Every page here is the Doubleday Complete Sherlock Holmes — the omnibus Doubleday, Doran published in 1930, gathering the four novels and fifty-six stories into one volume and opening with Christopher Morley’s preface, In Memoriam Sherlock Holmes, which stands on folios 8 to 10 of this text. It has scarcely been out of print since, and its pagination is the one most citations of the canon mean. That is why the page numbers here are its numbers: a folio between 8 and 1122 can be looked up in a copy on a shelf.
The scans. The words and the leaves come from two scans held by the Internet Archive: sherlockholmes0000unse for folios 8–480 and completesherlock0000unse_b9j7 for 483–1122. Two pieces of the Archive’s work made this possible and neither is glamorous. Its OCR gave the text of all 1,115 pages; and its page-number maps — which leaf carries which printed folio — are what let a line of prose be tied to the page it was printed on. Those maps were checked here against the printed running heads and agreed on 451 pages of 451 in the first volume and 612 of 614 in the second. An earlier scan of volume one had to be abandoned: it printed folios 289–320 twice and omitted 321–352 altogether.
The corrections. A scan is not a text. Each story was aligned against the sherlock-holm.es transcription, and where the two disagreed the shapes a scanner confuses — rn for m, li for h — decided which was the error: 2,022 scanner slips repaired that way. Where the two editions simply differ, as editions do — grey for gray, note-book for notebook — the scan stands, because the scan is the book; the uncorrected scan is kept beside it as a version of its own.
Two readings, and the difference named. The Doubleday is the authority here, but it is not the only text of these stories. Where the sherlock-holm.es transcription reads differently — grey for gray, note-book for notebook, arm-chair, cheque, enquiry, traveler — that reading can be put on screen from the Text control: 413 such places across 287 leaves. And a form only that text prints is still findable here. 141 such words are indexed by leaf, so a search for arm-chair or cheque answers instead of coming back empty, and the note says which reading has it and why the two differ — the editions hyphenate it differently, or spell it differently.
Deciding which of those it is was not left to the audit’s own labels, which are not clean enough to publish: more than half these rows sat in a pile it left for a person on purpose, and scanner damage was filed under its spelling heading — anda for and, wil for will, pear for dear. Each pair is judged on the two words themselves, and only two kinds survive: the same letters with one hyphen between them, or a spelling the table of orthographic patterns can get from one form to the other. A pair it cannot reconcile is two different words — warming against warning, led against let, from against form — and those are the scanner confusing letter shapes, or a single word nobody has adjudicated. 1,789 such places are dropped rather than dressed up as a reading.
What is not offered, at all, is the raw scan. What the photograph shows and the book does not is scanner damage — aftemoon, admirablel, My pear Watson — and that is not a reading of Sherlock Holmes but a photograph that came out badly. It is repaired without comment and never shown. This was built the other way round first, and it cost 107KB of payload and eight hundred and thirty-six indexed words to make the scanner’s mistakes findable. If a misprint is ever found in the book and corrected here, it goes on a list of its own and a reader can take it straight back off again; that list is empty today, because every correction so far has been the scanner’s.
The first printings, and where they differ. Every story but two reached print in The Strand before it reached this book; A Study in Scarlet appeared in Beeton’s Christmas Annual and The Sign of Four in Lippincott’s. The Variants control marks the places where what the magazine printed is not what the Doubleday prints — 1,449 of them on 690 leaves — and a marked word says what the other printing had. Nothing is changed on the page: the Doubleday still stands, and the mark is a note about it. Eight further marks are the book’s own misprints, which this text corrects and which the note names: the Doubleday sets unbrella, aqueezed, instand, choses. Where Moskowitz wrote a note on a place, the mark carries it in his own words — 28 of them, and some record things nothing else here knows: that the instand misprint on folio 645 “appears in older editions of Doubleday and was apparently corrected sometime after April, 1974”, which dates the printing the Archive scanned.
These come from Lester Moskowitz, The Canonical Variations: Doubleday versus The Strand (Battered Silicon Dispatch Box, 2012), and the debt is worth stating plainly: nobody here read two editions against each other, he did. It is not a complete collation and must not be read as one. His own preface says he left out most common spelling variations, capitalisation and minor punctuation, and that he was not consistent about even that. So a passage with no mark on it is a passage he did not record a variation for — which is not the same as a passage where the two printings agree.
His book also settled something about this text. Where the scan before correction read what he says the Doubleday prints, two witnesses to the leaf stand against the one transcription this project had been correcting against — and 20 of our own corrections turned out to be the error rather than the repair. Bengalore is back to Bangalore, Brackwell to Blackwell, and four leaves that had quietly lost I’m to I have it again. A further 42 places where his Doubleday and ours disagree are written down and unresolved, because each one needs a person to look at the leaf.
With thanks to Claude. The slip audit, the numeral and end-stop checks, and a good deal of this page were built in conversation with Claude, which turned up errors that had survived every earlier pass — and, in fairness, wrote a few of its own that had to be caught afterwards. The corpus is the better for the argument.