We claim these translations are original and accurate. Here is the arithmetic behind the claim — run it yourself; the method is described in full.
We break every translation into overlapping three-word phrases and compute how many are shared with the closest previous translation of the same text (Jaccard similarity — the standard plagiarism-detection measure). For calibration, we apply the identical measure between the classic public-domain translations, which nobody disputes are independent works.
| Comparison (Iliad, Book 1) | 3-gram phrase overlap |
|---|---|
| Plainspoken vs. Butler (1898 prose) | 4.7% |
| Plainspoken vs. Cowper (verse) | 1.1% |
| Plainspoken vs. Pope (verse) | 0.5% |
| Baseline: Butler vs. Cowper (independent classics) | 1.3% |
| Baseline: Butler vs. Pope | 0.6% |
Our prose sits a few points above the verse baselines for one honest reason: we and Butler both translate the same Greek faithfully into prose, and Greek dictates some renderings ("the tables of the money changers" can only be phrased so many ways). What matters is the order of magnitude: a paraphrase of an existing translation scores 40–60% on this measure. Ours scores 4–8%.
Phrase statistics can hide copying, so we also scan for the longest contiguous word-for-word run each paragraph shares with the prior translation. In 240 randomly sampled paragraphs across Josephus and Philo, the longest run found anywhere was 14 words — and the flagged runs concentrate where convergence is forced: quoted scripture (Philo quotes Genesis constantly; two honest translators of the same verse converge) and fixed formulae. For contrast, our internal standard elsewhere in the corpus is stricter still: in a 2,211-verse test we mechanically eliminated every 8-plus-word run against twenty-one prior translations.
Every ~15th translation batch is audited by an independent AI referee that re-reads the Greek and hunts for mistranslation, omission, and invention, quoting the Greek for every check. Across all sampled audits of the volumes on sale: zero fabricated content, zero omissions, and a small number of flagged stylistic notes (each fixed). Sample referee finding, verbatim: "v41 Greek ὀργισθεὶς ('being angered') is rendered 'Moved with anger,' correctly following the harder critical reading, whereas the traditional translations render the easier variant 'moved with compassion' — the new translation tracks the given Greek."
We commissioned a blind comparative review of six famous passages (our Homer vs. Butler and Butcher-Lang, against the Greek). Verdict: accuracy "never beaten, occasionally sharper"; a genuinely different and defensible reading at the epic's most scrutinized word (πολύτροπος, Odyssey 1.1); and a consistent house voice — epithets preserved, archaism stripped, Greek syntax-shapes kept at emphatic moments.
Method summary: AI fleets translate directly from public-domain Greek texts (Perseus, Niese, Cohn) and never see an English translation; independent AI referees audit random batches against the Greek; finished text is mechanically scanned for verbatim overlap with every prior translation we hold. The Greek is always the authority.
← back to the library