7 What is reconstructed
The analysis works only from what is printed, and scores itself against the record afterwards.
procedure
(spelling-evidence b [names]) → (listof page-evidence?)
b : book? names : (listof string?) = '("A" "B")
Counts the test forms page by page and assigns each page to the workman it
agrees with.
procedure
(full-report b [r names]) → string?
b : book? r : (or/c press-run? #f) = #f names : (listof string?) = '("A" "B")
The whole report: bibliographical description, the scheme of imposition, the
stints recovered from spelling, the same evidence with the answers, the
skeletons recovered from damaged running titles, the casting off recovered
from crowded pages, the state of the case with its standing-type ledger, the
press variants — and MCKENZIE.
7.1 Backwards: the editor’s problem
Given a printed text, recover the copy behind it. This is what an editor does, and the module reports a word-by-word confusion matrix rather than a score, because the interesting result is where it fails.
Run from the Folio text of Much Ado against the quarto it was set from, the best blanket rule ceilings at 70–76 per cent. No per-word rule does better, because no per-word evidence exists. That is the honest form of Greg’s distinction between substantives and accidentals: he arrived at it by counting, and so does this.