One is the loneliest number

A publisher can trust one translator’s taste or one model’s output — and hope. Or a publisher can build a translation the way peer review built science: draft, grade, refuse, revise. The council exists because one voice cannot hear itself. The method is expensive in compute and in care. It is also the only way to produce a translation whose every line has been measured and whose failures are recorded as plainly as its scores.

The name Seventy-One is a bow to the Septuagint, the ancient Greek translation of the Hebrew scriptures. The legend holds that seventy-two scholars were sent to Alexandria to translate the Torah — six from each of the twelve tribes. But the name the world remembers is “the Seventy,” the round number that stuck. We add the one: the council is never a closed room.

Three seats

The point of a council is disagreement that resolves — rancour without discord.

The council seats multiple intelligences — each a different architecture, each trained to differ deliberately. When several systems with nothing in common but the Greek converge on the same line, that convergence means something no single oracle can certify. The seats are only the beginning: behind them sits a quiet engine of specialist agents and structured retrieval that feeds the council its sources, tracks every reference, and records every refusal. What the reader sees is a finished line. What the press holds is an audit trail.

3
Council seats
337
Sources read
905
Reference passages indexed

For example: The Odyssey

The craft

Three stages to a line

Stage one

Faithful Draft

The council drafts the line from the Greek with a single priority: say what the original says, nothing added, nothing lost. Unapologetically literal. Beauty is not yet the point.

Stage two

Expressive Refinement

The faithful draft is handed to a second pass that lifts it into living English. Here the music enters: rhythm, cadence, image, the sound of a sentence that belongs in a poem.

Stage three

Elegance Polish

The refined line is tuned a final time for the qualities that separate literature from competent writing — the exact word, the held beat, the turn that surprises and then seems inevitable.

The rubric

Named axes, weighted

Each line is graded on weighted axes. Literary Quality carries the heaviest weight, because a translation that is faithful but charmless has still failed to translate Homer.

30%
Literary Quality

The heaviest single weight. A translation faithful but charmless has still failed to translate a poem.

25%
Fidelity

Beauty built on drift is beauty built on sand. The line must carry what the Greek carries.

15%
Scholarly Accuracy

Names, places, the grammar of the original — held to the standard of the commentaries.

15%
Voice Consistency

A translation that changes register mid-poem has two voices. The council holds to one.

15%
Accessibility

A reader who does not know the Greek should still hear the poem clearly, the first time.

The proof

Every number is a record of what the council agreed on — and what it refused.

0.956
Mean convergence
24
Books complete
128k
Words
21/24
Books above 0.93

The press ships the scorecard alongside the book so the reader knows exactly what was tested. A scorecard earns trust by reporting the refusals as plainly as the passes. Book 24 of the Odyssey was entirely missing from the source corpus — the poem had no ending. The council generated it at 0.95 convergence. The poem ends. It did not, before this method. It does now.

Open The Odyssey

Downstream

What the council makes possible

The council’s real product is structure. Every chapter the press ships is held as clean, measured, portable text — the same files the reader renders, the same files the scorecard graded. That structure is what lets one book take every form reading now takes: a page that stays quiet, a voice that narrates the chapter, a chat that answers against the text itself, a bundle your own tools can open, an endpoint your agents can query.

Craft upstream is what makes the living book downstream worth having. A chat grounded in a measured translation answers differently than one grounded in a scrape.