9. The long work

The charter asked for one long work, translated serially, "to surface problems short pieces never do: sustained voice, cumulative characterization, motif and terminology consistency, fatigue of method." The project translated six, in six languages, and closed each with a craft report. A seventh — a Serbian short novel of 1881 — was chosen on the last scheduled day and never begun.

The serial regime adds six rules to the ordinary close rendering: spans defined by paragraph index into a stored source; an append-only text, revised only "through a numbered, dated erratum"; a cumulative log; a binding register of names, terms, voice rules and decisions, with a list of questions the next span must settle; a cadence rather than a deadline; and the rule that "decisions made by drift are recorded as such." Its own statement of limits proved prophetic: "the register can become an alibi."

What only length surfaced

A register that is blind backwards. On Canth, the first span was re-rendered blind under the register as it stood at the end of the book: of 234 places where the two renderings differed, thirteen were attributable to something learned later, "and eleven of those thirteen are sites the closed register would now render differently and which no erratum ever caught." The craft report calls this "the serial method's structural defect" and immediately bounds its number: one hand rendering the same source twice "parts from itself in almost as many places as a stranger does," so what survives is eleven specific sentences. On Mikszáth the count was repeated, with a further caution: a two-paragraph argument in the log for one rendering "produced the choice the habit had already produced," and the record could not tell reasoning from recall.

A lexical chain has to be opened at the first occurrence. The Bengali stem meaning "mad" occurs nine times in The Post Office and the translation renders it three ways; the report does not repair it, because a chain "has to be opened at the first occurrence or it cannot be closed at all." On Mikszáth the principle produced the arm's best rules: the villagers who have never seen an umbrella call it an implement, a canvas platter, a canvas tent — "normalising any of them to umbrella deletes the fact that these people are looking at an object they have no name for" — while a returning verb must be given the same English every time although "English has a locally better word at each of those three sites … and taking any of them dissolves the return."

The source is a variable. Three free e-texts of Canth turned out to be one transcription of one posthumous reprint, and collation found five corruptions, one of which turns she would not willingly tell it into she would not tell it to her man. For the Nights, the two Arabic witnesses differed by "an obscene verb silently censored, twice" and 190 words missing at one span; the report's first conclusion is that "the source is a variable, and the collation is part of the translation." On Tagore, 335 words in twenty-four speeches were missing from the project's own copy, and only length found it.

The founding premise can shrink. The Tagore arm was built to study the loss of Bengali's graded second person; the whole-play census found ten such sites in 430 units, and nine minimal pairs put to three arbiters came back "same social relationship" in sixty cells of sixty. The loss was "real, total, and small."

A rule has a standing price. At the Nights' eighth span, "of four sound-figure losses, three were the price of terms frozen in earlier spans." Its refusal to regularize the night boundaries the manuscript places carelessly "turned out to be the arm's most productive single decision" — the formula turned out to place the seam — and the report generalizes: "an inconsistent source is an experiment; regularising it destroys the evidence." The sage who is Dubān in every English Nights became Ruyan, because that is what the witness says, a choice the report says "must be made in public."

A reviser who has read the whole book is not uniformly better at its first page. Canth's second pass made things worse twice, collapsing sour tears into ordinary ones. The long-works entry tells a translator to treat the second pass as a set of corrections rather than a fresh reading.

What length could not show

Two things. Affect, the cumulative experience of a reader — the charter's own reason for wanting a long work — was returned as a null: "a translator's log records choices and affect is a property of reception, so a serial-translation regime has no way of making a cumulative effect visible to the person producing it." And quality: every craft report repeats that the jury never scored a long work. What the six left behind is a method — a binding register, an erratum discipline, a collation habit, a census-based alarm — and a measured account of its price, which section 14's framework carries forward for anyone translating a whole book with a system like me.