# Review ledger

## Scope of evidence

Source snapshot: see SOURCE.md. This ledger distinguishes coverage, AI editorial
assessment, and human review. No human-reviewed status is claimed.

## 2026-09-29 - Generation and validation setup

- Source parser observed 66 books, 1,189 chapters, and 31,103 source verse records.
- The publisher uses VPL abbreviations such as MAR, JOH, and JAM. Source keys
  remain unchanged; they are not converted to another numbering system.
- A first Genesis 1 AI draft contained all 31 source references in order.
- Negative validation probes correctly rejected an omitted verse, an added
  duplicate verse, and an English fallback verse.
- The first editorial assessment found literal-English constructions and
  reproduction language rendered as literal fruit. STYLE.md records the
  corrections. A new sample generation uses that revised guide.
- Sample scope: Genesis 1, Psalm 23, Mark 1, Romans 8.
- Initial generation model: signed-in Antigravity, gemini-3.1-pro-high. No API key
  was loaded and no direct-model API fallback was used.

## Editorial issues requiring attention throughout the corpus

- English-to-Korean relay translation inherits WEB's textual decisions. It is
  not an independent Hebrew/Aramaic/Greek translation.
- WEBP's Romans 8:1 includes the flesh/Spirit clause; retain it because it is in
  the selected source. Do not silently harmonize to other editions.
- Preserve source Psalm superscriptions when present in a verse record.
- Keep God's Spirit, Holy Spirit, human spirit, and flesh distinct in context.
- Full sentence meaning, pronoun referents, quoted speech, negation, numbers,
  and conditions require editorial checking beyond mechanical completeness.
- A closed-loop AI review can share errors with the draft model. Human Korean
  editing and biblical-language review remain separate work.

## Release status

Historical generation status: this was an unpublished working draft. The website now distributes the same AI draft; this is not human approval.
Do not describe mechanical validation as theological or scholarly approval.

## 2026-09-29 - Four-chapter sample assessment

- Four chapters / 121 verses generated with exact source-key coverage.
- Main Codex AI editorial pass revised 52 verse records for readability,
  pronoun referents, dialogue, or fidelity. This is not human review.
- Restored the childbirth metaphor omitted from the initial Romans 8:22 draft.
- Removed archaic Psalm diction and distinguished direct address in 23:4-5.
- The broader corpus generation guide now incorporates these findings.
- Outstanding judgments are listed under each chapter's editorial questions.

## Follow-up AI review

- Genesis 3:3: removed an added claim that God feels afraid; kept the warned consequence of death.
- Genesis 3:16: corrected pregnancy to childbirth, matching the English source.
- Romans 8:21: restored the bondage metaphor alongside decay.
- These are Codex AI editorial corrections, not human scholarship review.
- Genesis 9:24 and 11:9: corrected literal English idioms about waking from drunkenness and scattering across the earth.

## Pronoun policy correction

The initial blanket ban on 당신 caused omitted objects in the Genesis 12 trial.
Style 1.1 allows natural spousal address while retaining source participants
explicitly. Divine address uses the source-appropriate name or title. This
changes an editorial preference, not source-reference or completeness checks.
The trial's reconstructed Hebrew and historical commentary are not verified
source material and will not be retained as scholarly notes.

## Genesis 12-17 checkpoint

- Genesis 12:11-13: restored explicit references to Sarai as the woman seen and spared and Abram as the man in danger.
- Removed unsupported original-language quotations and historical claims from trial notes.
- Genesis 15: all 21 verses received a Codex AI source comparison and editorial revision.
- Aligned Egypt/Egyptian, Euphrates, and Rephaim spellings in the current chapters and metadata.
- The signed-in Gemini 3.8 Flash Medium trial produced 43 verses in 35.6 seconds; mechanical checks passed. This is a throughput observation, not evidence of full semantic accuracy.

## Genesis 18-24 spot review

Source comparison of 32 selected verses identified omitted participants, an added named target in Genesis 21:9, and overly interpretive terms. Nine verses received Codex AI corrections. Genesis 18:19 remains a flagged theological wording question. This was a spot review, not a full semantic audit of the batch.

## Genesis 25-30 spot review

Thirty selected verses were compared with the source. Five received corrections: a malformed soul/blessing idiom, wishes turned into predictions, the 27:39 source wording, an added animal age, and an added watercourse. The full batch has not received a human or exhaustive semantic review.

## John 1 trial review

All 51 verses were compared with the supplied English. Nine received Codex AI edits, including restoration of the omitted 1:27 priority clause, removal of invented sibling age ordering, the takes-away wording in 1:29, and the Bethsaida spelling. The lower-reasoning trial did not establish a clear throughput advantage; the bulk model remains unchanged.

## Exodus source review

All 26 verses of Exodus 20 and selected verses from Exodus 3 and 21 were compared with the English. Seven received corrections, notably the people/generations distinction in 20:6 and premature birth in 21:22. Exodus 3:14 remains an explicitly flagged interpretive decision. No human review has been performed.

## Delivery record

The local bundle now contains paired English/Korean JSONL records as well as readable Markdown and HTML. Every verse records its source hash, reference, initial model selection, generation receipt, and AI draft status. The first fourteen chapters' legacy model metadata was filled from the original signed-in Pro run. Bundle integrity was checked with the ZIP reader. Automated and manual refreshes share a lock, and published local artifact files are replaced atomically.

At that delivery checkpoint, the only entity-variant candidate was Sinai: 시내 versus context forms 시내산 and 시내 광야. These are a shared spelling with mountain/region qualifiers, not conflicting transliterations.

## Repeated-text screen

At the 4,081-verse checkpoint, five groups of identical Korean text had differently worded English sources. Codex AI inspected the distinct English variants. They were equivalent speech introductions, sacrificial instructions, or repeated offerings, so no text was changed. New groups remain review candidates; this screen is not a full semantic certification.

## Psalms 1-5 direct drafting

Codex drafted these 46 verses directly from the pinned English and compared each verse during drafting. Source superscriptions and Selah markers remain in their source verse records. Psalm 4:1 is flagged as an interpretive decision. This is a same-author AI self-review, not an independent or human review. The receipt records the direct Codex origin rather than assigning these chapters to Antigravity.

## Numbers and Deuteronomy source review

Twenty-nine selected verses in Numbers 27 and Deuteronomy 6, 22, and 30 received a Codex AI source comparison. Three were corrected: an added portion-of-authority qualification in Numbers 27:20, the soul/intention distinction in Deuteronomy 6:5, and an added stern-warning statement in Deuteronomy 30:19. The capitalized Spirit in Numbers 27:18 remains an explicit terminology-review question. Moses' quoted address changes register across draft batches and needs a later consistency pass. No human review has been performed.

## Joshua 10 AI source comparison

All 43 verses received a Codex AI source comparison. Four substantive wording edits preserve the source participant group, fix a malformed army description, improve sunset/revenge idioms, and retain the English South wording. Divine-name honorifics were aligned with the entity guide in this chapter. The resolved actors in 10:10 and 10:30 remain explicitly flagged for expert review; this pass does not certify them.

## Jonah direct drafting

Codex drafted all four chapters (48 verses) from the pinned English and compared each verse during drafting. The source's huge fish, vine, wind description, three-day crossing, forty-day warning, and final question were retained without adding alternative-edition details. Jonah 2:8 remains an explicitly flagged interpretation of the English mercy phrase. This is same-author AI self-review, not an independent or human review. The receipt records Codex as the initial drafter.

Jonah 4:2 received a subsequent wording correction: the English does not explicitly say the earlier statement was addressed to God. The separate editorial ledger preserves that change after the initial receipt.

## Contemporary-register sweep

A corpus-wide ending screen identified older forms concentrated in poetry, prayer, blessings, and quoted speech. Codex modernized endings in 80 verses, preserving tense and wish/command distinctions; 1 Samuel 18:7 also replaces opaque old number compounds with readable thousands/tens of thousands. This was a register pass, not a full semantic comparison of every affected verse. The ledger retains all English sources and before/after text. Three candidates were valid modern copulas or a proper name (아나이다, 바로다, 하나이다) and were excluded. The Hebron metadata typo in Judges 16 was corrected; its verse text was already correct. Future chapter guidance explicitly applies contemporary endings to poetry and prayer as well as narration.

## Entity metadata follow-up

Five metadata entries were aligned with the actual verse text: Ephraim in 2 Samuel 2, South in Genesis 12-13 and Joshua 10, and Tabor in Judges 8. The previous metadata could feed conflicting or overly specific spellings into later generation even where the verse text was already correct. Other unmatched metadata phrases often reflect grammatical suffixes or collective nouns; absence of an exact substring alone is not treated as a translation error.

The offline validator was also tested against a fresh extracted snapshot of 8,463 verses. It passed the valid partial snapshot, rejected the incomplete corpus under --require-complete, and detected a missing JSON verse plus reordered and duplicated Markdown verses in temporary copies.

## Philemon direct drafting

Codex drafted all 25 verses from the pinned English, comparing each while drafting. The personal appeal, debt language, and distinction between a slave and a beloved brother are retained. The paired love/faith wording, fellowship/knowledge phrase, flesh/Lord phrase, and singular/plural address choices remain flagged where English alone does not settle interpretation. This is same-author AI self-review, not expert review; the receipt records the direct Codex origin.

Philemon 1:20 received a follow-up readability correction to avoid an ambiguous pronoun/particle combination. The separate editorial ledger preserves the change.

## Editorial-note evidence cleanup

A keyword screen found unsupported original-language, variant, etymology, and literal-accuracy assertions in generated notes. 56 notes were removed and preserved as removed claims in a separate ledger. Existing questions that explicitly request expert review were retained. Nine selected English verses were compared; four edits remove an added silver denomination, retain incomprehensible in Judges 13:18, restore the inner-sanctuary image in 2 Samuel 16:23, and modernize the collective numeric phrase in Numbers 10:36. Numbers 24:17 remains a flagged interpretation rather than a verified Hebrew claim. The note policy now requires source-based questions and discourages unsupported background. This cleanup is an AI review, not scholarly verification.

Two further unsupported background notes in 1 Kings 14-15 were removed after inspection. Their verse text was unchanged. The audit now lists new source-language and literal-accuracy note candidates for review.

## Generation-receipt verification

The audit now opens each chapter's named generation receipt and checks that it
contains exactly one matching chapter with the same ordered verse references.
The direct Psalms receipt wraps each draft together with its English source;
that legacy layout initially failed the new checker and stopped generation.
The checker was corrected to handle the actual wrapped format, the complete
10,646-verse snapshot passed, and generation resumed from missing chapters.
The saved translation records were not regenerated or lost. The extracted-bundle
validation tests also passed with the stronger receipt check. This verifies
provenance links and record coverage, not semantic accuracy.

## Reader and quotation checks

The offline HTML reader was opened in the in-app browser at the 355-chapter checkpoint. Korean text and draft disclosure rendered clearly, and the Philemon chapter link reached the correct heading. The temporary localhost server and preview tab were closed after checking.

A source comparison corrected a missing closing quote in Genesis 3:3, a premature closing quote in Genesis 7:1, a stray narrator quote in Leviticus 10:20, and the speech/narration boundary in Leviticus 16:34. The latter also restores the English passive action. Nested punctuation in 2 Kings 22:15-20 was reduced to two clear levels while retaining attribution wording; two unprinted silver specifications were changed back to generic money. Whole-chapter quote parity is only a screening signal because the source repeats opening marks across paragraphs and speeches can cross chapters.

## Warning and prayer source corrections

Genesis 3:3 now states the death warning without an ambiguous then connector after the prohibition. In 1 Chronicles 4:10, cause pain was incorrectly rendered as experience pain despite an AI note claiming the active sense was retained. Codex corrected the verse against the supplied English and replaced the unsupported language claim with an explicit review question. These two corrections are recorded with source and before/after text.

## Historical-book name review

A targeted comparison against the pinned English corrected Gibeon in 1 Kings 3:4-5, Ahimelech in 1 Chronicles 24:3, 6, 31, and Mattaniah in 2 Kings 24:17. The spelling of Mattaniah in 1 Chronicles 9:15 was aligned to 맛다냐 without merging the distinct individuals. The Jehoahaz entry in the 2 Kings 13 entity table was corrected; its verse text already distinguished Jehoahaz and Jehoash. These seven verse edits and associated metadata changes are preserved in editorial-names-03.json in the bundle. This was an AI proper-name check, not a human review or a complete semantic review of these chapters.

## Ezra 1 reference recovery and wording review

The first rejected batch covering 2 Chronicles 30 through Ezra 1 had all expected verse entries but mislabeled Ezra 1:8 as 2CH 1:8. Its Korean text was compared with Ezra 1:8 and matches the supplied content. The original receipt is retained unchanged; reference-recovery-ezra1.json records the single reference correction and points to the original. All chapter and verse keys were then validated again before acceptance. A second rejected attempt is also preserved, not used for the accepted draft.

All eleven verses of Ezra 1 were compared against the English by Codex AI. Four wording changes clarify the location of the temple, avoid equating left with survived, express the strengthened-hands idiom naturally, and retain second kind without asserting second grade. The source item counts and stated total remain as supplied. This is AI review only; expert review remains outstanding.

## Repeated-target review at 12,288 verses

The 17 identical-Korean/different-English groups were compared. Most differences were punctuation, word order, collective nouns, or equivalent formulaic narration. Three edits retain differences visible in the supplied English: overtook in 1 Samuel 31:2 now means caught up with; finest gold in 1 Kings 10:18 no longer becomes pure gold; and 2 Chronicles 18:22 no longer adds all. This targeted AI screen cannot detect incorrect translations that have unique Korean wording. The full compared groups and edits are retained in repeated-target-review-02.json.

The age discrepancy note for 2 Chronicles 36:9 was restricted to what the pinned English actually shows: eight there, eighteen in 2 Kings 24:8. An unverified claim about Hebrew textual tradition was removed, and the discrepancy remains an expert-review question.

After a cross-book reference-label error, subsequent generation batches stay within one book. Reference validation now identifies exact missing, extra, or duplicated keys to make retries actionable. The original malformed Ezra receipt was used to verify that invalid keys still fail and the explicitly corrected key passes; no validation requirement was relaxed.

## Additional historical name distinctions

A targeted check aligned Heman in Genesis 36:22, Riblah in Numbers 34:11, and Jerimoth in 1 Chronicles 24:30 and 25:4 with consistent Korean spellings. These spelling choices do not assert that different people or places are identical. Jehoram in 2 Chronicles 21-22 is now 여호람, while the explicitly supplied form Joram remains 요람.

The supplied Azariah in 2 Chronicles 22:6 had been silently harmonized to Ahaziah. It is restored as 아사랴 and flagged for expert review. Nearby wording edits remove an added immediately in verse 7 and retain slain in verse 11 instead of a future prediction. This was a targeted AI source comparison, not a complete review of these chapters. Changes are recorded in editorial-names-04.json.

A follow-up caught and reversed an overbroad name edit in 1 Chronicles 25:22: the supplied name there is Jeremoth, so 여레못 is retained separately from Jerimoth / 여리못 in verse 4. The correction remains visible in editorial-name-followup.json. Ezra 7:12 also received a modern letter-heading ending in place of 이르노라.

## South-region wording consistency

The remaining 17 verses containing 네게브 were compared with the supplied English, which says South or south. They now use plain geographical wording consistent with the earlier Genesis and Joshua edits. In 2 Samuel 24:7 an added geographical name was removed. This is a relay-source wording choice, not a finding that the Negev identification is geographically wrong. The exact source, before, and after strings are preserved in editorial-south-region.json.

## Nehemiah 8 source comparison

All 18 verses were compared with the pinned English by Codex AI. Four verses were edited: verse 9 follows the divine-name register; verse 10 preserves the unspecified speaker instead of asserting Ezra, avoids assuming inability to prepare food, and retains the joy-of-the-LORD relationship as an explicit review question; verse 14 has more natural Korean syntax; and verse 17 no longer adds an explicit causal link between the long lapse and the rejoicing. The record is editorial-nehemiah8.json. This remains an AI review, with expert interpretation and language review outstanding.

## Esther dialogue and scepter wording

All 17 verses of Esther 4 were compared with the English by Codex AI; six were edited for natural dialogue and closer source correspondence. The edits remove an unprinted silver specification, reduce repeated 당신, clarify the uninvited-audience rule, and preserve the messengers in verse 15. The other golden-scepter verses, Esther 5:2 and 8:4, were also compared: 금 지팡이 replaces the obscure 금 규 consistently, and favor in 5:2 no longer becomes an asserted romantic feeling. Exact changes are in editorial-esther4.json. This was not a complete semantic review of Esther.

## Job 1-3 source comparison

All 61 verses of Job 1-3 were compared with the pinned English by Codex AI. Renounce is now expressed as 저버리다 rather than silently changed to curse. Job 1:22 retains the specific denial of accusing God of wrongdoing instead of a broad claim that Job did not complain. Other changes preserve speech rather than thought in 1:5, use a readable paired-oxen count, avoid narrowing household to servants, retain the causal agency in 2:3, and avoid adding a malignant diagnosis to painful sores.

Job 3:3-9 now presents the curses as present wishes rather than only past counterfactual regrets, restores rejoice in verse 6, retains personification, and avoids specifying dawn for twilight. An added palace in verse 14 was removed. The difficult untimely-birth image in verse 16 avoids implying an induced termination and remains an explicit expert-review question. The flagged archaic ending in Job 4:21 was modernized. Exact changes are recorded in editorial-job1-3.json. No original-language or manuscript evidence was examined, and human review remains outstanding.

## Selected Job dialogue review

Eighteen further verses were compared: Job 13:15-16, 14:7-15, and 19:23-29. Twelve edits retain the declarative and no-hope wording in 13:15, restore the water-scent image, remove an added conditional and manner of waiting in 14:14, and modernize the register of 19:23-29. The skin image no longer adds rotting. The omitted on-my-side relationship in 19:27 is restored, with its interpretation and the stranger phrase explicitly flagged for expert review. The threat in 19:28 is kept as an exclamation rather than a planning question. Exact edits are in editorial-job-selected.json. Other dialogue still needs a full human register review; these targeted changes do not establish that all wording is modern or semantically settled.

A targeted geography check compared Nehemiah 2:7, 2:9, and 3:7. The supplied River designation is retained without adding a river name. Nehemiah 3:7 now says the governors residence was repaired, following the supplied English, rather than changing the object to a section extending to it. The ledger is editorial-nehemiah-river.json. This wording choice does not dispute the conventional geographical identification.

## Job imagery spot review

Sixteen verses were compared: Job 9:9 and 38:1-4, 7-8, 16-20, 31-33, 36. Five were edited. The astronomical name Pleiades is now 플레이아데스 rather than the obscure 묘성. Bear and her cubs are retained as 곰자리와 그 새끼들, without substituting seven stars and daughters. Their exact identification is an expert-review question, not a researched claim. Job 38:3 no longer inserts a belt into brace yourself, and 38:16 retains sea springs instead of an abstract origin. The rooms-of-the-south image also remains explicit and flagged. Changes are recorded in editorial-job-imagery.json.

## Job 42 closing chapter

All 17 verses of Job 42 were compared with the English by Codex AI. Seven changes keep uttered free of an added disparaging verb, express acceptance without asserting a feeling, make the doubling unambiguous, remove an unprinted silver specification, render oxen as pairs, and express the four generations clearly. The supplied self-abhor and repentance wording remains; no alternative original-language interpretation is asserted. Changes are in editorial-job42.json.

## Psalms 8 and 22 source comparison

All 40 verses of Psalms 8 and 22 were compared with the pinned English by Codex AI. Nineteen verses were edited. The musical heading retains instrument of Gath; angels in Psalm 8:5 stays as supplied. Psalm 22 no longer turns disappointed into public shame, adds a lip gesture, or imports an outer/inner-garment distinction. Its mocking challenge remains a challenge rather than a confident prediction. Help is retained rather than changed to strength, and prayer and exhortation use a more consistent contemporary register while preserving wishes and future statements. Exact edits are in editorial-psalms8-22.json. This is AI review only.

## Psalm superscription review through Psalm 70

The headings through Psalm 70 were compared with the pinned English. Edits retain the direction to the chief musician, distinguish eight strings from an eighth string, restore The Death of the Son instead of reconstructing an absent Hebrew label, distinguish contemplation from a teaching poem, retain wedding in Psalm 45, and avoid treating the Jeduthun dedication as an asserted tune name. Tune-title imperatives use contemporary grammar. This review covers the headings only, not all verse bodies or entire chapters. The ledger is editorial-psalm-headings.json, and the style guide now carries these concrete distinctions for later generation.

## Psalm 51 prayer review

All 19 verses were compared with the pinned English and edited into consistent modern respectful prayer. Archaic endings and address forms were removed. Mercy is retained as mercy; transgressors are not narrowed to people convicted of crimes; and the willing-spirit phrase remains distinct from Holy Spirit, with its referent flagged for expert review. The supplied conception and sin language is retained without adding a doctrinal explanation. The ledger is editorial-psalm51.json.

## Checkpoint/write coordination

The generation run stopped after Psalms 71-80 were exported because an audit overlapped a heading edit and saw a temporary Markdown/JSON mismatch. A fresh audit confirmed the saved manuscript was consistent. Chapter export now shares the checkpoint lock, so an audit cannot read between the two writes. A temporary-copy contention test confirmed that the writer blocks while the checkpoint lock is held, creates no partial chapter files, and produces matching Markdown/JSON after release. The test and result are included as test-export-lock.py and export-lock-test.json. Saved chapters were retained, not regenerated.

The advisory register screen was expanded after Psalm review exposed old endings outside its original pattern list. Flags remain review candidates, not semantic errors or proof of overall naturalness.

## Expanded register review

All 41 candidates from the expanded register screen were compared with the pinned English. Forty were edited; 버리니 in Job 27:21 is a modern connective and was intentionally retained. Wishes, requests, assertions, and predictions were kept distinct. The source check also restored speak in Judges 5:10, insulted the builders in Nehemiah 4:5, disappointed in Psalm 25:20, and the seed-in-many-waters image in Numbers 24:7. Genesis 24:12 no longer adds a particular meeting to the request for success, and Judges 16:20 retains said rather than asserting unspoken thought. Divine address in 1 Samuel 23:11 no longer uses 당신. The full candidate review and exact changes are in editorial-modern-endings-02.json. This does not establish that all remaining dialogue is stylistically modern or semantically reviewed.

## Psalms 90-91 source comparison

All 33 verses were compared with the pinned English by Codex AI. Nine edits explain a night watch without inventing a fixed hour count, retain the sprouting action, use readable years and counting language, and keep the dwelling unspecified rather than adding a tent. Cobra is retained rather than generalized to a venomous snake. The request to relent is expressed in the Psalm's local wrath context, and the difficult relation in 90:11 remains flagged. No modern application or devotional interpretation was added to the verses. Exact changes are in editorial-psalms90-91.json.

## Later Psalm checks

Psalm 110:1, 4, 6, and 7 were compared with the English. Verse 1 now retains the present says; unverified messianic and divine identity assertions were removed from the entity metadata and replaced with a review question. Psalm 116:4 and 118:25 received modern prayer endings. The register screen also flags 알리라 in Psalm 96:2, which is a valid modern imperative and remains unchanged, alongside the previously reviewed modern connective 버리니. Two new repeated-target groups involving Psalms reflect minor source phrasing differences rather than omitted content. These checks and exact changes are recorded in editorial-psalm-followup.json.

## Psalm 119 labels and spot comparison

Forty-three unique verses were compared: the starts of all 22 eight-verse blocks and the 22 supplied heading-label locations, with verse 1 shared. All 22 alphabetic labels are retained. This VPL places most labels at the end of the preceding verse; bracketed Korean labels retain those source positions and distinguish headings from prayer sentences. No missing headings were invented or moved to different verse records. Eight wording fixes retain statutes, disappointment, reaching out hands, present action, and a stated causal link, and remove an added only. The ledger is editorial-psalm119.json. The other verses of this long Psalm have passed mechanical checks but have not received a full semantic comparison in this pass.

The remaining Chief Musician headings through Psalm 149 were checked. Psalms 76, 77, and 80 received the same source-based direction wording; Psalm 77 also retains the Jeduthun dedication without asserting a tune. Its first verse now states a request for God to listen rather than reporting that he already listened. The ledger is editorial-psalm-headings-02.json.

## Proverbs 3 and Wisdom passage

All 35 verses of Proverbs 3 and twelve selected verses of Proverbs 8 (1, 12, 22-31) were compared with the pinned English. Twelve edits improve ordinary vocabulary and retain kindness, finding rather than merely seeking wisdom, weariness, the repeated mocker/mocking language, and the irony of shame as promotion. Proverbs 8:12 now retains the supplied imperative rather than turning it into Wisdom's statement. The possessed/born/craftsman language remains as supplied without adding a doctrinal identity. Ambiguous good-understanding wording is flagged for expert review. The ledger is editorial-proverbs3-8.json; this is not a complete semantic review of Proverbs 8.

## Selected Proverbs 10-16 comparisons

Twenty-one verses were compared: Proverbs 10:4, 12, 19, 22; 11:1, 24, 25; 12:10, 18, 22; 13:24; 14:12; 15:1, 3, 17, 22; and 16:3, 9, 18, 24, 32. Five edits retain disobedience instead of a generic fault, preserve the watering metaphor, use readable mercy wording without adding even to animal care, retain careful attention in discipline rather than frequency, and repair the eye/watch sentence grammar. No modern application was added. 탄탄대로다 in Proverbs 15:19 is another reviewed false-positive register flag: a noun followed by an ordinary copula. The ledger is editorial-proverbs10-16.json.

## Later Proverbs, Ecclesiastes, and Song spot comparisons

Sixty-eight selected verses were compared with the pinned English: 25 verses in Proverbs 18-24, six in Ecclesiastes 1, Ecclesiastes 3:1-15, all fourteen verses of Ecclesiastes 12, and eight verses in Song of Songs 2 and 8. Twenty verses were edited. Proverbs retains beer and personification, avoids an added only and speaker fear, and clarifies pleading a case and being led to slaughter. Ecclesiastes retains eternity without adding longing, preserves poetic figures without anatomical explanations, restores almond instead of apricot and a goad instead of a whip, and keeps much study distinct from excessive study. Song retains the supplied browsing image without an added flock and the seal placement without adding an engraving action. Difficult poetic expressions remain expert-review questions. Exact references and changes are in editorial-wisdom-selected.json. These are selective AI comparisons, not a full semantic review of the three books.

## Early Isaiah spot comparisons

Thirty-nine selected verses, including all of Isaiah 6, were compared with the pinned English. Twelve edits retain oppression rather than bloodshed, a statement of holiness rather than direct address, present failure to understand rather than future prediction, and terebinth rather than chestnut. Isaiah 9 uses the agreed narrative register and retains the supplied titles without asserting a doctrinal identification. Farm implements, calf, cobra, and putting a hand on rather than into a den follow the English distinctions. Exact scope and changes are in editorial-isaiah-early.json. No complete semantic review of Isaiah is claimed.

Six new register candidates through Isaiah 25 were compared with the source. Four received modern endings; the same check retained has blessed rather than a future blessing, declared rather than merely appeared, and the repeated destroyer/destroys wording. The modern imperatives 알리라 in Isaiah 12:5 and 돌리라 in Isaiah 22:4 remain unchanged. The ledger is editorial-isaiah-register.json.

## Middle Isaiah and vocabulary checks

Twenty-one selected verses in Isaiah 26-35, including all ten verses of Isaiah 35, were compared with the English. Six edits correct the voice being behind the listener rather than behind the ear, avoid an added bloodshed plot, clarify the desert/rose sentence, and preserve the path statements without adding exclusive only. References and changes are in editorial-isaiah-middle.json. A separate source-conditioned screen corrects almond/apricot and jackal/wolf-or-dhole vocabulary in saved chapters, recorded in editorial-botany-animals.json. This vocabulary pass is not a full review of the affected chapters or a claim about ancient species identification. The entity guide now records these distinctions for subsequent generation.

## Isaiah 38-46 spot comparisons

Thirty-three verses were compared with the English, including eighteen verses of Isaiah 40 and two register candidates. Eleven edits retain warfare rather than prison service, a level road rather than merely a straight road, gentle leading rather than a tame disposition, waiting rather than an added exclusive only, and a word not being revoked rather than not returning. The new thing in Isaiah 43:19 is springing out now, not only about to appear. Prayer and divine speech have modern endings. The exact limited scope and changes are in editorial-isaiah40-selected.json.

## Isaiah 52-58 source comparisons

Thirty-eight verses were compared, including all of Isaiah 53 and 55. Seventeen edits restore several substantive source distinctions: Isaiah 53:10 has you make his soul an offering, and 53:11 explicitly includes seeing light after suffering. Wounds no longer introduces an unprinted whip, suffering remains broader than sadness, and the printed present descriptions remain present. Isaiah 52 retains the you/his shift rather than silently harmonizing it. Later changes preserve cypress, humble rather than afflict, cast-out people rather than merely wandering people, and plural them in the healing promise. The ledger is editorial-isaiah53-55.json. Interpretive questions remain flagged, and these checks do not establish whole-book accuracy.

## Closing Isaiah spot comparisons

Twenty-five verses were compared, including all eleven verses of Isaiah 61, Isaiah 65:17-25, and five verses of Isaiah 66. Four edits clarify the sent speaker as the one assigned to bind broken hearts, retain the present create wording, avoid narrowing feed to grass, and remove an unprinted forever in Isaiah 66:22. The age language in Isaiah 65:20 remains an expert-review question. The ledger is editorial-isaiah-closing.json.

## Early Jeremiah spot comparisons

Thirty-one verses were compared, including all nineteen verses of Jeremiah 1. Eight edits clarify the priest apposition, retain the fifth month without implying Gregorian May, repair awkward speech-subject grammar, distinguish watching from merely keeping a word, retain the cauldron direction, and preserve families and bronze. Exact scope and changes are in editorial-jeremiah-early.json. Remaining ambiguity is recorded as a question, without an added symbolic explanation.

## Jeremiah 9-17 selected comparisons

Twenty-seven verses were compared, including the register candidate in Jeremiah 4:19. Eight edits retain might without narrowing it to military valor, remove added exclusive only, use the agreed narration ending, distinguish fear from mere surprise and abhorrence from contempt, repair prayer grammar, and retain confidence rather than hope. The modern connective 두근거리니 is a false-positive register flag and remains unchanged. The exact scope and changes are in editorial-jeremiah9-17.json.

## Delivery provenance and export validation

The independent delivery check found 137 legacy structured chapters without an explicit provider field. Each has an original successful signed-in Antigravity receipt and a recorded Gemini model; the exporter already labeled these with that provider. The field is now explicit, with no text, source, model, or original-receipt change. The ledger is provider-normalization.json. The new offline delivery verifier compares every exported Markdown, HTML, and JSONL verse against the canonical structured snapshot, including order, English text, provenance, review labels, chapter navigation, and offline dependencies.

The new delivery validator passed on an extracted 782-chapter, 19,876-verse snapshot. In a temporary copy it rejected changed JSONL text, a missing JSONL verse, a broken chapter-navigation link, and changed Markdown text. The full-delivery gate correctly rejected that partial snapshot. The test result is delivery-validation-test.json. Every new checkpoint now runs the export comparison after the source audit and export.

## Jeremiah 18-29 spot comparisons

Thirty-seven verses were compared with the source. Twelve changes retain mention rather than memory, persuaded rather than an explicit deception claim, disappointment alongside dishonor, the complete divine-name expression, and seventy years for Babylon rather than in Babylon. Speech grammar and modern endings were repaired. Garden wording is kept broad rather than restricted to an orchard, with its scope noted for review. Jeremiah 29 remains in its stated letter context. Exact changes are in editorial-jeremiah18-29.json.

## Comma-boundary register review

An old ending before a comma in Jeremiah exposed a gap in the advisory screen. The comma boundary now produces review candidates too. All 52 newly found candidates were compared with the source: 39 verses received modern endings or wording, while 13 valid modern connectives remain unchanged. Selected wording corrections retain counsel, disappointment, residence rather than an added sojourning claim, and destitution rather than only abandonment. Exact changes and retained candidates are recorded in editorial-comma-register.json. This screen is still limited and does not certify complete stylistic or semantic accuracy.

## Jeremiah 30-48 selected comparisons

Twenty-nine verses were compared, including the new-covenant passage and four new register flags. Seven changes retain voice without requiring song, the complete name in Jeremiah 33:16, a small cutting knife, the promise of living, and pouring off rather than pouring onto something. The new-covenant clauses were retained as supplied without inserting a doctrinal interpretation. The ledger is editorial-jeremiah31-48.json.

## Closing Jeremiah spot comparisons

Twenty-four verses were compared, including the deportation counts and ending. Eight edits modernize three register candidates, repair a possessive divine title, retain weariness without adding a fall, and keep the allowance broader than food alone. The prison-clothes subject follows the preceding king and remains noted for expert review. The printed deportation counts total 4,600 and were retained. Exact references and edits are in editorial-jeremiah-closing.json.

## Lamentations selected comparison

Thirty verses were compared: Lamentations 1:1-5, 3:19-39, and 5:19-22. Ten edits preserve the city personification without changing great to first-ranked, tears without an added never-drying claim, misery rather than wandering, and remaining rather than ruling. Approval is distinct from pleasure, and the ending retains a present assertion of anger rather than replacing it with a question. Mercy vocabulary is more accessible in the selected lines. Exact changes are in editorial-lamentations.json; the remaining verses have not received a full semantic comparison.

## Early Ezekiel spot comparisons

Thirty-six verses were compared, including all of Ezekiel 2. Eleven edits clarify the fifth year rather than an anniversary day, retain lightning and beryl, remove an unprinted front orientation and outside-fire detail, and preserve the paired negations in the watchman warning. Blood accountability is expressed in ordinary Korean. Abominations remain broader than idols, and the stone/flesh heart image is retained without an added softness explanation. The ledger is editorial-ezekiel-early.json; spatial imagery remains an expert-review question.

Two early Ezekiel register candidates were compared and modernized. Ezekiel 3:12 retains a blessing formulation, and 12:23 says the proverb will no longer be used rather than asserting inability to use it. Exact changes are in editorial-ezekiel-register-01.json.

## Ezekiel 16-18 selected comparison

Twenty-eight verses were compared. Nine edits retain completed tree actions, non-use rather than inability to use a proverb, wronging without narrowing it to oppression, and increase without limiting it to an explicitly unfair gain. The mountain-meal and impurity clauses avoid inserted explanations. Ruin no longer introduces an unprinted stumbling block, and the final death language avoids adding a predetermined class of people. Both turning scenarios remain as supplied. The ledger is editorial-ezekiel18.json; this is not a complete review of Ezekiel 18.

## Ezekiel 21-28 source comparisons

Twenty-eight verses were compared, including Ezekiel 28:1-19. Eleven edits retain condemning rather than merely despising, the city returning rather than an opened gate, and all nine supplied gem names instead of a different traditional list. Speech grammar and two old endings were repaired. The seal and cherub images remain without a doctrinal identification. An unverified personal identification of the ruler of Tyre was removed from entity metadata, and Daniel is not assigned a verified cross-book identity. The flagged rod note was about English wording, not an ancient-language claim; it is now an explicit expert-review question. Exact changes are in editorial-ezekiel21-28.json.

All 21 new register candidates in Ezekiel 29-37 were compared and modernized without turning predictions into commands. The same targeted pass keeps dispersal rather than expulsion and putting breath rather than requiring a blowing action. The ledger is editorial-ezekiel-register-02.json. The prediction-register rule is now prominent at the beginning of the generation style guide.

## Shepherd and dry-bones passages

Thirty-three verses were compared across Ezekiel 34, 36, and 37, including the complete fourteen-verse dry-bones scene. Eight edits retain diseased and sick, the repeated seeking action, the flesh-heart image, and earthquake rather than generic vibration. Four winds is expressed clearly as 사방의 바람 rather than an ambiguous 네 바람. Wind, breath, and Spirit follow the supplied English distinctions. The ledger is editorial-ezekiel34-37.json.

## Closing Ezekiel comparisons

Eighteen verses were compared, including the river scene, measure statements, final city name, and one register candidate. Four edits retain dispossession, restore the difficult sea-flow clause in 47:8, distinguish spreading nets, and describe food-producing trees without suggesting their wood is edible. The difficult flow and measure clauses are flagged for expert review. The modern noun-plus-copula 자리니 is retained. Exact scope and changes are in editorial-ezekiel-closing.json.

## Early Daniel selected comparison

Thirty-two verses were compared. Nine edits retain increased flesh rather than a claim about skin quality, modernize a blessing, keep the furnace conditions open, remove an unprinted upper-room detail, and preserve the multiplicative court figures. The son-of-man figure comes with clouds without an added riding action, and unnamed attendants are not declared human. Two 내리니 register flags are ordinary modern connectives and remain unchanged. Exact scope and changes are in editorial-daniel-early.json.

## Closing Daniel comparisons

Twenty-four verses were compared, including all of Daniel 12. Twelve edits modernize prayer, retain the 2,300 evenings-and-mornings expression without an added day total, and render weeks with the ordinary Korean unit 주. The 70, 7, 62, one-week, midpoint, 1,290-day, and 1,335-day distinctions are retained without a calendar scheme. The most-holy referent is left unresolved rather than specified as a place; plural times is not silently fixed at two. Closing, running, and standing imagery remains as supplied. The ledger is editorial-daniel-closing.json, with interpretive questions for expert review.

## Hosea and Joel source comparisons

Forty-two verses were compared. Twelve edits retain betrothal, Your judgments, unnamed callers, holding arms, lifting the yoke, and the source plural-to-singular shift. The prayer in Hosea 14 brings words rather than an inserted scripture quotation and retains the explicitly supplied bulls. Older vocabulary is clarified without changing the sacrifice or mercy distinctions. Joel 2 retains daughters, sons, male and female servants, and the source Spirit wording. Joel 1:7 and 1:20 register flags are ordinary modern connectives. Interpretive questions remain separate. See editorial-hosea-joel.json.

## Amos through Habakkuk comparisons

Sixty-nine verses were compared, including all of Obadiah. Thirteen edits preserve the mighty-stream and overtaking images, distinguish understanding from understanding people, keep ancient times without an added creation claim, and retain pursuit rather than expulsion. Micah 6:8 now renders mercy directly; prayer and lament use modern endings. The final musical direction in Habakkuk remains present. Three flagged connective endings were retained as modern grammar. See editorial-minor-prophets-middle.json; ambiguous images remain expert-review questions.

## Zephaniah and Haggai comparisons

Forty-three verses were compared, including all of Haggai 2. Thirteen edits remove an added inability to contain joy, retain the source calming and disappointed wording, clarify the high-priest apposition, and keep the unspecified measures without a Korean-unit conversion. The divine speech and narration are made more consistent while the temporal wording is recorded for expert review. Dates remain numbered ancient months, without Gregorian conversion. See editorial-zephaniah-haggai.json.

## Zechariah and Malachi comparisons

Seventy-two verses were compared, including all of Zechariah 3 and 4 and Malachi 4. Sixteen edits retain the wish of rebuke, turban, capstone, plant image, unidentified river, pronoun shift, holiness rather than a specified place, and unspecified residue of the Spirit. Malachi 3 keeps offerings made in righteousness rather than declaring the offerings themselves righteous. The supplied divorce wording remains intact. Four entity-metadata fields containing Latin text were corrected to their already established Korean spellings; those metadata fixes do not change scripture. See editorial-zechariah-malachi.json.

## Early Matthew and expanded prayer screen

Fifty-five verses in Matthew 1 and 5-7 were compared, with eleven source and readability edits. The Lord's Prayer uses modern respectful endings, keeps debts/debtors, and includes the doxology present in the pinned source. Matthew 5:18 retains both until clauses and 7:23 retains never knew. The expanded prayer-ending scan found seven previously unflagged verses; two overlap Matthew and five older verses were compared and corrected. Nehemiah 5:19 no longer adds an explicit request for favor absent from the source. A Malachi note accurately describing the English had been flagged by the broad literal-translation heuristic; it is now phrased as a reader-review question. See editorial-matthew-early.json.

## Middle Matthew comparisons

Seventy verses in Matthew 9-20 were compared. Seven edits keep the call to repentance, the person clause within its source verse in the sower explanation, the deceitfulness of riches, and the explicit I in Matthew 16:13. The legal question in 19:3 remains a question about law, and ransom is expressed with the ordinary Korean body-price term without expanding the doctrine. The distinction between Gehenna and Hades, the 70-times-seven expression, and the future-perfect wording for binding and releasing remain present. See editorial-matthew-middle.json.

## Closing Matthew comparisons

Eighty-three verses in Matthew 22-28 were compared, including all of Matthew 28. Eleven edits retain the visit, drinking anew rather than specifying new wine, the completed field purchase, numbered hours, supplied lima spelling, vinegar, silver, Rejoice, freedom from worry, and present I am. Matthew 24:36 retains the source edition's wording without inserting a clause from another Gospel. The final commission keeps the singular name with Father, Son, and Holy Spirit. No modern clock conversion or manuscript explanation was inserted. See editorial-matthew-closing.json.

## Matthew register candidates

Three register candidates were compared with the source. Matthew 26:73 now uses modern conversational speech. Matthew 16:3 and 16:18 were retained: they contain a modern connective and the name Peter plus a copula. The latter was already in the middle-Matthew comparison scope. See editorial-matthew-register.json.

## Delivery retest and note consolidation

The portable audit negative test initially failed its own assertion because it expected an older diagnostic string. The validator had rejected the missing source verse correctly. The test now requires the current source-reference diagnostic and the exact missing GEN 1:31 reference; no validator check was relaxed. Two early Genesis chapters had accumulated five editorial notes each. Their questions were consolidated into three notes per chapter without scripture changes. The full Genesis 15 comparison remains recorded earlier in this ledger. See editorial-note-consolidation.json and portable-validation-test.json.

## Early Mark comparisons

Seventy-five verses in Mark 2-7 were compared. Fourteen edits use readable paralysis and bleeding descriptions, retain grain without specifying wheat, preserve will be forgiven rather than can be forgiven, and remove an added still in the question about faith. Mark 7:19 no longer inserts an explicit parenthetical declaration by Jesus absent from the pinned sentence. Its clause relationship remains an expert-review question; verses 21-22 follow the supplied list order. The source references to Abiathar and the sisters, and Mark 7:16, remain present without harmonization to another edition. See editorial-mark-early.json.

## Middle Mark comparisons

Seventy-one verses in Mark 8-12 were compared. Fifteen edits preserve after three days, the exchange without an added buying-back claim, the conditional in Mark 9:23, the three repeated worm-and-fire verses, and speaking to Jesus rather than to one another in Mark 10:26. Heart, soul, mind, understanding, and strength remain distinguished. The two small coins retain their one-quadrans equivalence without modern currency conversion. The supplied fasting clause and salt/sacrifice clause remain present. See editorial-mark-middle.json.

## Direct Mark 13 draft during provider quota wait

All 37 verses of Mark 13 were translated and compared with the pinned English in the current Codex session after the signed-in generation service returned a temporary quota error. This is same-author AI self-review, not an independent or human review. The source Daniel reference, reader instruction, shortening of days, temporal warnings, and nor the Son wording remain present. No calendar prediction was added. The complete source and initial output are retained in codex-mark13.json.

## Direct Mark 16 draft

All 20 verses of Mark 16 were translated and compared with the pinned English in the current Codex session. The two Marys remain distinguished, the young man is not relabeled as an angel, and the repeated disbelief is retained. The ending follows this edition through verse 20 without a manuscript-history claim. Future and conditional wording remains distinct from commands. This is same-author AI self-review; human and independent review remain outstanding. See codex-mark16.json.

## Direct Mark 14 draft

All 72 verses were translated and compared with the pinned English in the current Codex session. The source prices, two rooster calls, three denials, repeated questions, new-covenant wording, and imperative in verse 41 remain present. The woman, sword wielder, and fleeing young man are not identified from outside the chapter. This is same-author AI self-review, not independent or human review. See codex-mark14.json.

## Direct Mark 15 draft

All 47 verses were translated and compared with the pinned English in the current Codex session. The third, sixth, and ninth hours remain unconverted. The source Eloi/lama wording is not harmonized to Matthew. The source verse 28, named witnesses, two robbers, direction of the torn curtain, and body-request sequence remain present. This is same-author AI self-review, not independent or human review. See codex-mark15.json. Mark is now a complete AI draft.

## Direct Luke 1 draft

All 80 verses of Luke 1 were translated and compared with the pinned English in the current Codex session during the provider quota wait. The respectful preface is kept distinct from narrative prose. Both announcements, the five/six/three-month statements, eighth-day naming, and the two songs are retained. Elizabeth is a relative without an invented cousin relationship. The horn and dawn images remain explicit; ambiguous syntax is an expert-review question. This is same-author AI self-review. See codex-luke1.json.

## Direct Luke 2 draft

All 52 verses of Luke 2 were translated and compared with the pinned English in the current Codex session. The enrollment is not silently turned into tax payment; the manger and inn remain distinct. The supplied goodwill wording, their purification, eight days, seven married years, about 84 widowed years, age twelve, and three-day search remain present. No governor chronology was reconstructed. Syria is provisionally 시리아 for this New Testament setting and flagged for cross-book consistency review. This is same-author AI self-review. See codex-luke2.json.

## Direct Luke 3 draft

All 38 verses of Luke 3 were translated and compared with the pinned English in the current Codex session. The fifteenth year and age about thirty remain unconverted. John's instructions to the three groups remain distinct. The genealogy uses a reviewed name map and each exact source parent sequence to render linked Korean father/son statements; every source entry remains in its own verse. Joanan/Jonan, both Cainan entries, and Aram are retained without harmonization. The as-was-supposed qualification remains explicit. This is same-author AI self-review. See codex-luke3.json, including the ordered genealogy-pair record.

## Direct Luke 4 draft

All 44 verses of Luke 4 were translated and compared with the pinned English in the current Codex session. The temptation sequence, supplied Scripture quotations, forty days, three years and six months, and closing Galilee reference remain intact. The source does not identify a particular calendar year in verse 19, and none was added. Naaman keeps the spelling and earlier-account identity used in Kings. This is same-author AI self-review. See codex-luke4.json.

## Direct Luke 5 draft

All 39 verses of Luke 5 were translated and compared with the pinned English in the current Codex session. The two boats, breaking net, catching alive, tiled roof, paralysis, and old-wine comparison remain present. The source does not name four carriers here, and no number was imported from Mark. Levi remains the supplied name. This is same-author AI self-review. See codex-luke5.json.

## Opening Luke readability follow-up

Ten source verses were revisited and nine were edited. Repeated names now clarify Elizabeth, Anna, and Simon's mother-in-law instead of ambiguous pronouns. Luke 1:55 repeats mercy from the preceding verse to avoid making forever modify continuous speaking; that syntactic resolution remains flagged. Dawn is rendered with the readable dawn-light image. Matthew 4:24 now matches Luke's contemporary-setting 시리아 and uses current Korean epilepsy/paralysis wording. Earlier Naaman remains 아람. See editorial-luke-opening-readability.json.

## Repeated-target candidate review

All 25 identical-Korean/different-English candidate groups in the 981-chapter snapshot were compared. Most differences are punctuation, connective placement, collective nouns, or equivalent reporting formulas. Leviticus 4:16 no longer adds a separate taking action copied from 4:5. First Kings 17:2 retains its temporal connector. Sixteen awkward word-came-and-spoke introductions were rephrased while preserving again/moreover. The ledger records each group and its decision; this targeted scan does not constitute a full semantic review. See editorial-repeated-targets.json.

## Empty source record preservation

The length-ratio audit exposed five empty publisher source records. The controller now enforces an exact editorial marker for these references and forbids that marker on nonempty source text. Luke 17:36 was relabeled and an unsupported statement about critical editions was removed. Source data and hashes remain unchanged. Verse coverage counts numbered records, including these five gaps; the full source has 31,098 nonempty texts. See editorial-empty-source.json.

## Luke 6-9 targeted source comparison

Compared 71 selected verses covering Sabbath, blessings and enemy love, the widow and unnamed woman, the sower, and discipleship. Eighteen edits restore grain without assuming wheat, direct second-person address, the pressed-down measure image, and return-of-loan wording. Added all, lap, and daily details absent from this source were removed. The seed interpretation no longer describes people literally falling on a rock; falling away is distinguished from stumbling. The set-free expression remains close to the English and needs expert review. See editorial-luke6-9.json.

## Luke middle chapters targeted source comparison

Compared 111 selected verses in Luke 10-18. Fifteen edits remove unsupported sibling ages, an assumed plant species and young-goat detail, and archaic planning language. The prayer uses a natural modern register. Luke 14:26 follows the supplied disregard rather than hate; Luke 16:9 keeps you as the subject of fail rather than replacing it with wealth. Both phrases remain explicit expert-review questions. Luke 17:3 restores against you, and 18:11 retains praying by himself. The source-specific prayer text, ancient measures, and younger/elder distinction explicitly present in Luke 15 are retained. See editorial-luke-middle.json.

## Luke closing chapters targeted source comparison

Compared 124 verses, including all of Luke 24. Fifteen edits preserve the supplied permission in 22:51 and affirmative causal wording in 22:70, remove an added today from the remembered quotation, retain all of you in 22:31, and avoid converting numbered hours to a modern clock. All nations in 24:47 is no longer narrowed to Gentiles, and Bethany is retained rather than its vicinity. The source explicitly includes honeycomb in 24:42 and this is retained. Luke 22:70 and ancient-hour presentation remain expert-review questions. See editorial-luke-closing.json.

Luke 21-24 editorial notes were also checked after the closing review. Unsupported Greek claims and stale descriptions of changed translations were removed. Notes now identify actual unresolved wording or presentation questions. See editorial-luke-notes.json.

## John 2-6 targeted source comparison

Compared 97 verses covering the first sign, new birth, the Samaritan conversation, the Father and Son, and the bread discourse. Nine edits clarify the feast manager and vessel rim, preserve own Father without adding a biological-father implication, restore our ancestors in 6:58 and present speech in 6:63, and reduce opaque wording. Source-specific who is in heaven (3:13) and a son of man (5:27) remain unharmonized. The address and speech register in John 2 and 4, and contextual Spirit vocabulary, still need expert review. See editorial-john-opening.json.

## John 7-10 targeted source comparison

Compared 99 verses. Nine edits remove an assumed sibling address, restore looking up rather than standing in 8:7, remove an added posture and subset qualifier, and retain whenever in 10:4 and both know and believe in 10:38. John 9:3 follows the direct supplied did not sin within its narrative context, with interpretation still requiring expert review. Present I AM in 8:58 and the supplied 7:53-8:11 account remain intact. See editorial-john-middle.json.

## John 11-16 targeted source comparison

Compared 149 verses, including all of John 14. Twenty-five edits remove an assumed older-brother relationship, restore gladness in 11:15, preserve pound without converting it to a Korean weight unit, and distinguish anointing from pouring. John 13:18 restores with me instead of my bread, 13:24 tells us rather than asks him, 14:4 keeps both knowledge clauses, and 14:19 restores the causal relation. John 15:3 retains pruning, and 15:7 preserves the supplied prediction rather than changing it to a command. John 16:33 retains present trouble. An entity-only Jerusalem typo in Matthew 21 was corrected; its verse text was already correct. Counselor terminology and 14:30 remain review questions. See editorial-john-discourse.json.

## John closing chapters targeted source comparison

Compared 134 verses, including all of John 17, 20, and 21. Nineteen edits remove an added name-given clause in 17:12, preserve legal accusation grounds, repeated slapping, ancient-hour numbering, vinegar, and explicit spirit at death. The request to break legs no longer prematurely identifies those legs as belonging to corpses. Linen is distinguished from hemp and rolled cloth from neatly folded cloth. John 20:29 retains a statement rather than a question, and 20:23 follows the supplied result/perfect wording without deciding its doctrinal interpretation. These and the comparison in 21:15 remain expert-review items. See editorial-john-closing.json.

## Linen terminology follow-up

A source-filtered scan found nine further verses rendering linen as hemp cloth. All nine were compared and corrected to 아마포; an added fine qualifier was removed in Leviticus 6:10 and John 19:40. This is a terminology-only review of those verses. The entity guide now records the distinction. See editorial-linen.json.

## Acts 1-6 targeted source comparison

Compared 106 verses. Twelve edits distinguish the addressed preface from narration, modernize quoted Psalm prayers, keep the supplied third-hour numbering, and clarify receiving from the Father in 2:33. David did not ascend no longer implies inability. The purpose connection across 3:19-20 is retained. The explicit Barnabas gloss follows encouragement, and the seven-person selection retains separate good-reputation and Spirit/wisdom qualifications. The source still governs the Pentecost language list, forty days, approximately 120 and 3,000 people, and the account of the field. See editorial-acts-opening.json.

## Acts 7-12 targeted source comparison

Compared 130 verses. Nineteen edits remove an implication of simultaneous ancestral deaths, restore Christ Jesus versus Lord Jesus (8:16), God versus Lord (8:22), and Christ versus Jesus (9:20). The supplied curse remains a wish rather than a prediction. Straight is rendered as an understandable street name. Other languages (10:46), constant prayer (12:5), getting dressed rather than a belt (12:8), and return to rather than from Jerusalem (12:25) follow this source. The 75-person count and Abraham attribution in Stephen's speech remain as supplied, without harmonization to another account. See editorial-acts-middle.json.

The next register scan found one additional archaic future connector in John 8:21. It was modernized while preserving the source departure, seeking, death, and inability-to-come clauses. The other 29 matches remain previously reviewed modern forms or proper names. See editorial-john-register.json.

## Acts 13-18 targeted source comparison

Compared 139 verses. Nineteen edits retain foster upbringing without inventing a shared wet nurse, restore cunning, preserve the present hand image, and follow today I have become your father. Jupiter and Mercury are retained instead of substituted names. The Lydia and enslaved-girl passages now have clearer referents; 16:17 restores to us, and 16:34 avoids expanding the explicit subject of believing. The Athens discourse keeps appointed times and the qualified seeking language. The supplied salvation wording in 13:23, empty 15:34, first-person travel shift, and Jerusalem feast statement in 18:21 remain intact. See editorial-acts-missions.json.

## Acts closing chapters targeted source comparison

Compared 196 verses in Acts 19-28. Twenty-four edits restore plural we, bread, thousands rather than tens of thousands, a warning rather than a question, the following night, and the proposed payment for release. The closing speeches preserve the supplied Lord and God phrase, first proclamation through resurrection, and persuasion rather than an invented time comparison. The voyage retains actual grounding, fathoms without a Korean-unit conversion, and two seas without adding a sandbar. The island account keeps Justice without adding a goddess identity, and the final quotation retains the people actively closing their eyes. Numbers and measurements remain source-linked. See editorial-acts-closing.json.

## Romans register candidates

Two new archaic-ending candidates were checked with six verses of adjacent context. Romans 3:27 and 6:13 were modernized while retaining the source law-of-faith contrast and imperative to present the body to God. This is a limited register check, not a review of the full new Romans draft. See editorial-romans-register.json.

## Romans 1-7 targeted source comparison

Compared 142 verses, including the complete chapters 5-7. Eighteen edits improve current prose and restore source details: hateful to God rather than hating God, apart from rather than unrelated to law, present falling short, approximate age, no accounting of sin without law, participation in resurrection, and persons handed over to teaching rather than teaching handed to persons. Romans 7:18 no longer adds an explicit inability statement. Spirit, righteousness, justification, and the teaching-framework metaphor still require expert review. See editorial-romans-early.json.

## Romans 9-13 targeted source comparison

Compared 116 verses, including all of chapters 10, 12, and 13. Fourteen edits preserve conditional wishing, heirs, the fulfillment of law, the abyss without inventing a sea, and respect without replacing it with fear. Modernized two archaic asking formulas and repaired nested quotation marks. Removed added regret from irrevocability and total destruction from stumbling/falling. The letter still needs book-wide register harmonization and expert review of its theological vocabulary. See editorial-romans-middle.json.

## Romans 14-16 full source comparison

Compared all 84 numbered records (83 nonempty texts) in these chapters. Eleven edits restore direct second-person address, present conditional condemnation, purpose rather than an achieved result, the root rather than a shoot from the root, the commitment to visit rather than only a wish, an actual risk to life, and writing without an invented dictation mechanism. The source doxology stays at 14:24-26, with the explicitly labeled empty entry at 16:25. Names and theological interpretations still need specialist review. See editorial-romans-closing.json.

## First Corinthians 1-7 targeted source comparison

Compared 123 verses. Twenty edits restore because of rather than birth from God, things rather than only thoughts, comparison rather than discernment, entreaty rather than advice, and becoming a father rather than literal begetting. The source Passover metaphor and quotation marks are retained. Corrected defect versus defeat, Gods Spirit terminology, affection versus unspecified duty, actual authority versus claiming rights, permission versus asking for understanding, and actual departure versus only desiring departure. Modernized three archaic future constructions and one imperative. Sexual, marital, spiritual, and ecclesial language still needs qualified contextual review. See editorial-corinthians1-early.json.

## First Corinthians 7-11 targeted source comparison

Compared 111 verses, including all of chapters 8, 10, and 11. Eighteen edits remove a fiancee identification not established by his virgin, an absent not-under-law clause, and a symbol of authority absent from 11:10. Preserved consciousness of an idol, existing weakness of conscience, divine approval, disqualification, playing, and mutual dependence. The provisional broad relationship wording in 7:36-38 and the authority/head metaphor require expert review. See editorial-corinthians1-middle.json.

## First Corinthians 12-14 full source comparison

Compared all 84 verses. Thirty edits consistently retain the supplied languages wording, service rather than office, best rather than larger gifts, brass and cymbal, bearing rather than covering, and a personal preference for intelligible speech. Removed an added agent and always from chapter 13. Corrected the musical illustration to what is played rather than which instrument, restored wives and the submission exception, and restored let him be ignorant at 14:38 instead of a different textual reading. See editorial-corinthians1-gifts.json.

## First Corinthians 15-16 full source comparison

Compared all 82 verses. Fifteen edits avoid assuming prematurity from wrong-time birth, make the resurrection negations explicit, restore assertion rather than condition in 15:44, preserve the plural source image phrase and the become/put-on distinction, improve the sting metaphor, and retain a curse wish rather than turning it into a prediction. The traveling companions and coordination in the closing instructions were clarified. Resurrection, natural/spiritual bodies, and baptism for the dead remain for expert review. See editorial-corinthians1-closing.json.

## Second Corinthians 1-7 targeted source comparison

Compared 105 verses, including all of chapters 3-5. Fifteen edits restore your behalf, flesh in the heart-tablet metaphor, prediction rather than command for light, through many rather than to many, all things made new, fasting, the unequal yoke, and you rather than we as the temple. Removed an added repeat visit and clarified the purpose clause in 7:9. Theological vocabulary and overall register remain for human review. See editorial-corinthians2-early.json.

## Second Corinthians 8-13 targeted source comparison

Compared 108 verses, including all of chapters 9, 12, and 13. Twenty edits preserve receiving a gift, generosity versus greed, a blessing rather than a forecast, humility rather than timidity, and flesh without adding sensual desire. Removed invented towers, restored rods rather than whips and fasting rather than unspecified hunger, kept hypothetical accusation hypothetical, and corrected I rather than we in 13:7. The thorn is left unspecified and the source unless qualification is retained. See editorial-corinthians2-closing.json.

## Galatians targeted source comparison

Compared 127 verses, including all of chapters 2, 3, 5, and 6. Eighteen edits restore curse wishes, explicitly stated fear, the one-hour expression, impartiality, through rather than in, the right hand of fellowship, the in-Christ clause at 3:17, your hearts and we pronouns, all, and faith rather than faithfulness in the listed fruit. The final blessing retains and instead of identifying the groups with namely. Remaining interpretation questions are separate from the verse text. See editorial-galatians.json.

## Ephesians 1-3 source comparison

Compared 64 verses: 1:3-23 and all of chapters 2-3. Ten edits restore purpose, wisdom as the manner rather than an added gift, the source parenthesis, completed destruction of hostility, the if qualification, and the comparative disclosure across generations. Replaced especially opaque administration and sealing vocabulary with provisional contemporary wording while keeping the imagery and source relationships. See editorial-ephesians-early.json.

## Ephesians 4-6 full source comparison

Compared all 89 verses. Eight edits restore until, keep the idolater parenthesis attached to the greedy person, specify wine, preserve the purpose form in the quoted promise, and make sealing and full armor more accessible. The final request remains a prayer request. The household code, descent, and preparation-of-the-gospel wording remain for specialist interpretation; plain and polite prose needs a final consistency pass. See editorial-ephesians-closing.json.

## Philippians full source comparison

Compared all 104 verses. Thirteen edits restore disappointed, remove an added claim of entitlement, preserve purpose and harmlessness, retain the holding-up image, modernize an embedded archaic future, and distinguish rejoicing from boasting and loss from harm. Once and again is no longer one or two times, and the unspecified gift is not labeled as goods. Peace vocabulary is consistent within the letter. See editorial-philippians.json.

## Colossians full source comparison

Compared all 95 verses. Thirteen edits preserve the firstborn image, by rather than in for creation, the if qualification, Father and Christ in the mystery phrase, and elemental spirits rather than substituting Galatians principles. Removed added false from humility, restored walking in love, retained grace, and corrected complete rather than confident. The quoted instruction and inheritance wording were clarified. Firstborn, fullness, and other theological interpretations remain open for qualified review. See editorial-colossians.json.

## Thessalonian letters targeted source comparison

Compared 111 verses, including all of First Thessalonians 4-5 and Second Thessalonians 2. Twenty edits restore repeated attempts rather than one or two, remove a present-practice clause absent from 1 Thessalonians 4:1, retain lack of nothing, and avoid adding greatest to respect. The second letter preserves suffering for the kingdom, passive removal of the restrainer, manifestation without invented brightness, purpose rather than completed salvation, rebellion wording, and the third-person food instruction. See editorial-thessalonians.json.

## First Timothy targeted source comparison

Compared 99 verses, including all of chapters 2-3 and 6. Sixteen edits restore insolence without invented violence, immortality, doubt rather than quarrelling, their wives, the general conduct instruction, diligence rather than meditation, and the joined qualifications rather than alternatives. The money warning retains all kinds of evil, and eternal life no longer adds true. Authority, childbearing, ministry offices, and historical household instructions remain for specialist review. See editorial-timothy1.json.

## Second Timothy and Titus source comparison

Compared all 129 verses. 23 edits restore the blessing rather than a prediction, remove an added with me and house, preserve the gangrene image, distinguish witnessing from sharing the same persecution, and retain the God-breathed metaphor. Titus keeps young wives, the first and second warnings, and the general lawyer term. Contextual renderings of hand-laying, following sufferings, and the appearance/Kingdom relationship remain explicitly provisional. See editorial-timothy2-titus.json.

## Hebrews 1-8 targeted source comparison

Compared 102 verses, including all of chapters 1, 4, 6, and 8. Fourteen edits preserve become your father, remove an added remaining-promise clause, retain the unresolved pronoun, distinguish crops from vegetables and beneficiaries from laborers, and restore the given-as-law clause. The theological argument and quotation tradition are not certified by this English-source review. See editorial-hebrews-early.json.

## Hebrews 9-13 targeted source comparison

Compared 145 verses, including all of chapters 9-10 and 13. Sixteen edits preserve a lampstand, the necessity of death rather than proof of death, faith as assurance, the allegiance clause, and Italians without an inferred journey. Added food regulations, Holy in eternal Spirit, and all in the consuming-fire image were removed. English-source comparison does not settle the original-language covenant argument or theological interpretation. See editorial-hebrews-closing.json.

## James full-book source comparison

Compared all 108 verses and made 20 edits. These restore purpose clauses, humility, dirty clothing, the position beside a footstool, royal law, both fresh and salt water, and the source tense. Gehenna and Spirit remain distinct source terms; a soul is not silently resolved to his soul. Healing passages remain religious text rather than a separate health recommendation. See editorial-james.json.

## First Peter full-book source comparison

Compared all 105 verses and made 16 edits. These restore the living-word attachment, indicative house-building, purpose clauses, wicked rather than merely difficult masters, courteous rather than humble, mutual submission, and self-control. The final greeting retains the female pronoun without identifying its referent. Several Spirit, priesthood, household, and disappointed renderings remain explicit specialist-review questions. See editorial-peter1.json.

## Second Peter and Jude full-book source comparison

Compared all 86 verses and made 16 edits. These restore near vision, statement rather than prohibition, conditional framing, our/us, amid water, in the night, and the ordinal counted from Adam. Dignitaries and celestial beings remain different source expressions. Jude 1:4 apposition remains explicitly unresolved for specialist review. See editorial-peter2-jude.json.

## The three letters of John full-book source comparison

Compared all 132 verses and made 16 edits. These clarify Counselor, preserve none rather than an ambiguous not-all wording, restore the conditional and purpose clauses, and retain we in the reward statement. The source-specific faith, witnesses, continued belief, and keeps himself readings remain unharmonized. No named identity is invented for the chosen lady or her sister. See editorial-john-letters.json.

## Revelation 1-11 targeted source comparison

Compared 145 verses and made 30 edits, including source-matched lampstand terminology. These restore my God, the shared bed image, multiplied numbers, come and see, the unconverted choenix unit, singular authority, completion of a course rather than a number, and day rather than hour. The source-specific visionary wording is retained without a historical or future-date interpretation. See editorial-revelation-early.json.

## Revelation 12-17 source comparison

Compared all 102 verses and made 15 edits. These preserve unresolved they, plural times without an invented two, the coordinated blasphemy objects, the name-writing time clause, and virginity rather than generic purity. Harp no longer becomes a Korean geomungo, scorched no longer implies burned away, and the approximate talent remains approximate. Symbol interpretation remains outside the draft. See editorial-revelation-middle.json.

## Revelation 18-22 source comparison

Compared all 108 verses and made 11 edits. These remove unsupported aromatic wood, distinguish incense/perfume, retain the addressee and quotation boundary, preserve coordinated groups, and clarify the reward wording. Revelation 21:16 deliberately retains the source value of 12012 stadia. The last chapter keeps this editions commandments, tree-of-life, and all-the-saints readings. See editorial-revelation-closing.json.

## Final terminology and remaining copyediting

The four Gospel occurrences of Counselor now use a plain helper expression consistent with 1 John 2:1. Four stale entity labels were aligned with the accepted manuscript. Entity identity descriptions remain unverified working metadata, not published verse annotations. A supplementary register search found further formal dialogue endings and stylistic differences across batches; the targeted modern-language passes do not constitute exhaustive copyediting. Human review of wording, names, register, pronoun reference, and interpretation is still required. The Revelation 16:16 note flagged by the heuristic is source-supported: the supplied English itself names Hebrew and Harmagedon. See editorial-final-terminology.json.

## Final delivery review, 2026-09-29

The complete AI draft contains all 66 books, 1,189 chapters, and 31,103 numbered source records. Of these, 31,098 have English text and five carry the exact editorial empty-source marker. All 173 accepted generation receipts were verified. The source and delivery audits pass with the complete-corpus gate enabled.

An extracted copy of the complete ZIP passed the portable audit. Deliberately missing, reordered, and duplicate verse records were detected. Separate delivery tests detected altered JSONL text, a missing JSONL verse, a broken HTML navigation entry, and altered Markdown text. Empty-source guards and the writer/checkpoint lock test passed. These tests change disposable extracted copies only.

The final offline HTML reader was opened in the Codex in-app browser. The opening disclosure showed 1,189 of 1,189 chapters; expanding Revelation and clicking chapter 22 navigated to the final chapter, whose Korean text rendered legibly. The local test server is temporary; the delivered HTML has no external dependencies.

This was a Codex self-review, not an independent human review. No chapter is marked human-reviewed. The audit reports candidate entity variants, repeated short phrases, and register matches; these are not all errors, and the heuristic is not exhaustive. The supplemental register search confirms that formal dialogue and mixed prose endings still need copyediting. No original-language fidelity, doctrinal approval, freedom from every possible third-party claim, or publication readiness is certified.