Sacred Translation Standard

Rule 6 — Translate Meaning, Not Words Alone. Literal translation is the default. When literal wording obscures meaning, preserve the author's meaning before preserving individual lexical choices.

Rule 7 — Preserve Literary Devices. When possible, reproduce: rhythm, parallelism, repetition, alliteration, irony, rhetorical questions, wordplay. If a literary device cannot be reproduced, document it.

Rule 8 — Maintain Terminological Consistency. Technical theological vocabulary should normally receive consistent English renderings unless context clearly demands otherwise. Maintain a translation glossary.

Rule 9 — Preserve Ambiguity. Where the original intentionally allows multiple interpretations, avoid resolving the ambiguity unless context requires it. Commentary belongs in notes, not in translation.

Rule 10 — Read Aloud Test. Every translated paragraph should be read aloud. If it cannot be proclaimed naturally, revise the English.

Rule 11 — Honor the Reception Tradition. Consult how the text has historically been understood within its tradition. For Catholic works this includes: the Latin tradition, the Fathers, the Doctors, the Councils, and liturgical usage. Reception history informs translation but does not automatically determine it.

Rule 12 — Preserve Technical Theology. Avoid replacing historically significant theological terminology with contemporary paraphrases unless meaning would otherwise be lost.

Rule 13 — Distinguish Translation from Interpretation. The translator should transmit the author's thought. Interpretive explanation belongs in notes.

Rule 14 — Intellectual Humility. Where uncertainty exists: say so. Do not create false certainty.

Rule 15 — Every Decision Must Be Explainable. Every non-obvious translation choice should have a documented rationale. Future editors should always be able to answer: Why this wording? Why not another? Which sources were consulted?

Rule 16 — Human Responsibility. Technology — including AI — is an editorial assistant. A qualified human editor bears responsibility for every published sentence.

Rule 17 — Preserve Editorial History. Every revision should record: previous wording, reason for change, supporting evidence, approving editor.

Rule 18 — Translation Is Never Anonymous. Each edition should identify its translators and editors. Authority comes from accountable scholarship.

Rule 19 — Revision Requires Evidence. No wording should change merely because it "sounds nicer." Every revision should improve one or more of: fidelity, clarity, literary quality, consistency.

Rule 20 — The Translation Should Endure. Prefer English that is likely to remain understandable and dignified for generations. Avoid both unnecessary archaism and fleeting contemporary idioms.

Rule 21 — The Principle of Restraint. When the original achieves greatness through simplicity, resist the temptation to embellish.

Rule 22 — The Principle of Charity. Assume the author wrote intentionally. Before concluding that a passage is awkward, repetitive, or redundant, first ask whether those features are deliberate.

Rule 23 — The Principle of Transparency. Readers should be able to distinguish: what the author wrote, what the translator inferred, and what the editor explained. These should never be conflated.

Rule 24 — The Principle of Literary Equivalence. Seek to reproduce not merely the information conveyed by the original, but also its literary experience. When the original is majestic, the translation should be majestic. When it is terse, the translation should be terse. When it is lyrical, the translation should be lyrical. When it is severe, the translation should be severe.

Rule 25 — The Translator Is a Steward. The translator is neither an author nor a commentator. The task is to receive faithfully, understand deeply, and transmit clearly.

Rule 26 — Originals First. The base text of every translation is the original language (Greek, Latin, Hebrew, Syriac) in a NAMED public-domain edition — e.g., Migne's Patrologia Graeca/Latina, the Clementine Vulgate, pre-1929 critical editions. English is translated from the original, never from an intermediary translation. Where no usable public-domain original edition is accessible, the work may ship in the Modernized tier (Rule 30) — never silently. Where an author himself worked from an intermediary (e.g., Aquinas quoting Greek Fathers through Latin versions in the Catena Aurea), we translate the work the author actually made and document the chain.

Rule 27 — Witness Triangulation. Every translated unit is checked against at least one further English translation (Newman, Waterworth, Percival, Pusey, ANF/NPNF, and their peers) serving as a witness. Witnesses refine; they do not govern — the original governs meaning, the reception tradition (Rule 11) informs it, and the witness catches misreadings in both directions: the machine's errors and the Victorian translator's. Material divergence from a received rendering requires a documented rationale (Rules 15, 19).

The witness is withheld from the drafting context (v1.2): it is consulted only after the draft and its fidelity check against the original are complete, and by a different reader than the one that drafted. This is a procedural guarantee, not a claim of independence. A public-domain witness is, by the very fact of being public domain, likely present in the training data of any machine-assisted pipeline — being public domain is what causes a text to be digitized at all, so this rule's own selection criterion favours the witnesses least independent of the drafter. No procedure undoes that. What this rule guarantees is that the witness text was not in front of the drafter.

An in-copyright translation may serve as a consultation-only witness — consulted, recorded as consulted, never reproduced (founder ruling, 2026-07-31). Note the inversion: such a witness is often more independent in the training sense, being less digitized. Where one is consulted, close agreement with its distinctive phrasing is a risk to be reviewed, not a success signal.

Rule 28 — Edition Provenance. Every unit records its exact source edition and locus (e.g., "PL 76, col. 1302"; "PG 57, col. 21"; "DS 1652") and the edition's own rights status. Modern critical editions whose apparatus remains under copyright are not used as base texts unless licensed, and are never quoted silently.

Rule 29 — The Dogmatic Precision Tier. Conciliar definitions, canons, anathemas, creedal formulae, and other dogmatic loci receive the strictest handling: the original language is displayed alongside the English at all times; technical terms are rendered per the Denzinger reception tradition; traditional English formulations are preserved unless demonstrably deficient; and any change at this tier requires the strongest documented evidence and named-editor signoff. A modern rendering of a dogmatic text never appears without its original.

Rule 30 — Two Labeled Tiers. Every published text declares its tier: "Translated from the original (edition, witnesses consulted)" or "Modernized from [translator, year]." The reader always knows which. No unlabeled hybrids.

Rule 31 — Assisted-Pipeline Disclosure. Extending Rule 16: the production pipeline (machine draft from the original → fidelity check against the original with the witness withheld from context → witness comparison → human editor of record) is published, and every published sentence carries a named human editor of record (Rule 18). The pipeline is an implementation detail; the accountability is not.

Disclosure of a limit (v1.2). Where a machine assists, the witnesses this standard requires are likely present in its training data. Our guarantee is procedural, not epistemic: the witness was not in the drafting context. We do not claim the machine had never encountered it, because we cannot, and a standard that claimed otherwise would be asking for exactly the trust this document exists to make unnecessary.

Rule 32 — Attribution Is Verified Separately (v1.2). In an anthology or florilegium — the Catena Aurea above all — the attribution of an excerpt is a factual claim about a named person, and it is verified as its own gated stage, before and apart from translating the excerpt. The asymmetry is the reason: a reader can weigh whether "nature" or "property" better renders proprietas, but cannot detect that a passage was never written by the Father whose name sits above it. Base editions get this wrong at rates that translation review does not catch — the first pilot's calibration found a wrong or misleading attribution or locus in half its units, one of which no witness caught.

Where the base edition's attribution or locus cannot be corroborated against an independent source, the excerpt is published with its attribution marked unverified, or it is not published. No unit reaches the human editor of record with an unverified attribution presented as fact.

Every translated unit (the LWM section, {§n}, is the unit of record; metadata rides as a sidecar per unit and renders as expandable "translation notes" in the reader) carries:

• Source text(s): Greek, Hebrew, Latin, or other witnesses consulted.
• Base edition + locus (v1.1): exact edition and column/section/DS number, with the edition's rights status.
• Tier label (v1.1): translated-from-original vs modernized-from-PD (Rule 30).
• Translation confidence: High / Medium / Low.
• Literary features: poetry, parallelism, rhetorical question, chiasm, alliteration, etc.
• Key terms: links to the controlled glossary (Rule 8).
• Alternative renderings: plausible translations considered and why they were rejected.
• Historical witnesses: significant traditional renderings (e.g., Vulgate, the PD English translations, KJV/DR where scripture is quoted) with divergence notes (v1.1, Rule 27).
• Pipeline record (v1.1): stages passed (draft, fidelity check, witness comparison) and checker outcomes (Rule 31).
• Editorial rationale: a concise explanation of the final choice.
• Editor of record (v1.1): the accountable human (Rules 16, 18).
• Revision history: every subsequent change with supporting evidence (Rule 17).