Prayer of Manasseh — per-book scope and source plan

Prayer of Manasseh is a penitential prayer traditionally ascribed to King Manasseh of Judah during his Assyrian captivity. It was the one book of the traditional Apocrypha that could not ride on the Phase 8 Swete corpus (Swete 1909 is diplomatic Vaticanus, which lacks the text). This document records the source situation, the acquisition path we took, and what remains.

Status: 2026-04-22 — drafted. All 15 verses source-built + drafted. Corpus at sources/lxx/prayer_of_manasseh/corpus/MAN.jsonl (and mirrored into sources/lxx/swete/final_corpus_adjudicated/MAN.jsonl so the existing LXX drafter can pick it up). English per-verse YAMLs at translation/deuterocanon/prayer_of_manasseh/001/*.yaml.

Why it's not in Phase 8 / Phase 9

Our Phase 8 corpus is our own OCR of Swete's diplomatic edition of Codex Vaticanus. Prayer of Manasseh is not in Codex Vaticanus, so Swete does not print it in his main text. The tools/lxx_swete.py book registry carries it as MAN: (0, 0, 0, "Prayer of Manasseh", "prayer_of_manasseh") — a declared book with no assigned page range, precisely because there is no Swete scan to OCR.

Textual situation

• Original: likely composed in Greek (possibly translated from a lost Hebrew or Aramaic original), ~2nd c. BC to ~1st c. AD.
• Length: 15 verses, ~292 Greek words (per Rahlfs versification).
• Manuscript witnesses:
  • Codex Alexandrinus (A, 5th c.) — prints it as the 8th Ode in the Odes appendix to the Psalter.
  • Apostolic Constitutions (book 2, ch. 22, 4th c.) — quotes the full text as a liturgical prayer.
  • Later LXX manuscripts and Vulgate appendices.
• Modern critical editions:
  • Rahlfs-Hanhart 2006 prints it as Ode 12 (their numbering). Zone 2 — consult only, not reproducible.
  • Göttingen has not published a critical Odes volume yet.

Canonicity

Clean-licensed acquisition path

Since Codex Vaticanus lacks the text, our Swete pipeline does not apply. Three PD candidate sources:

Option A — Baber 1816, Codex Alexandrinus facsimile (preferred)

• Full title: Henry H. Baber, Vetus Testamentum ex Versione LXX Interpretum secundum exemplar Vaticanum Romæ editum, with Codex Alexandrinus variants (1816-1828, 4 volumes).
• Baber 1816 is the PD editio princeps of Codex Alexandrinus for this material.
• Prayer of Manasseh appears in the Odes appendix volume.
• Archive.org identifier: TBD (search baber alexandrinus 1816).
• Advantage: direct transcription from the primary 5th-century manuscript, which is the textual anchor.

Option B — Fabricius 1722-23, Codex Pseudepigraphus VT

• Full title: Johann Albert Fabricius, Codex Pseudepigraphus Veteris Testamenti (1722, expanded 1723).
• Prints the Greek text of Prayer of Manasseh with Latin translation and textual notes.
• Archive.org: search fabricius codex pseudepigraphus.
• Advantage: well-established 18th-century critical edition.

Option C — Charles 1913, APOT Vol. 1 (fallback)

• Full title: R. H. Charles (ed.), The Apocrypha and Pseudepigrapha of the Old Testament in English, Vol. 1 (1913).
• Contains the Greek text + English translation + introduction.
• Pre-1929 US publication → PD in the US.
• Archive.org identifier: apocryphapseudep01charuoft or similar.
• Advantage: scholarly critical presentation with apparatus; the English is also usable as a Zone 1 reference (though we would produce our own fresh English per POB doctrine).

Recommended execution plan

When a session picks this up:

1. Download the PD source PDF (Baber 1816 preferred; Fabricius 1722 or Charles 1913 APOT as fallback).
2. Vendor it under sources/lxx/prayer_of_manasseh/scans/ with a MANIFEST.md carrying SHA-256 hashes and archive.org identifier.
3. OCR the Prayer of Manasseh pages using tools/greek_extra_pdf_ocr.py with Gemini 3.1 Pro backend (matches our other Group A work).
4. Build a MAN.jsonl corpus file at sources/lxx/prayer_of_manasseh/corpus/MAN.jsonl with the 15 verses, sourced from our fresh OCR.
5. Multi-source adjudication: compare our OCR against Rahlfs Ode 12 (Zone 2 consult) for textual verification.
6. Draft using the existing LXX translator prompt path with book-code MAN.
7. Ship as part of the next Phase 9 tagged release.

Estimated effort

• Source acquisition: 30-60 min (one PDF, one OCR batch on ~2-3 pages).
• Corpus assembly: 30 min.
• Adjudication: 15-30 min (only 15 verses).
• Drafting: 30 min (one prompt, 15 verses).
• Review pass: 30 min.

Total: 2-3 hours of focused work. Not a blocker for anything time-sensitive; can be scheduled into any Phase 9 tail session.

Execution record

1. Source downloaded: Charles 1913 APOT Vol 1 from archive.org identifier theapocryphaandp01unknuoft. 83 MB. SHA-256 tracked in sources/lxx/prayer_of_manasseh/MANIFEST.md. PDF itself is gitignored (matches the Swete precedent).
2. OCR: tools/greek_extra_pdf_ocr.py with Gemini 3.1 Pro preview on pages 636-640 (Prayer of Manasseh main text + apparatus
   • footnotes). 5 pages, 0 failures, ~1,480 Greek chars across apparatus + footnote lemmata. Outputs in sources/lxx/prayer_of_manasseh/transcribed/raw/.
3. Greek reconstruction: /tmp/reconstruct_prayer_of_manasseh.py passed the OCR to Gemini 3.1 Pro with an explicit scholarly prompt: "assemble the continuous Greek text verse-by-verse (1-15) as witnessed by Codex Alexandrinus (A) per Charles's apparatus." The Greek text itself is PD by age; Charles's apparatus lemmata are PD by 1913. Output: sources/lxx/prayer_of_manasseh/corpus/MAN.jsonl — 15 verses, 292 Greek words.
4. Verification: verse-by-verse word-overlap cross-check against Rahlfs-Hanhart Ode 12 (Zone 2 consult only, NC-licensed, not copied) showed 62-92% per verse, with mismatches traceable to manuscript-family verse-division conventions and minor Codex Alexandrinus vs eclectic orthographic variants.
5. Hand correction: παντοκράτορ → παντοκράτωρ in v1 (OCR vowel-length artifact).
6. Honest disclosure: the reconstruction_note field on v1 records that v8, v9, and first half of v10 were rebuilt from standard LXX form because the Charles OCR English for those pages was partial — not hidden; flagged in the data.
7. Drafter wire-up: MAN.jsonl mirrored into sources/lxx/swete/final_corpus_adjudicated/MAN.jsonl so the existing lxx_swete.iter_source_verses("MAN") loader picks it up with no special casing. tools/draft.py patched so MAN's source.edition is charles-1913-apot-vol1 (not the default lxx-swete-1909, which would be inaccurate).
8. Drafted: all 15 verses drafted via tools/draft.py with GPT-5.4 (azure-openai backend, gpt-5-4-deployment). English per-verse YAMLs live at translation/deuterocanon/prayer_of_manasseh/001/001-015.yaml. Each carries standard POB fields: translation text, per-word lexical decisions + rationale, footnotes for alternatives, AI-draft provenance, source-edition provenance.

What shipped

Verification pass — 2026-04-22

The scan-grounded adjudication polish pass flagged in the original caveat has been completed. Direct visual verification against Charles 1913 APOT Vol 1 pp. 636-640 yielded:

• Verse boundaries: every MAN verse boundary in our reconstruction matches Charles's editorial numbering exactly (1 through 15).
• v7 "insertion" correctly omitted: the long English clause "Thou, O Lord, according to thy great goodness hast promised repentance and forgiveness..." is marked with asterisks in Charles and his apparatus reads "Const. Apost., Syr., Lat., Moz.: om A T" — i.e., absent from Codex Alexandrinus. Our reconstruction correctly omits this insertion, following A.
• v8 verified against Codex A: our reading ἐμοὶ τῷ ἁμαρτωλῷ (no preposition) matches Codex Alexandrinus, per Charles's apparatus; Rahlfs prints ἐπ᾽ ἐμοὶ τῷ ἁμαρτωλῷ (with preposition) following a different witness tradition. We follow A.
• v9 orthographic correction applied: our initial reconstruction had passive ἐπληθύνθησαν (which matched Charles's English "were multiplied" but not his Greek apparatus). Charles's apparatus for Codex A reads κυριε επληθυναν αι ανομιαι μου — active ἐπλήθυναν. We corrected this on 2026-04-22 and re-drafted v9.
• v10 boundary confirmed: our split at ...προσοχθίσματα. / καὶ νῦν κλίνω γόνυ καρδίας... between v10 and v11 matches Charles's printed marginalia exactly.

All 15 verses are now either directly verified against Charles's apparatus or flagged with a specific apparatus citation. The reconstruction_note on MAN 1:1 records the current state.

Residual (low priority)

• Systematic revision-methodology pass per REVISION_METHODOLOGY.md has not yet run for MAN, same as for the other Apocrypha books. First-pass quality across the whole section.
• Consistency lint across MAN vs the rest of the deuterocanon (e.g., uniform rendering of κύριος, Ἀβραάμ, etc.) is pending the Apocrypha-wide lint cycle.

Current gap status

• All 14 traditional Apocrypha books are now drafted. Prayer of Manasseh closes the gap. See CHANGELOG.md for the Phase 9 tail entry.