2.1 Casting off🔗ℹ

A house setting by formes must know in advance where the copy for each page begins, so the copy is measured out beforehand — by counting words and computing, not by eye. Verse casts off almost exactly; prose is much harder, so the strain falls where the prose is. What makes the practice workable at all is that early spelling is elastic: “In the days when abbreviations were to some extent optional copy was to some extent elastic” (Harry Carter).

The error has to run both ways. This closed a segment whenever the next unit would carry the estimate past the page, so a page was never allotted more copy than it held and every error ran short. Measured across the samples, every page came out spun out or exact and none was crowded.

That is not a small inaccuracy. It silently disabled three separate mechanisms downstream: the branch that drops copy for want of room, the report’s count of dropped lines, and the catchword mismatch — each of which then looked like working code that simply never fired.

The man marking up the copy judges by eye how much manuscript makes a page, and when the surplus looks small he commits it and is sometimes wrong. So a unit that overruns may still be taken in, with a probability that falls away as the surplus grows: a line over is easily missed, and the rare four- or five-line misjudgement is what actually costs text, because there is not that much white on the page to take out. He is wrong oftener with prose than with verse, which is Gaskell’s point and the reason slip exists.

On the Much Ado prose at the default accuracy this now gives twelve crowded pages against twelve spun out. Dropped copy stays rare — the crowding devices absorb most of the strain, which is what they are for — but it happens, and when it does the catchword left facing the gap no longer answers.

2.2 Justification🔗ℹ

Nothing in a line of type is elastic. The line must fill the measure exactly, so a compositor short of room strikes off a terminal -e, sets & for and, yᵉ for the, puts a stroke over a vowel to stand for a following nasal (thē), reduces a doubled consonant, abbreviates the speech prefix, and — if he is that sort of man — runs verse together as prose. A compositor with room to spare does the reverse.

The asymmetry is historical: the period’s default was the fuller spelling, so there was always more room to expand than to contract. This is why crowded pages abbreviate visibly while gaping pages merely look generous.

2.3 Spelling habit🔗ℹ

These are the discriminants Thomas Satchell isolated in Macbeth in 1920, which Willoughby extended and Hinman built the Folio’s stints upon. The split is crossed, not full forms against short ones: A is the doe/goe man but the here man. A workman who merely spelt fully would be undetectable, because justification alone would produce that.

Okes’s men, from Blayney’s Texts of King Lear, are in the same table: B sets -our where C sets -or (“the spellings ‘conquerour’ and ‘labour’ … strongly support … compositor B”, i. 159), and C “refused five opportunities to use an apostrophe” that B would have taken.

Habit strength is a per-word table rather than a single number, because one number cannot hold Hinman’s own figures: in quire L, A sets doe/goe 81 per cent of the time but here only 50 per cent. A scalar strong enough for the first is far too strong for the second.

2.4 The elided ending🔗ℹ

rul’d for ruled — the contraction the English trade actually used, and the one this program was longest without. Across five scenes of Much Ado the Folio has 114 medial apostrophes against the quarto’s 14, turning some thirty -ed endings into -’d. R. G. White noticed the practice without counting it: the Folio is “carefully printed for the day, even as to punctuation, contracted syllables, and capital letters” (Furness, Variorum, p. 293).

It does not elide after t or d, where the ending is a syllable that must be sounded: wanted and ended keep their -ed.

The instructive part is where it belongs. Put among the justification devices, it fired only when a line needed squeezing, and produced 15 apostrophes where the Folio has 114. It is a habit before it is ever a convenience, so it sits with spelling preference, and Okes’s C is the one workman who declines it. Because A, B and Okes’s B all elide, the form is no evidence of which of them set a page, and pattern-witness excludes it from attribution: a form several workmen would all have set testifies to none of them.

2.5 Stints, and how the frame changes hands🔗ℹ

This program used to change compositors every forme — a tidy alternation, and the one shop whose records survive contradicts it flatly. McKenzie, from the Cambridge Vouchers: when two or more men worked on a book “they did not work together setting sheet and sheet about. What usually happened was that one took over where the other left off and then composed as many sheets as the master found convenient or as other commitments allowed” (i. 107).

His quarto Virgil shows the shape. Bertram set A–E; Crownfield F–3G; Michaelis 3H–3Z; Bertram resumed to 4F; Délié set the single sheet 4G; Crownfield took over at 4H; and Bertram finished the book after Crownfield retired at 4O. Four men, long blocks, one man returning three times, and the odd single sheet dropped in where somebody was briefly free.

make-house therefore takes #:stint-sheets, the mean length of a stint in sheets, and builds a plan of contiguous blocks with a single sheet about one time in six. The next man is whoever is free rather than the next in rotation.

Left unset it follows the size of the shop, which is Gaskell’s rule (p. 41) and the thing that reconciles the authorities. Where there were “no more than two or three” compositors the tendency was for a man to concentrate on particular books “and to set at least whole sheets or whole formes” — McKenzie’s Cambridge, and long stints. Where there were more, copy went out in small “takings” or “takes” “to whoever was ready for them”, so that “the setting of sheets, formes, and even individual pages were on occasion shared”. A large house genuinely does approach the rapid alternation this program used to do unconditionally; the mistake was doing it for every house.

Honesty about the demonstration: the rule is implemented and the block structure does change shape with the shop — a two-man house opens with a twelve-page block where a five-man house sets even blocks of four. But on the 24 pages of the Much Ado sample the number of changes of hand comes out much the same either way, so this is an evidenced parameter rather than a measured effect.

The consequence for attribution is the uncomfortable part. Those boundaries fall where the shop’s other commitments put them, so — as McKenzie says — “the compositorial pattern within any such book will rarely have any internal significance”. It records the house’s other work, not anything about the book being analysed.

The nuance that partly rescues the method: the shop is chaotic, but a single book’s composition is usually “a simple matter of progression from sheet to sheet by consecutive compositors”. Stints are blocks, and blocks are findable.

2.6 The compositor’s freedom with his copy🔗ℹ

The expansion and contraction devices are not an inference. The manuscript copy for two Cambridge books survives beside the printed sheets, and Knell can be watched expanding q to que, fut: to futuro, salib: to salibus, p to per, Gram: to Grammat: — and contracting et to & (i. 118). McKenzie’s conclusion is that “a compositor was evidently not only free to expand or contract forms in his copy but that authors probably relied upon him to do so.”

That is direct archival warrant for the ampersand device and for the whole expand/contract ladder, from a shop that kept its records.

2.7 Space-metal, and why justification is quantised🔗ℹ

A gap between two words is a piece of type: a body cast a shade lower than the face so that it takes no ink. Em quad, en quad, thick, middle, thin, hair. They are picked from boxes, they run out, and they are distributed with everything else.

Two figures make the case for taking them seriously. A quarto page here holds 1,311 letters and 253 word-gaps, so 16% of everything set is white; and the thick space, the normal word space of the house, is as common in a fount as the letter e and needs a box about as deep.

The consequence is that justification is quantised. A compositor cannot make a gap of any width he likes; he can only set combinations of the bodies he has. This program spent a long time dividing the white by handing out single units of 1/120 em until the arithmetic came out — a body no founder ever cast — with the result that 86% of its gaps were widths no combination of real spaces could make. What a compositor does is Moxon (1683)’s account, and it is stepped: he sets with one space between words, and if the line will not fill he “puts a Space more between every Word”, and if still not, another. So a line fills the measure to within less than a hair rather than exactly, and that residue is real — it was taken up by the pressure of the lock-up.

Blayney (1982) makes space-metal the hinge of his whole reconstruction. Okes had never printed a play before Lear, and “what Lear used in the quantities most unprecedented in the pica books of 1605-7 was space-metal”. A play is short lines, quadded-out ends and marginal prefixes: it eats quads where prose eats letters. The program reproduces the asymmetry without being told to — provisioned from prose demand, Areopagitica never empties the em-quad box and Much Ado empties it outright.

2.8 The lay of the case🔗ℹ

The English divided lay, after Gaskell’s fig. 23 (reproducing Smith’s Printer’s Grammar, 1755). The essential thing is that the lower case is divided: two blocks with a gap between them, and no hand strays across the division. So n adjoins h and m, i adjoins the long ſ, p adjoins q — but e and i are not neighbours, though a single undivided grid would make them so.

2.9 Damaged and worn type🔗ℹ

Type wears. A fount bought new is clean; one that has printed twenty books has sorts with broken serifs, battered shoulders, split hairlines, and these injuries are peculiar to the individual piece of metal. That is what makes Hinman’s method possible: a particular damaged e recurring in two formes proves they were set from the same case within one distribution cycle.

So the pieces are tracked individually. sort-piece carries an identifier and its own damage, drawn from a vocabulary keyed to letter anatomy — a serif can break on a letter that has serifs, a bowl can fill on a letter that has a bowl.

2.10 Imposition and the skeleton forme🔗ℹ

Pages are printed several to a sheet, in formes. The running titles and furniture are not reset for each forme; the assembled skeleton is lifted from a printed-off forme and re-used, each running title carrying its own accumulating damage. A broken letter recurring at intervals therefore tells you which skeleton was used when — Hinman’s method.

The pages of a forme are not consecutive, and this is the fact from which most of the rest follows. A folio in sixes is three sheets quired one within another, so the outermost sheet carries the outermost pages (McKerrow, Introduction, ch. ii):

The general rule, which the tests assert for every format rather than for these cases only: on any sheet of a quired gathering a page’s fellow in the forme is (pages + 1) - p. For a folio in sixes that is 13 − 1 = 12.

2.11 The sheet, and the size it makes🔗ℹ

Format is how often the sheet was folded. Size is how big the sheet was. Neither gives a leaf a dimension on its own, and “quarto” says nothing whatever about how big the book is. For most of this program’s life only the first was modelled, so a leaf had no dimensions at all and the stylesheet was left to invent them — it drew a quarto at 1.99 tall to wide where Gaskell’s own figures give 1.31.

A fold halves whichever dimension is currently the longer, and the leaf is the result stood upright. That one rule is the whole relationship between format and size, and it reproduces Gaskell’s Key III (p. 86) exactly for pot, demy and royal in every format, which is what the test submodule asserts. From a foolscap sheet of 420 × 320 mm:

The proportion alternates rather than settling. A folio and an octavo are nearly the same shape and nothing like the same size; only the quarto is squat.

Foolscap is the default because Gaskell (p. 68) has the sixteenth century’s ordinary printing paper in that range, growing to demy by the eighteenth. –paper selects among foolscap, pot, crown, demy and royal; each carries a note recording where its figure comes from, because a sheet size is a citation and not a constant — the same name meant different paper in different countries and centuries, which is the whole difficulty of Gaskell’s Table 3.

MARGIN-CANON is '(2 3 4 6). The type page does not sit in the middle of the leaf: it sits toward the inner and upper corner, so that the two type pages of an opening read as one block and the tail carries the weight. This is the one number in the module Gaskell does not supply — a convention and not a measurement, and a parameter for that reason.

It is also where a finding sits that has not been tuned away. The canon only yields those proportions when the type page shares the leaf’s shape, and ours does not: on foolscap the quarto’s type page is 160 × 89 mm, a ratio of 1.80, on a leaf of 1.31. So the outer margins come out wider than the head and tail, which is the wrong way round for a hand-press book. Either the measure is too narrow for the paper or the stock is wrong for these books, and Blayney’s Okes quartos are where to settle it.

Watermarks, countermarks and chain-lines are not modelled. Note before adding them that in this period the mark will not give the size: Gaskell lists the sixteenth-century foolscap group as carrying the Strasbourg lily, the pot and the grapes indifferently, and warns (p. 68) that the marks “were not used exclusively for particular sizes, especially during the sixteenth century”.

2.12 Standing type, and what a gathering costs in metal🔗ℹ

Nothing can be printed until both pages of a forme stand in type, and type that stands cannot be distributed back into the case. So the order of setting has a price, and the program counts it. Setting the same play both ways:

A house setting straight through the copy cannot perfect the first sheet until page 12 is set, so eleven pages stand locked up and the case runs to 46 per cent empty. A house that casts off and sets by formes can print each sheet and return it to the boxes before beginning the next.

But that is a calculation, not a motive, and the archives do not support turning it into one. McKenzie went through the Cambridge Vouchers looking for exactly this practice and concluded that setting by formes “was followed occasionally but was certainly not normal” (Cambridge University Press, i. 115). Worse for the economic story, where work was shared he found the likely reason was the opposite of the one modelled here: “the principal reason for the shared setting was not to make more economical use of a limited supply of type but to find work for a waiting compositor” (i. 116). Labour scheduling, not metal.

But Hinman argues the opposite, for the Folio, and he is not guessing either. His case is that Jaggard simply did not own enough type: “shortage of type may have rendered it impracticable to set the book in the conventional way. If so, the size of the supply was one of the factors which made the alternative method desirable” — setting by formes needing “only enough type to set four pages”.

So the two authorities disagree, and the disagreement is not a muddle. They are describing different shops fifty years apart: a London trade house printing an unusually large folio against its type stock, and a university press with time on its hands and men to keep busy. Economy of type can be decisive in one and irrelevant in the other. What cannot be done is to take either as a general law of the hand press, which is what this program’s documentation was doing.

So the table above says what setting by formes would have saved a house that did it. Whether any given house did it for that reason is a question about that house. Hinman’s evidence that the Folio was set by formes is separate and stands on its own footing — it rests on type recurrence, not on economy — and McKenzie names that same test as the one that would have settled his own question: “One distinctive sort appearing in both formes of a sheet would of course outweigh all the ambiguous testimony of the Vouchers, for it would prove conclusively that one forme could not have been set until type from the other had been distributed.” At Cambridge the evidence was lacking. In the Folio it was not.

This program tracks the distinctive sorts that would constitute that evidence but does not yet reconstruct forme order from them. Until it does, the standing-type figures are arithmetic about a practice, not proof of one.

Capital B is the sort under most strain in Much Ado either way — Beatrice, Benedicke, Balthasar, Borachio against a bill of 224. That is the kind of pressure that produces a substitution.

The sorts figure counts every piece in a standing page, spaces and quads included, so it is a type-body count rather than a letter count: comparable between runs, but not directly against a founder’s bill in pounds weight.

2.13 Catchwords, and why they can disagree🔗ℹ

The catchword is taken from the copy, not from the next page. The compositor finished a page, looked at his manuscript for the word that came next, and set it in the direction line — before the next page existed.

McKerrow’s proof is the mismatches: we find “a correct catchword in cases where the opening words of the next page are wrong, owing to the compositor having mistaken the point at which he left off and consequently omitted or repeated a word or two. The catchword must therefore have been taken from the MS.” Hence his editorial rule — where a catchword disagrees with the page it faces, “the reading of the former may well be given the preference, for it was the earlier set up”. The catchword can be the better witness to the copy than the text it points at.

This program took the catchword from the next page’s first printed word, which guarantees the two always agree. add-catchwords now prefers the copy reading where the following page dropped any.

This was inert until the casting off was made to err in both directions (see Casting off). With that repaired the diagnostic appears where it should — at –cast-off 0.6 on the Much Ado prose, E4r catches but from the copy while E4v opens Change, because five lines went missing between them.

2.14 Signatures, and who puts them there🔗ℹ

Signing is the compositor’s act, not the imposer’s. McKerrow finds the Elizabethan men “normally finished a page of work at a time, adding catchword and signature (if necessary) before proceeding to the next one”, rather than setting long columns and dividing them into pages afterwards. So the signature in the direction line is put there by whoever set the page.

And the number of leaves signed was that man’s own habit. McKerrow: “there cannot be said to have been in early times any definite practice ... we may have anything from the first two to every leaf.” Hinman saw the same at Jaggard’s, where in a quarto one man might sign the first two leaves and another the first three.

This program used to sign half the leaves of every gathering, uniformly, because the format said so. Now each workman has his own count. Set one man to work and the book is regular; set five and it is not:

Which man signed how many, neither source says. Assigning particular counts to Hinman’s A and B would manufacture exactly the kind of evidence this program exists to test, so the count is drawn per workman from the run’s seed. The phenomenon is attested; the attribution is not.

Both the Bowers signing statement and the summary in the report now count what the sheets actually show rather than asserting the format’s rule, which is what a descriptive bibliographer does and what this program was not doing.

2.15 The preliminaries, and why they have a series of their own🔗ℹ

The front matter was printed last and bound first, and everything else about it follows. Gaskell: “the preliminaries were not included in the main signature series of new books because it was usual to print them last; reprints, however, sometimes began the main signature series at the beginning of the preliminaries” (p. 8). McKerrow from the shop floor: “in composing a new book from MS the normal course was to begin at the beginning of the text and proceed straight on to the end, setting up the title-page and preliminaries last” (p. 128). A compositor who has already signed his text A to L cannot give the front matter letters without collision, so he gives it a series of its own — and Gaskell’s exception proves it, because in a reprint the extent is known in advance and the separate series is not needed.

The forms, in Gaskell’s order of frequency (p. 52), are all in imposition: * ** ***; the symbols * † ‡ § “without logical order”; the main series from A with the preliminaries a b c, “always quite common”; and the main series from B with the preliminaries signed A, “a characteristically English habit … to allow for a sheet of preliminaries signed A”. Leaves that carry nothing at all are cited as McKerrow’s π (p. 156), “easily recalled by the p of ‘preliminary”’, which is a citation mark only and never reaches the direction line.

Gaskell gives the order and no numbers, and neither does McKerrow. PRELIM-SCHEMES weights them, and the weights are a guess at a distribution whose ordering alone is attested. Nothing in this program should be read as evidence for them.

A short preliminary gathering is half a sheet, worked and turned: “all the pages for a half sheet were imposed in one forme; this forme was first printed on one side of the whole sheet, then the heap of paper was turned … and printed from the same forme on the other side” (Gaskell, p. 83). One forme, not two, and one pull per two copies — which is why A2 is the commonest preliminary arrangement in Blayney’s checklist by a wide margin.

The overflow is the interesting case, and McKerrow reads it the way this program generates it. Of two editions of the Masque of the Gentlemen of Gray’s Inn, the first collating ?, A4, a4, B–E4, F2, he says that “even from the make-up alone we might guess that Ed. 1 is the earlier, for the work itself begins on B1 and this is preceded by A and a, the latter signature strongly suggesting that the preliminary matter was more than the printer had expected and allowed for” (p. 182). The second series is not a style. It is a misjudgement made visible — the house allowed one sheet signed A and the front matter would not go in it — and because the program knows whether the overflow happened, the inference can be scored.

2.16 Telling what the preliminary matter is🔗ℹ

The hard part is not the signing. Both authorities agree the question has no answer in the text: the boundary is a printer’s decision, not a property of the copy. McKerrow has the case. Tottel’s 1575 Treatise of Moral Philosophy puts its Table among the preliminaries; East reprinting it in 1584 began the text at C1 in imitation, “then found he had room for the Table in the last gathering of the book and placed it there, with the result that his preliminaries now only” half filled a gathering (p. 78). The same matter, in the same words, is preliminary in one edition and terminal in the next, and the reason is how much room happened to be left.

So divide-copy guesses, says that it is guessing, gives its evidence, and can be overruled. Marked-up copy is believed. Plain copy is matched against a closed vocabulary of period headings — the list is McKerrow’s “the title, dedication, preface, and, if there is one, list of contents” (p. 25) — and only where the heading stands before the text begins, because every one of those phrases occurs inside texts as well.

The confidences are not flat, and the differences are the point. The epistle dedicatory can hardly be anything else and scores 0.92. The names of the speakers scores 0.60, because McKerrow prints two editions of one masque in which it is preliminary in the first and the head of the text in the second. The argument scores 0.50, because it heads a preface in one book and every act of a play in another. A flat confidence would hide exactly the cases worth doubting.

Two caps, and the second was found by running it rather than by thinking:

• A walk that meets a heading outside the vocabulary stops there, which is what keeps a play whose acts are each called “The Argument” from being read as its own front matter.

• A block that grows past 2,600 words with no further heading to close it is abandoned whole and given back to the text. Plain copy gives no signal where the last block ends, so before this a table of contents ran on to the end of the book. Cutting it short instead would have put half a table among the preliminaries and half at the head of the text; the block boundary is the one thing the copy really tells us, and the extent is only a bound taken from what front matter is like elsewhere.

Whether the Table then goes to the back is decided by arithmetic on two make-ups rather than by a rate: is there room in the white leaves the text has already left, and does moving it save leaves at the front? Both yes and it moves, which is East; either no and it stays, which is Tottel. The report says which, and why, in both cases — because “nothing moved” and “there was nothing that could move” are different facts about a book.

2.17 The title-page🔗ℹ

Supplied by hand, a title-page would be a cheat at the most formulaic thing a hand-press book contains, and formulae can be measured. Blayney’s Appendix II is a checklist of about ninety books from one London shop between 1604 and 1609, each with a substantive transcript. Counted over those transcripts:

Two things the transcripts settle that guesswork gets wrong. The address is the bookseller’s, not the printer’s — “Printed by N. O. for Roger Iackson, dwelling in Fleetstreet neere to the Conduit” is Jackson’s shop — and it belongs to the printer only where there is no bookseller to own it. And “dwelling in” and “and are to be sold at his shop in” are two ways of saying the same thing, which is why they come out twenty and twenty.

The date is quadded: the last line of an imprint is short and the figures were spread to fill it, which is why about half of them read 1 6 0 8. rather than 1608. The spaces are metal, and this program charges for them.

The page is handed back as copy, not as a page, for two reasons. It must go through the same compositor as everything else, so that its spelling, its long s and its accidents are his rather than the program’s; and it must be settable at a moment of the run’s choosing, because Blayney found the Lear title-page was “the first part of the book to be set” and then distributed. Set first, printed last: the awkward case for the type accounting.

The type is charged to the text case, which is wrong and is reported as wrong. A real title-page drew on titling founts kept apart from the body fount — Blayney records which line came from which — and typecase keeps one case. The error is about forty words a book, all of them large.

2.18 The last sheet, and what becomes of the white paper🔗ℹ

One sentence governs the end of every book:

“as it costs practically as much to print part of a sheet as a complete one, it was always to the printer’s interest to make up a complete sheet whenever he could.” (McKerrow, p. 159)

So a text that stops two leaves short of the end of its last sheet, in a house with two leaves of preliminaries still to print, does not print a separate half-sheet and leave two leaves white. It prints the preliminaries in the white leaves and cuts them out:

“he will as a matter of course impose these preliminaries in the middle of his last sheet, which may therefore run, as actually printed (supposing it to be in fours), Z1, [*], *2, Z2, the two centre leaves being cut out to be used as preliminaries. Such a book will be described as *², A–Y⁴, Z², quite correctly.” (p. 158–9)

And where the front matter is a single title-leaf: “the printer would be quite likely to print it Z1, Z2, Z3, [—], cutting off his last leaf to form the title.”

A gathering is therefore printed whole and bound short. page-refs takes per-leaf signature overrides so that one sheet can carry two series at once, and plan-bound-leaves is what the collation formula counts. McKerrow’s reason for pressing the point is that the alternative makes editors record leaves that never existed — “*1 and Z4 wanting, probably blanks, thus inventing two blank leaves which in fact never existed” — which is exactly what this program used to produce.

Cut from the centre the preliminaries come off as a conjugate fold; cut from the tail they come off disjunct. Which of the two is recorded, because it is the fact Bowers used to prove the case: of Sandys’s Ovid he observes that the printed preliminary leaves “are always disjunct and have any watermark on the outer edges of the two leaves, an impossibility if they had been printed as a fold in the cut-off.” The paper that would betray it is not modelled; see the roadmap.

It is a tendency, not a law, and the authorities guard it from both sides. McKerrow: “we must not assume that a printer would in every case economize his labour and paper in this fashion: it might sometimes have been more convenient to have the two extra leaves as covers or end-papers.” Bowers: “Even when normal printing practice might lead one to expect economical machining without blanks, it is dangerous, lacking proof, to assume their absence.” Hence CUT-OUT-SHARE, and white paper otherwise.

2.19 Cancels, and why the cause is not simulated🔗ℹ

A cancel is a leaf cut out and another pasted to the stub. The obvious objection to simulating one is that they happen for reasons no printing house can generate — the Privy Council took exception to Eastward Ho. McKerrow gets there first and makes the decision:

“Into the purpose of these cancels we need not enter. There may have been in the original print something so grossly incorrect that it was too much for even the easy-going printer of the day — or for the author; or, as often in early times, there may have been something that the authorities found objectionable. The point at present is the aid that bibliography gives us in detecting them.” (p. 223)

So the cause is a parameter and the trace is a simulation. Three causes, of which only the first is modelled in the strong sense:

• An error the program made itself and its own corrector missed. The run knows what the error was, which page it is on, and that the proof went by without it, so nothing is supplied from outside. –cancel-rate.

• A change of imprint — the same setting with the bookseller’s name altered, which is why a cancel title is commoner than any other kind. –imprint-change.

• Anything else — –cancels N, a count and not a model. The report says so in those words.

The trace is complete, because that is the part that can be got wrong. The leaf is cut out “leaving a stub of paper to which the new leaf could be pasted” (p. 223). The replacement is printed in the white paper at the end of a gathering — Gaskell has Rousseau’s publisher writing “to encourage the author to use up the blank leaves of final sheets for printing cancels”, and McKerrow that “spare leaves at the end of a book were used to print matter that was to be bound elsewhere in it, such as titles or cancels” (p. 156) — or it costs a half-sheet of its own, which the report distinguishes because it is the expensive case. A cancellans is a leaf, so the paper it can use must be blank on both sides.

And mckerrow-signs generates five of McKerrow’s six detection tests (p. 224) as properties of the particular leaf, so that the analysis half can one day be scored on finding them: the setting of headline, signature or text differing; a different number of lines to the page; a signature where its fellows carry none; a part-signature away from the first leaf of its gathering; two press figures in what should be one forme. The sixth is the paper, and is not claimed.

Bowers’s caution holds throughout: “It may be taken as an axiom that no blank not interrupting continuous text would be torn by the printer for excision.” A cancel is a deliberate act on a printed leaf. Blanks stay.

2.20 The heaps, and why a copy is not a random draw🔗ℹ

Two authorities meet here, and between them they turn the making-up of copies from a shuffle into an inference.

Gaskell gives the mechanism (pp. 143–4). The heaps stand in signature order and are gathered from the top of each. For a sheet perfected inner forme first they come off “in the reverse of the printing order, so that the first book to be gathered contained the last printed sheets”; for one perfected outer forme first the heap “had to be turned over to show the first page of the signature, which brought the first-printed sheet to the top. This heap was then gathered in the printing order”. So the copies lie in one linear order, each variant divides that order at the point the corrected proof came back, and which side of the division is corrected says which forme of that sheet went to press first.

Greg gives the test. His calculus assumes simple transcription, and warns that where “the grouping is throughout random or if inconsistent forms are of frequent occurrence, the relationship of the manuscripts cannot be accounted for on the hypothesis of simple transcription; some sort of conflation has somewhere to be assumed” (The Calculus of Variants, p. 43).

A made-up copy of a printed edition is conflation by construction. It descends from no other copy; it is assembled from as many heaps as there are sheets, which is Greg’s many-one relation exactly. Drawn independently the groupings cross. Gathered as Gaskell describes they are prefixes and suffixes of one order, hence nested or disjoint — which is precisely Greg’s condition for consistency: “given any two constant groups, either these or their complements are either mutually exclusive or one wholly includes the other” (p. 12).

variant-groupings returns the grouping each press variant makes, and greg-consistent? applies the condition. Over 25 runs of ten copies:

The last row is what this program did until the heaps were modelled — and the fact that it fails Greg’s test is the point, not an embarrassment: the test is a detector for the very thing the code was getting wrong.

The parameter itself carries no authority. Gaskell hypothesises “a case of remarkable regularity” and hedges it in the same breath: after drying, “the chances were that … the sheets would be in the same order as before, although this was not certain to happen.” How much order the rack destroys is a knob, and the report says so.

What is not built, and is the next real exam for the analysis half: making the inference rather than showing the answer. The direction of each grouping says which forme of its sheet went to press first, and this program knows the truth.

2.21 Gathering, folding and the binder’s errors🔗ℹ

The one stage at which the book diverges from the printing. Everything before it is the same for every copy of an impression; from here on the copies are individuals, and a bibliographer describing one has to tell the faults of the copy from the faults of the edition.

Two hands in two places. The warehouseman gathers, in the printing house: the heaps “were set out in signature order on a long table, with the first recto pages upwards and to the near side”, and the gatherer “took off the top copy of the last sheet of the book and then walked along the line of sheets, taking off one copy of each in turn” (Gaskell, pp. 143–4). The books “were then collated to ensure that each was made up correctly”, and only then folded, pressed and baled. The binder folds and sews later and elsewhere.

Between them they can drop a sheet, take two, put one in backwards, or sew them out of order — and the whole apparatus of signatures exists to stop two of those:

“It was necessary, when assembling the sheets of a book, to get them the right way up and in the right order; and to this end each sheet was signed on the first page with a letter of the alphabet so that they could readily be arranged in alphabetical order; similar signatures were also placed on the rectos of a few leaves after the first of each sheet in order to help the binder with his folding.” (Gaskell, p. 79)

That sentence is the design of binding. The kinds of fault come from the sources, as does the fact that made-up books were collated before they went out. The rate does not. Neither Gaskell nor McKerrow gives one, so BINDING-ERROR-RATE is an explicit parameter carrying no authority whatever, and the report prints the disclaimer beside every fault it lists. It is a knob, not a finding.

What is not invented is the shape. An unsigned gathering offers the binder no help with either the order or the way up, so it goes in wrong oftener — and survives the warehouse’s own check oftener, because the check is the signatures too. A preliminary series signed π is therefore the most misbindable thing in a book, which is a consequence of Gaskell’s sentence rather than an invention on top of it.

2.22 The corrector🔗ℹ

Before copy reaches the frames it is read through and marked up — capitals altered, house spellings imposed, page endings crossed. So the compositor does not set from what the author wrote, and a spelling that differs from the manuscript may be the house’s rather than any workman’s. Because house habits fall evenly across every stint, no amount of counting can separate them from a compositor’s own practice.

2.23 At press🔗ℹ

The press does not stop for a proof. At three or four impressions a minute and fifteen to thirty minutes for the reader, some 60 to 120 sheets of an edition of about 1,200 were printed before the marked proof came back — so the uncorrected state is typically 5 to 10 per cent of the copies. The corrector generally worked without the copy, so he catches foul case readily and misreadings hardly at all, and sometimes he made it worse.

Whether the corrector had the copy in front of him is a question with two answers, and this module used to give only one. Moxon describes the proper method: the master-printer appoints someone “well skill’d in true and quick Reading, to Read the Copy” aloud while the corrector follows the proof. On that method a misreading is as catchable as a turned letter.

Hinman found the Folio was not corrected that way — “the copy was in all probability seldom if ever used to correct perfectly obvious mistakes” that the context would yield sense for. But seldom is not never, and he can point to the corrections that prove the copy was sometimes at the reader’s elbow: a two-line speech that “cannot have been restored save by reference to the copy”, and a line bearing no resemblance to the one it replaced.

So consults-copy governs how often the copy is called for, and the two methods fail differently. Sense catches foul case and leaves a plausible misreading standing; the copy catches the omission that sense cannot see. Hinman also noticed the consequence of a scare — having found a considerable omission, the reader grew “somewhat more careful ... at least for a time” — so a serious catch raises vigilance for the next forme.

Building this turned up a bug of some standing. Misreadings are recorded when the compositor reads his copy, before the word is placed, so they carry no page or line and the press loop never saw them: no misreading was correctable by any method, and catches-misreading did nothing at all. They are now found on the page, where the copy reading and the read reading disagree.

first-proof is the chance a forme was proofed and mended before the pressmen began. Such corrections leave no variant: every copy shows the mended reading. So the catalogue of press variants, however many copies are collated, records only the corrections made too late.