency 'I' appears 46 times in C alone, while 'S T' occur at moderate levels in C(9, 5) and D(7, 7). The cluster is completed by low-density italic 'C F G swash-G P W' and Y-face 'M'

that recur in the typical manner. The 68mm roman 'C' (C4:8[2], D1:5,11) and 'I' (C4:16) do not seem to reappear as part of the cluster, although this may be an oversight. Textual demands in high-frequency words such as "Bishop", "Church", "Doctor", "Epistle" and "God" in Woeful Cry of Rome STC1833 (1605) pulled italic 'B swash-B E' and two Y-face 'D' (B3:34,35) into the text and markedly increased the frequency of appearance of resident low-density cluster members (excepting italic 'W' and Y-face 'M') through additional fouling. An attempt to purge at some time following Woeful Cry is implied by the state of the font in Regiment of the Church STC1827 (1606). Regiment begins with italic 'I' fouling to B4 where the first four italic 'F' ("Father") intrude; only nine 'S' have been set, all roman. Additional fouling that persists throughout Creede's section of the book obtains by D with the full cluster, except that the italic 'swash-B swash-G E' and Y-face 'D' do not appear by this point. It is possible that only one case was completely purged after Woeful Cry and used to set most of Regiment B, but that seems less likely than an incomplete purging of both cases. In any event, both were fouled by the completion of Regiment. Creede's section (A, B(i)) of d'Olive (1606) exhibits the same italic cluster and Y-face 'M'. The large size of the fouling cluster and the progressive fouling in Regiment suggest the sequence in which the two books were printed. d'Olive was not entered in the Stationer's Register and the date of printing has not been determined beyond the imprint date of 1606. Regiment was entered 12 August 1605 but bears an imprint of 1606. The state of Creede-4 with respect to fouling at the beginning of each book seems significant. In Regiment, the text is clean except for 'I' (the first on B1v after five S-face 'I') until the four italic 'F' in B4. Pressure on the sorts in the cluster varies up to and including B4. S-face cluster members appear as follows: B(5), C(9), D(5), G(9), F(10), P(6), S(9), T(22). d'Olive is fouled by the end of the second page (A2v) with italic 'B I P S' before sort-pressure has a chance to develop. This is clear from the order in which the roman and italic capitals were set in their respective sorts. The italic 'I' at A2:6[text] is the first of the sort to be set, the 'S' at A2:8 is the second, the 'P' at A2v:6 is the first, the 'B' at A2v:29 the second, and the 'G' at A4v:29 the second. In short, Creede-4 began d'Olive with the cluster in residence. It seems certain that Regiment was printed first, and quite likely was followed closely by d'Olive. The increased recurrence rate of some members in comparison to Whore suggests that, despite the effort, each subsequent attempt to purge the font left a few more foul-case letters in residence.[55]

The importance of the compositor as a key variable cannot be overestimated, judging from the relation between Windet-S and Windet-F. The alternation of Windet-F with the two cases of Windet-S in Fawne Q2 provides confirmation of proper distribution of substituted roman capitals. One compositor set text from the Windet-F cases and from one of two Windet-S cases (Windet-S2) distinguished by fouling with Windet-F. "Dulcinea" required extensive substitutions of roman 'D' in the Windet-S2 sections of G through E of Q2.[56] For whatever reason, the compositor substituted from both the Windet-S2 and Windet-F cases. Hence, the route for possible mis-distribution

was: Windet-S2 and Windet-F to speech prefixes; speech prefixes to the Windet-S2 case, into which the roman type on these pages was being distributed. Windet-S2 contained a few resident Windet-F 'D' that reappear in the text at G2(2), G3v(2), F3(1), E2v(1). The Windet-F 'D' substitutions include G2(3), G3v(2), E2v(4). If these had been distributed into the Windet-S2 case, we would expect the rate of reappearance of Windet-F 'D' to increase in the latter gatherings (F-E). Moreover, the Windet-F 'D' fouling appears in Windet-S2 sections of Sophonisba at the same low rate (C2v:19, C4:33, D2:11, G3: 15). Presuming that Sophonisba was printed after Fawne Q2, the substitutions in Fawne Q2 were transient and correctly redistributed.[57] At the same time, the resident cluster of Windet-F capitals in Windet-S2 (A D G H M O T) is tolerated and recurs in Sophonisba at about the same low rate. Given the virtual impossibility of purging fouled lower-case letters, it is not surprising that the Windet-F cluster (a b g k w x y z) remains resident in Windet-S2 in both books. Purging foul-case ligatures is equally difficult, yet this compositor exhibits the ability to sort out the S-face 'ſh' and condensed Windet-F 'ſh' ligatures. Overall, Windet-S2 sections of G are heavily fouled with about 68 Windet-F letters, including two condensed 'ſh' (G3v:36). In resetting G4v from the Windet-S2 case, this compositor substituted ten condensed 'ſh' (G4v: 23-40) and then purged them. Only three appearances occur in later S2 sections of Fawne Q2 (E1:27,33, F3:39), and in heavily fouled S2 sections of Sophonisba (E3:35, E4:1, F2:1).

In general, compositorial intention must be considered a pivotal factor in determining the permanence of fouling in a font. However, the process of purging foul-case letters was subject to a deliberate choice on the part of the compositor, but that decision was not inevitable, since foul-case letters were routinely tolerated and produced resident clusters. Moreover, the effectiveness of the process was limited by the class of letters that were involved. It seems clear that deliberate large-scale fouling of a small section of text created favorable mnemonic conditions for immediate purging if the compositor so desired, although even obvious foul-case letters presented some difficulties in this situation. If a highly skilled compositor was intent upon purging foul-case letters, the kind of letters made a significant difference: purging some classes of letters was well-nigh impossible, regardless of skill or intention. Hence, the relation between the ease of purging, and compositorial skill and intention, has serious implications in font analysis and identification. Fouled lower-case letters can be trusted implicitly as font discriminants, roman capitals slightly less, and italic capitals only if they can be established as resident in the roman font over a sequence of texts.

Given the difficulties of purging even obvious letters, resident foul-case roman types thus tend to reappear fairly consistently over a period of time, either in a given sort or two, or in clusters if several sorts are fouled, a fact which enhances their value as font discriminants as well as evidence of presswork despite the inherent ambiguity of their specific identities. This is also true of wrong-face letters introduced through very low-level replenishment. As will be noted below, replenished wrong-face letters can occur in such low

proportions that they approximate the recurrence behavior of low-density foul-case letters even though they appear more frequently than the latter. Hence, it is expedient at this point to qualify the concept of "foul-case" to include such replenished roman letters. In general, low-density fouling exhibits a random unpredictability with respect to reappearances within a text or across several texts. Thus foul-case evidence must be approached from the perspective of the statistical principles of randomness and probability rather than in simple numerical terms. The rate of appearance in a sequence of gatherings can be described in terms of the minimum-maximum frequency per unit of text, i.e., 1 appearance per three gatherings, 2-5 appearances per gathering of approximately 300 lines and so on. Since recurrence is random, these can occur anywhere within the unit, i.e., page two of the first gathering, four within the first 20 lines and none after, none until two in lines 298-299, etc. It is surprising to find such theoretical extremes regularly occurring in texts.

This random behavior of low-density fouling can present a problem in font identification if the target font is compared to a similar font in only one or two other books. Members of a cluster, or the cluster itself, can skip gatherings and entire texts and thus create a presumption that a different font is in the other book. The recurrence of the S-face 'g k x' cluster in Eld-Y1 illustrates how the principle of randomness translates into non-reappearance of foul-case letters even though they are probably in the font. S-face 'k' first occurs in Mal Q2 (1 appearance), then in Whore (3), Q3 (1), Fools (4), and Eastward Hoe! Q1 STC4971 (4); S-face 'g' first occurs in Mal Q1 (2), not in Q2, then in the later books in increasing numbers as fouling progresses toward the end of the lifetime of Eld-Y1; S-face expanded 'x' occurs first in Mal Q3 (2), skips Fools, then recurs in Eastward (3). Although the 'k' reappears consistently across the sequence of texts, the sporadic absence of the 'g x' portion of the resident cluster could suggest that the fonts are not the same. The problem is amplified within the gatherings of a single text because non-reappearance seems much more frequent, especially with clusters of 2-3 letters. In Whore, for example, the 'k' skips H-K, the 'g' appears in all four (G-K), and the 'x' does not appear.[58] The 'w' of the Windet-F 'k w' cluster in Windet-S1 likewise skips F and I(standing) in both editions of Fawne.[59] Randomness is especially noticeable in regard to the capitals. Textual demands upon upper-case sorts vary considerably, even in play-texts, a fact which can lead to the non-recurrence of a fouling capital even though it is probably in the font. Eld-Y1 is fouled early on with S-face capitals 'D E G H K O P S' that recur fairly consistently in most texts from 1603 onward except for the 'E K P'. S-face 'K' first appears in Antichrist STC7120 (1603) ([2]B5:5) in a text with low demand upon the sort, is not seen in Mal Q1-2, but then emerges in later texts (Whore, G2:7; skips Mal Q3; Fools, B2:27; Eastward Q1, B1v:28, B2:28, E1v:28, F3v:37, I1v:8). The 'E P' are similar. The 'E' appears in Mal Q1 (B1v:14, standing E1:30), skips Q2, then appears in Whore (I1v:33), Mal Q3 (H1v:16), Fools (G3v:6) and Eastward (B4v:27, C2:24, I1v:13); the 'P' appears in Mal Q1 (B2v:16, standing D2:3), skips Q2, Whore, and Q3, then

appears in Fools (H2:14) and Eastward (C2v:13, I1v:32). It seems valid to assume that the 'E K P' sorts are continuously fouled from 1603-05 for two reasons, despite the fact that they skip texts. First, their eventual reappearance conforms to the behavior expected in the context of randomness. Second, identified foul-case capitals exhibit similar behavior.[60] The S-face type G1 in Eld-Y1, for example, appears in De Vnione STC13951 (March, 1604) (B1:29), skips several books, appears in Mal Q1 (July, 1604) (E1v:9), skips Siege of Ostend (September) and Whore (November), appears in Remaines STC4521 (November) (a1:7, e2v:4), skips two, appears in Mal Q3 (January, 1605?) (H3: 24), then Fools (?, 1605) (sig. G2) and Eastward Q1 (?, 1605) (C1:30, E1v:25, I2:7). The same rationale applies to low-density foul-case italic capitals in Short-Y. Low-density italic 'P' fouling occurs in R3 (L3:5, L3v:19, M1:1), skips 3H6 and reappears in Secretary (P2, O2:26, O2v:25, Q2:26, R3v:20, V2:3), and swash-italic 'P' in R3 (I1v:2), 1H4 (E3:38) and Secretary (F3:23). The frequent roman 'P' substitutions in speech prefixes in 1H4 had no effect on italic fouling in the sort, since the compositor distributed both 'P' correctly. It may cautiously be inferred from the recurrence of swash-italic 'P' in Secretary that he organized distribution according to textual units on 1H4 E3, where seven italic 'P' occurred in speech prefixes prior to the foul-case 'P' at E3:38, but the two types are not identifiable. 'F' is frequent in 3H6 (B2:25 and eleven others), skips R3 and 3H6, then reappears in Secretary (O4v:12, R1:33, Aa2:30); similarly, the 'K' fouling in 3H6 (B3:8, D6v:16) resurfaces in Secretary (F4:12). 'E' fouling in R3 (I3:36, I3v:4, L1:13) skips 1H4 but reappears in Secretary (Rr3v:21); 'B' appears in 1H4 (K1:25) and Secretary (H4v:4). Identified roman capitals in Short-Y recur in similar fashion. B1 appears four times in 3H6 (A3:24, B6:4, C6:18, D6:19) and 1H4 (B2v:17, C3v: 30, E1v:27, K3:14), once in R3 (H4v:19) but apparently not in Secretary. S-face T1 from the post-R3 replenishment occurs thrice in 1H4 (C1v:9, H4:33, I3v:18) and twice in Secretary (Ee4:9, Ff1:15).

Despite the probability that members of a low-density cluster will fail to reappear consistently as a whole, the odds are just as good that they will. Small capitals and low-density remnants of original capitals in fonts replenished with wrong-face capitals usually reappear, perhaps because the sorts were only partially replenished. The recurrence of normal 'W' and symmetrical 'M' of Fools EF in Whore EF and Courtesan EF illustrates the manner in which randomness influences both the numerical density of recurrences and their location within units of text, a tidy two gatherings in this instance. The normal 'W' skips E of Fools to appear four times in F (F1:16, F4:13,32, F4v:20); of the three in E of Whore, E3:21 and E4v:31 are distributed (E3v:24 left standing); a normal 'W' then is reset in Courtesan at E3:33; the 'W' at F4:24 is left standing. The appearances of the symmetrical 'M' are limited to three in E of Fools (E1v:19, E4:21, E4v:14), but four appearances occur in each gathering in Whore (E1v:18, E3:23,26, E3v:32, F1:20, F1v:12,21, F4:33); two in distributed E(o) do not recur in Courtesan. Even though the 'M' at F1:20 was distributed along with the rest of the page, Courtesan sets another symmetrical 'M' in exactly the same location—what are the odds on this happening?

However astronomical, another instance occurs in reset B in Fawne Q2, where the Y-face 'R' appears again at B1:32, accounting for the only two appearances of the letter in Q1-2. The few anomalous letters of Fools CD, I, reappear in d'Olive EF with random variation: the Guyot 'G' once per gathering in Fools (C2v:14, D2v:21, I3:18) but only once in d'Olive (F3v:34); the Y-face 'M' twice in Fools (C3v:18, I2:32) but twice in one gathering in d'Olive (E1v:1, E2:14, F3v:17); the 76mm 'Y' four times in Fools (D2:34, I1:21, I2v:25, I3v:2) but once in d'Olive (E3v:31). Finally, the small, low-riding 'W' in R3 (H3:38, H4:28) skips two gatherings before recurring (L1v:32, L2v:5).[61]

In short, the behavior of low-density foul-case letters must be approached in terms of a probable range of recurrence rather than in a simple numerical term, and the even probability of random non-recurrence must be borne in mind. In the instance of Eld-Y1, for example, the cluster members consisting of S-face 'D G H O S', Guyot 'S', and black letter query represent stable components of the resident population of their respective sorts. The principle of randomness indicates that, in theory, each or all may skip gatherings. But across several texts, they produce appearances per gathering consisting of usually 1-2 'D', 1-3 'G', 1-2 'H', 1-3 'O', 2-4 S-face 'S', 2-5 Guyot 'S', and about half of the time 1-3 black letter query. Even though individual letters in the cluster do skip gatherings, they all usually appear in texts of three or more gatherings. The same kind of behavior based on the probable range of recurrence can be noted in low-density clusters in Short-Y, Purfoot-Y, Windet-S1 and -S2, Danter-M, and Braddock-Y1.

The principle that the recurrence of low-density fouling is independent of pressure on a sort is implicit in the preceding examples, particularly in regard to reset portions of Whore and Mal Q1-2 ('S'), where sort-pressure is identical to that exerted in the original setting. However, the minute number of foul-case types noted in these examples obscures the randomness of recurrence in short units of text, such as the page or a gathering or two, that is characteristic of independence from sort-pressure. That randomness is more obvious in the recurrence of foul-case and replenished wrong-face letters that occur in higher proportions than low-density foul-case letters. Again, proportions must be viewed in terms of a range into which numbers are distributed. In general, fouling and replenishment of less than roughly 8-10% exhibits a randomness similar to low-density fouling in an experiential context, that is, the wrong-face letters are missing much of the time and it is necessary to search for them. A transition region occurs roughly between levels of 10-20%. As the proportion of a wrong-face letter moves upward through the region, recurrence exhibits an increasingly direct correlation between sort-pressure and frequency of appearance until the wrong-face letter behaves as a dependent variable. For example, Simmes-S was replenished with a Y-face 'g' and condensed 'p' but in different proportions. The random behavior of moderate fouling and low-level replenishment can be seen in recurrence of the Y-face 'g' across four texts. Hamlet Q1 (1603), Mal Q1-2 H, and Mal Q3 A-E are unrelated texts, while FG of Mal Q1-2-3 are essentially identical (give or take a few lines in Q3). The number of appearances of Y-face 'g' vary

in the sequence of gatherings, but as a variable that is independent of the total number of 'g' in the texts. The frequencies of the normal 'g' therefore are omitted in the tabulation (see Appendix, Table 1). The number of appearances is random, that is, is as likely to be low (1-2) as high (5-8). Moreover, the appearances in FG exhibit no correlation with respect to the three related and one unrelated texts. Although the 'g' consistently recurs in each sequence of gatherings set from Simmes-S, it frequently skips pages and is buried in disproportionate numbers of normal letters (Hamlet B, 3/131). This aspect of random behavior can be illustrated by a comparison of two fonts that are fouled in the same sort, a factor which levels any skewing that might arise due to the variable frequency of letters in an English text. The low-density S-face erect g2 of Fools EF and the Y-face 'g' of Simmes-S occur as shown in Table 2 (see Appendix), which groups appearances in related (Whore-Courtesan E and Mal Q1-3 FG) and unrelated texts (Fools E, Hamlet E) for each font. The randomness of appearance is obvious when the locations are examined in gatherings with similar totals. The two 'g' of Q1 F and Q3 G appear both early and late, the seven of Q2 G appear in only three pages while the eight of Hamlet E in 6 pages, only one 'g' occurs in page 3 of the ten gatherings, and so on. Likewise, the 'g' in related reset texts vary both in totals and locations. Fools EF differs from Simmes-S in regard to the slightly higher frequency of appearance of the 'g'. If both were S-fonts, this would provide a potential quantitative discriminant between the two. Other members of the low-level clusters in each would probably exhibit similar differences which, when combined with that seen in the 'g', would provide strong evidence that the fonts are not the same. The Simmes-S cluster, for example, includes the replenished 'p' and S-face x1. In contrast to the Y-face 'g', the replenished 'p', which accounts for about 22% of the sort in Simmes-S, exhibits a clear dependence upon sort-pressure across the sequence of gatherings in Hamlet Q1 and Mal Q3. Hence, a positive correlation obtains between the frequencies of appearance of the replenished and the S-face 'p'. However, the frequency of appearance within the pages of a gathering is influenced not only by sort-pressure, but also by additional factors such as the order of distribution and the alternation of the cases from which type is set. Thus the appearances according to pages can vary widely and appear random to a degree, although the relative densities of the two letters are constant for the most part (see Appendix, Table 3). Textual demands differ considerably, with seven gatherings of Q3 requiring more 'p' than eight of Hamlet. Although the sort-pressure swings radically among gatherings, the frequency distributions track rather closely. The slightly higher maximum in Q3 (25 in Q3 B, 21 in Hamlet G) is not significant statistically as an indicator of further replenishment between the two texts. Taken as a whole, the low-level 'g' and moderate-level 'p x' cluster could be expected to discriminate Simmes-S from other S-fonts. For example, The Supplication of Certaine Massepriests Q1 STC14432 (1604) was assigned to the English Secret Press by Woodfield (Surreptitious Printing) and the new STC although Simmes shared in several subsequent editions. The S-font composite of Q1 (A2-G2r) exhibits the same variants as in Simmes's sections of later editions, including the 'g p x' cluster. The 'g p' distributions included in the Tables leave little doubt that the font is Simmes-S.

Nonetheless, subsequent fouling or replenishment can shift the proportions of wrong-face letters in a sort and possibly cause confusion in comparing two fonts. For example, a moderate-level replenishment of Simmes-S with S-face 'p' after Hamlet Q1 would have this effect. In such a situation, Knowledge of compositorial fouling-purging habits as observed in other texts can support a hypothesis that progressive fouling is responsible for a shift in the number or the proportions of cluster members, as seen earlier in Creede-4. This kind of shift occurs in the progressive fouling of Windet-F with S-face 'w' between Spectres and Fawne. CD of Spectres exhibit ratios (S-face/normal 'w') of 11/187 and 10/213; the ratios in Fawne Q1 G(o) and reset portions of F (1, 2v-3, 3v-4) are 13/151 and 11/141.[62] The fouling behavior of the Windet compositor(s) in Fawne Q1-2 G, reset portions of Q2, and Sophonisba D, E4, F2, and G2v-3 leaves progressive fouling of Windet-F as the only plausible explanation for the increased level of foul-case 'w'. In another instance, progressive fouling in Eld-Y1 after Mal Q1-2 expanded the resident cluster with S-face 'A C L Y' that recur as usual in later texts. The additional letters first appear in Whore ('C': G4v:23; 'L': I1:38, I3:20; 'Y': I4v:15), then Mal Q3 ('A': H4v:12, I2v:39; 'C': H2:22; 'L': I1:14; 'Y': I2:4), and Eastward ('A': E2:12, F3v:16; 'C': C2v:22, F2v:37; 'L': E1:14, F2v:17). Finally, 'R' first appears in Eastward (A2:27, C2v:27, E1:36, F2v:22) toward the end of Eld-Y1's lifetime. Situations like this definitely require a survey of earlier texts to define the initial resident cluster and its rate of recurrence. Overall, the frequency of appearance of the original cluster is fairly consistent until Whore, when an increase can be noted along with the expansion of the cluster. That the two shifts occur together suggests progressive fouling of one font rather than two separate fonts. Eld's compositors were quite careful about fouling, setting texts in both Eld-S and Eld-Y1, and alternating the two fonts in Eastward Hoe! Q1-3, without producing anything more than low-density fouling in a few sorts, and successfully purging most of the foul-case letters from Eld-S after Sejanus. The shift in Eld-Y1 in Whore probably is attributable to the depleted condition of the font rather than compositorial habit. A different kind of shift occurs when large-scale replenishment modifies cluster composition. The resident cluster in Short-Y includes moderate levels of italic 'T' in 3H6 and R3, but the replenishment with S-face 'T' and purging of italic 'T' has a dramatic impact on the cluster. Similarly, introduction of the 'oo' ligature in Short-Y after 3H6 adds an obvious (although not wrong-face) letter to the font that is useful as a cluster member. Short-Y looks quite different as a result. A similar effect occurs in Simmes-S through normal attrition of wrong-face letters introduced by replenishment. The few Y-face 'M' in the 1604 Simmes-S (noted earlier) are remnants of low-level replenishment (4-7 appearances per gathering) around 1597 (see R3, A-G).

Finally, the relatively large number of recurrent identifiable S-face capitals in the Braddock-Y1 cluster from January, 1598, to October, 1600, reveals

the effect upon cluster composition and recurrence behavior exerted by the relationship between a periodic decision to purge unwanted letters and the size of a sort-population. It should be noted that compositorial intention to purge letters perceived as not belonging in a font applies equally to foul-case and heavily damaged letters. The residence of the identifiable S-face capitals is attributable to two factors: (1) the deliberately replenished wrong-face capitals were automatically tolerated; and (2), the extent of damage was inconsequential as a motivation to purge such letters routinely within or between texts: the very bold, heavily damaged A2 and gross T1-3 are brutally obvious on a small octavo page with 3-7 capitals at most, yet were tolerated in the cluster for two years.[63] The regularity and frequency of recurrence is striking, given the size of three of the sample texts (prose octavos, 3, 11, and 13 sigs.) with their extremely low demand on capitals and the slightly higher demand of one verse octavo (7 sigs.) as compared to the two mostly-verse play-texts (see Appendix). At some point in 1599, a decision to purge removed some damaged S-face capitals with differing impact upon cluster appearance. The cluster is unaffected by the loss of letters (such as A2, A5, A6) in sorts with high populations or the non-recurrence of A3 between Edward II Q2 (STC17483, 1598; F2:37) and MND (October, 1600; E1:13, F3v:19, G1v:13). However, the eventual purging of T1-3 (probably foul-case 96mm letters) dramatically affects the cluster, since these are the only oversized 'T' in the font. Similarly, the short-term transient 96mm 'C' apparently were purged at the same time with significant impact upon the cluster (A Short Forme STC12312, 1599, nine appearances). An earlier decision (early 1599?) to purge the large italic portion of the cluster had even greater impact upon cluster composition (see Appendix). Despite the effort, italic I1 (and possibly I2) escaped detection or were tolerated (I1: Ed.II, A3:13; C1:18, E2v:3, G1:33; Celestial Elegies STC 21225 [1598], C4v:6, D1v:10; MND, A3:13, E4:10 etc.), as was the italic Q. In the final analysis, a survey of a font across several texts or years is the only way to detect the kinds of shifts in cluster composition illustrated in the foregoing examples.[64]