9 The sources🔗ℹ

Everything this program does is drawn from somewhere, and the sources are not of one kind. The distinction matters more than an alphabetical list suggests, because it decides how much weight a claim will bear.

The manuals — Moxon (1683), Smith (1755) — were written by printers for printers. They are the only sources that describe the work from inside, and they describe it as it ought to be done. The lay of the case, the quantities of the spaces, the reaching, the reader who speaks the copy aloud: all of that is Moxon’s, and where this program models a procedure rather than an outcome it is usually following him.

The archives — McKenzie (1966), and the proof-sheets behind Simpson (1935) — are records made at the time for other purposes, which is what makes them evidence rather than inference. Nearly every correction this program has had to make came from the first of them. Its production times, its compositors’ output in ens, its finding that setting by formes “was followed occasionally but was certainly not normal”, and its Vouchers showing that men took over from one another in long blocks rather than alternating, between them overturned four separate assumptions in the code.

The analyses — Greg (1955), Hinman (1963), Bowers (1949), Blayney (1982), McKerrow (1927), and behind them Satchell (1920) — are reconstructions from the printed books themselves. This is the New Bibliography, and this program stands in an odd relation to it: it implements the method in order to test it, and reports where it fails. Hinman (1963) is the largest single debt — the compositor spellings, the type-recurrence method, the proof-reading, the pagination errors, and the caveat about justification that the whole method turns on.

The objection — McKenzie (1969) — is the paper this program cannot answer and does not try to. See The objection.

The texts and data are Best (1996) for the old-spelling transcriptions used in calibration, Furness (1899) for the collation they are measured against, Mulcaster (1582) for what the period held correct spelling to be, and Partnership (2015) for what was actually printed. Baron and Rayson (2008) is not used but solves the same variant-grouping problem, and its design settles a question met here independently.