feat(lenses): Genesis 1 proof-of-concept (#820, #1781)

[CraigBuckmaster]

Genesis 1 — proof-of-concept for Epic #820 / #1781

Single-chapter slice of the Genesis pilot. Ships 6 lens entries for Genesis 1 to validate the entire Phase 0 + Phase 1 + quality-gate pipeline against real content before scaling.

The remaining 49 Genesis chapters stay under #1781 — this PR is the smoke test that proves the pipeline works end-to-end. The regex bug it caught (see below) is exactly the kind of thing a single-chapter PoC is supposed to surface.

Coverage decisions

Curated sparse: 6 of 8 lenses applied. Skipped:

• devotional — Genesis 1 is theological foundation; devotional response is more naturally drawn from later texts. Forcing it here would invite filler.
• mission — the missional thread of Scripture is rooted in Genesis 12 (Abrahamic call), not Genesis 1. The accuracy auditor would flag a missional Genesis-1 entry as fabrication.

The 6 included:

Lens	What it surfaces
grammatical	Sevenfold formula, opening/closing markers, bara bookending vv 1 and 27
literary	Forming/filling structure across days 1-3 paired with 4-6
typological	Spirit hovering, seventh-day rest, light/dark separation
redemptive	Image-of-God commission grounds the whole arc to Christ
christocentric	John 1:1-3, Col 1:16, divine speech as the same Word
canonical	Creation-by-speech echoes; Rev 21-22 recapitulation

Each entry has a panel_filter and panel_order selected to surface the existing chapter panels that that lens most illuminates, in the order that best supports the reading.

Quality gate results (local)

$ python3 _tools/schema_validator.py
RESULTS: 138138 passed, 0 failed, 19 warnings
$ python3 _tools/build_sqlite.py
[OK] hermeneutic_lenses: 8 lenses, 6 contents
$ python3 _tools/validate_sqlite.py
RESULTS: 101 passed, 0 failed, 2 warnings
$ python3 _tools/lens_quality_scorer.py --chapter gen1
LENS QUALITY REPORT — 6 entries
Passing (>= 90): 6
Failing (< 90): 0
Average: 100.0/100

All entries 257–274 chars (within the 80–280 schema range). All score 25/25 on every DVCR dimension.

Tier-2 accuracy audit

accuracy_meta_extractors.extract_hermeneutic_lenses() produces 6 claims from this content (one per lens entry), ready for accuracy_auditor.py --tier 2 to verify in CI. I can't run the tier-2 audit locally (no API key in the dev container — that's CI's job).

If any claim fails the tier-2 verifier, the auditor will flag it as a position the named hermeneutical school doesn't actually advocate, and the entry needs revision before merge.

The exegesis I think is most likely to draw an audit flag (transparency note): the typological claim that light/dark separation foreshadows Revelation 21 is defensible (Beale, The Temple and the Church's Mission; many patristic readings) but it's the closest to a stretch in this batch. The other two anchors in that entry — Spirit hovering → new creation, seventh-day rest → Hebrews 4 — are rock-solid. If the audit pushes back on the Rev 21 anchor, I'll trim it.

Scorer fix included

lens_quality_scorer.py had a regex bug that this content surfaced: VERSE_REF_RE was matching numbers after a colon, so Psalm 33:6 got parsed as a verse reference and flagged 33-6 as out-of-range for Genesis 1 (which has 31 verses).

Fixed by tightening VERSE_REF_RE to require an explicit v / vv prefix for in-chapter validation, and adding a separate looser ANY_VERSE_PATTERN for the Density "has any verse anchor?" check. Verified across 10 representative cases:

Input	Should match?	Did match?
v 5	yes	✓
vv 5-7	yes	✓
v. 31	yes	✓
(vv 26-28)	yes	✓
Psalm 33:6	no	✓
Heb 4:9-11	no	✓
Genesis 1:3	no	✓
Col 1:15	no	✓
verses 26 and 27	no	✓ (no v/vv prefix)
vv. 5, 8, 13	yes (first match)	✓

This regex fix is part of the same PR because the lens content immediately depends on it — no point shipping content that would fail the scorer for the wrong reason.

What this PR does NOT do

• The remaining 49 Genesis chapters (still tracked under [#820 / 2] Pilot batch — Genesis (50 chapters) #1781)
• The 8 genre batches ([#820 / 3] Pentateuch rest — Exodus, Leviticus, Numbers, Deuteronomy (137 ch) #1782 through [#820 / 10] Revelation (22 ch) #1789) — they were always going to wait for [#820 / 2] Pilot batch — Genesis (50 chapters) #1781 to ship clean
• Any tier-2 audit results (CI runs them)

Rollback plan

Pure additive content + one-line scorer regex tightening. git revert is clean — nothing references the new content row except the app's lens-toggle UI, which currently shows a guidance string and (per #1790) optionally filters/reorders panels. Reverting just makes Genesis 1's lens toggle show no guidance, same as the rest of the corpus.

Refs #820, #1781.

[github-actions]

Content Pipeline Results

✅ All pipeline checks passed

Step	Status	Details
Schema Validation	✅	138138 passed, 0 failed
Build DB	✅	—
DB Integrity	✅	101 passed, 0 failed

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3973	❌ 0	3973
Suites	✅ 531	❌ 0	531

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 95.8s

[CraigBuckmaster]

Genesis 1 — proof-of-concept for Epic #820 / #1781

Single-chapter slice of the Genesis pilot. Ships 6 lens entries for Genesis 1 to validate the entire Phase 0 + Phase 1 + quality-gate pipeline against real content before scaling.

The remaining 49 Genesis chapters stay under #1781 — this PR is the smoke test that proves the pipeline works end-to-end. The regex bug it caught (see below) is exactly the kind of thing a single-chapter PoC is supposed to surface.

Coverage decisions

Curated sparse: 6 of 8 lenses applied. Skipped:

• devotional — Genesis 1 is theological foundation; devotional response is more naturally drawn from later texts. Forcing it here would invite filler.
• mission — the missional thread of Scripture is rooted in Genesis 12 (Abrahamic call), not Genesis 1. The accuracy auditor would flag a missional Genesis-1 entry as fabrication.

The 6 included:

Lens	What it surfaces
grammatical	Sevenfold formula, opening/closing markers, bara bookending vv 1 and 27
literary	Forming/filling structure across days 1-3 paired with 4-6
typological	Spirit hovering, seventh-day rest, light/dark separation
redemptive	Image-of-God commission grounds the whole arc to Christ
christocentric	John 1:1-3, Col 1:16, divine speech as the same Word
canonical	Creation-by-speech echoes; Rev 21-22 recapitulation

Each entry has a panel_filter and panel_order selected to surface the existing chapter panels that that lens most illuminates, in the order that best supports the reading.

Quality gate results (local)

$ python3 _tools/schema_validator.py
RESULTS: 138138 passed, 0 failed, 19 warnings
$ python3 _tools/build_sqlite.py
[OK] hermeneutic_lenses: 8 lenses, 6 contents
$ python3 _tools/validate_sqlite.py
RESULTS: 101 passed, 0 failed, 2 warnings
$ python3 _tools/lens_quality_scorer.py --chapter gen1
LENS QUALITY REPORT — 6 entries
Passing (>= 90): 6
Failing (< 90): 0
Average: 100.0/100

All entries 257–274 chars (within the 80–280 schema range). All score 25/25 on every DVCR dimension.

Tier-2 accuracy audit

accuracy_meta_extractors.extract_hermeneutic_lenses() produces 6 claims from this content (one per lens entry), ready for accuracy_auditor.py --tier 2 to verify in CI. I can't run the tier-2 audit locally (no API key in the dev container — that's CI's job).

If any claim fails the tier-2 verifier, the auditor will flag it as a position the named hermeneutical school doesn't actually advocate, and the entry needs revision before merge.

The exegesis I think is most likely to draw an audit flag (transparency note): the typological claim that light/dark separation foreshadows Revelation 21 is defensible (Beale, The Temple and the Church's Mission; many patristic readings) but it's the closest to a stretch in this batch. The other two anchors in that entry — Spirit hovering → new creation, seventh-day rest → Hebrews 4 — are rock-solid. If the audit pushes back on the Rev 21 anchor, I'll trim it.

Scorer fix included

lens_quality_scorer.py had a regex bug that this content surfaced: VERSE_REF_RE was matching numbers after a colon, so Psalm 33:6 got parsed as a verse reference and flagged 33-6 as out-of-range for Genesis 1 (which has 31 verses).

Fixed by tightening VERSE_REF_RE to require an explicit v / vv prefix for in-chapter validation, and adding a separate looser ANY_VERSE_PATTERN for the Density "has any verse anchor?" check. Verified across 10 representative cases:

Input	Should match?	Did match?
v 5	yes	✓
vv 5-7	yes	✓
v. 31	yes	✓
(vv 26-28)	yes	✓
Psalm 33:6	no	✓
Heb 4:9-11	no	✓
Genesis 1:3	no	✓
Col 1:15	no	✓
verses 26 and 27	no	✓ (no v/vv prefix)
vv. 5, 8, 13	yes (first match)	✓

This regex fix is part of the same PR because the lens content immediately depends on it — no point shipping content that would fail the scorer for the wrong reason.

What this PR does NOT do

• The remaining 49 Genesis chapters (still tracked under [#820 / 2] Pilot batch — Genesis (50 chapters) #1781)
• The 8 genre batches ([#820 / 3] Pentateuch rest — Exodus, Leviticus, Numbers, Deuteronomy (137 ch) #1782 through [#820 / 10] Revelation (22 ch) #1789) — they were always going to wait for [#820 / 2] Pilot batch — Genesis (50 chapters) #1781 to ship clean
• Any tier-2 audit results (CI runs them)

Rollback plan

Pure additive content + one-line scorer regex tightening. git revert is clean — nothing references the new content row except the app's lens-toggle UI, which currently shows a guidance string and (per #1790) optionally filters/reorders panels. Reverting just makes Genesis 1's lens toggle show no guidance, same as the rest of the corpus.

Refs #820, #1781.

[github-actions]

Content Pipeline Results

✅ All pipeline checks passed

Step	Status	Details
Schema Validation	✅	138138 passed, 0 failed
Build DB	✅	—
DB Integrity	✅	101 passed, 0 failed

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3973	❌ 0	3973
Suites	✅ 531	❌ 0	531

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 95.8s

[CraigBuckmaster]

Genesis 1 — proof-of-concept for Epic #820 / #1781

Single-chapter slice of the Genesis pilot. Ships 6 lens entries for Genesis 1 to validate the entire Phase 0 + Phase 1 + quality-gate pipeline against real content before scaling.

The remaining 49 Genesis chapters stay under #1781 — this PR is the smoke test that proves the pipeline works end-to-end. The regex bug it caught (see below) is exactly the kind of thing a single-chapter PoC is supposed to surface.

Coverage decisions

Curated sparse: 6 of 8 lenses applied. Skipped:

• devotional — Genesis 1 is theological foundation; devotional response is more naturally drawn from later texts. Forcing it here would invite filler.
• mission — the missional thread of Scripture is rooted in Genesis 12 (Abrahamic call), not Genesis 1. The accuracy auditor would flag a missional Genesis-1 entry as fabrication.

The 6 included:

Lens	What it surfaces
grammatical	Sevenfold formula, opening/closing markers, bara bookending vv 1 and 27
literary	Forming/filling structure across days 1-3 paired with 4-6
typological	Spirit hovering, seventh-day rest, light/dark separation
redemptive	Image-of-God commission grounds the whole arc to Christ
christocentric	John 1:1-3, Col 1:16, divine speech as the same Word
canonical	Creation-by-speech echoes; Rev 21-22 recapitulation

Each entry has a panel_filter and panel_order selected to surface the existing chapter panels that that lens most illuminates, in the order that best supports the reading.

Quality gate results (local)

$ python3 _tools/schema_validator.py
RESULTS: 138138 passed, 0 failed, 19 warnings
$ python3 _tools/build_sqlite.py
[OK] hermeneutic_lenses: 8 lenses, 6 contents
$ python3 _tools/validate_sqlite.py
RESULTS: 101 passed, 0 failed, 2 warnings
$ python3 _tools/lens_quality_scorer.py --chapter gen1
LENS QUALITY REPORT — 6 entries
Passing (>= 90): 6
Failing (< 90): 0
Average: 100.0/100

All entries 257–274 chars (within the 80–280 schema range). All score 25/25 on every DVCR dimension.

Tier-2 accuracy audit

accuracy_meta_extractors.extract_hermeneutic_lenses() produces 6 claims from this content (one per lens entry), ready for accuracy_auditor.py --tier 2 to verify in CI. I can't run the tier-2 audit locally (no API key in the dev container — that's CI's job).

If any claim fails the tier-2 verifier, the auditor will flag it as a position the named hermeneutical school doesn't actually advocate, and the entry needs revision before merge.

The exegesis I think is most likely to draw an audit flag (transparency note): the typological claim that light/dark separation foreshadows Revelation 21 is defensible (Beale, The Temple and the Church's Mission; many patristic readings) but it's the closest to a stretch in this batch. The other two anchors in that entry — Spirit hovering → new creation, seventh-day rest → Hebrews 4 — are rock-solid. If the audit pushes back on the Rev 21 anchor, I'll trim it.

Scorer fix included

lens_quality_scorer.py had a regex bug that this content surfaced: VERSE_REF_RE was matching numbers after a colon, so Psalm 33:6 got parsed as a verse reference and flagged 33-6 as out-of-range for Genesis 1 (which has 31 verses).

Fixed by tightening VERSE_REF_RE to require an explicit v / vv prefix for in-chapter validation, and adding a separate looser ANY_VERSE_PATTERN for the Density "has any verse anchor?" check. Verified across 10 representative cases:

Input	Should match?	Did match?
v 5	yes	✓
vv 5-7	yes	✓
v. 31	yes	✓
(vv 26-28)	yes	✓
Psalm 33:6	no	✓
Heb 4:9-11	no	✓
Genesis 1:3	no	✓
Col 1:15	no	✓
verses 26 and 27	no	✓ (no v/vv prefix)
vv. 5, 8, 13	yes (first match)	✓

This regex fix is part of the same PR because the lens content immediately depends on it — no point shipping content that would fail the scorer for the wrong reason.

What this PR does NOT do

• The remaining 49 Genesis chapters (still tracked under [#820 / 2] Pilot batch — Genesis (50 chapters) #1781)
• The 8 genre batches ([#820 / 3] Pentateuch rest — Exodus, Leviticus, Numbers, Deuteronomy (137 ch) #1782 through [#820 / 10] Revelation (22 ch) #1789) — they were always going to wait for [#820 / 2] Pilot batch — Genesis (50 chapters) #1781 to ship clean
• Any tier-2 audit results (CI runs them)

Rollback plan

Pure additive content + one-line scorer regex tightening. git revert is clean — nothing references the new content row except the app's lens-toggle UI, which currently shows a guidance string and (per #1790) optionally filters/reorders panels. Reverting just makes Genesis 1's lens toggle show no guidance, same as the rest of the corpus.

Refs #820, #1781.

[github-actions]

Content Pipeline Results

✅ All pipeline checks passed

Step	Status	Details
Schema Validation	✅	138138 passed, 0 failed
Build DB	✅	—
DB Integrity	✅	101 passed, 0 failed

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3973	❌ 0	3973
Suites	✅ 531	❌ 0	531

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 95.8s

[CraigBuckmaster]

Genesis 1 — proof-of-concept for Epic #820 / #1781

Single-chapter slice of the Genesis pilot. Ships 6 lens entries for Genesis 1 to validate the entire Phase 0 + Phase 1 + quality-gate pipeline against real content before scaling.

The remaining 49 Genesis chapters stay under #1781 — this PR is the smoke test that proves the pipeline works end-to-end. The regex bug it caught (see below) is exactly the kind of thing a single-chapter PoC is supposed to surface.

Coverage decisions

Curated sparse: 6 of 8 lenses applied. Skipped:

• devotional — Genesis 1 is theological foundation; devotional response is more naturally drawn from later texts. Forcing it here would invite filler.
• mission — the missional thread of Scripture is rooted in Genesis 12 (Abrahamic call), not Genesis 1. The accuracy auditor would flag a missional Genesis-1 entry as fabrication.

The 6 included:

Lens	What it surfaces
grammatical	Sevenfold formula, opening/closing markers, bara bookending vv 1 and 27
literary	Forming/filling structure across days 1-3 paired with 4-6
typological	Spirit hovering, seventh-day rest, light/dark separation
redemptive	Image-of-God commission grounds the whole arc to Christ
christocentric	John 1:1-3, Col 1:16, divine speech as the same Word
canonical	Creation-by-speech echoes; Rev 21-22 recapitulation

Each entry has a panel_filter and panel_order selected to surface the existing chapter panels that that lens most illuminates, in the order that best supports the reading.

Quality gate results (local)

$ python3 _tools/schema_validator.py
RESULTS: 138138 passed, 0 failed, 19 warnings
$ python3 _tools/build_sqlite.py
[OK] hermeneutic_lenses: 8 lenses, 6 contents
$ python3 _tools/validate_sqlite.py
RESULTS: 101 passed, 0 failed, 2 warnings
$ python3 _tools/lens_quality_scorer.py --chapter gen1
LENS QUALITY REPORT — 6 entries
Passing (>= 90): 6
Failing (< 90): 0
Average: 100.0/100

All entries 257–274 chars (within the 80–280 schema range). All score 25/25 on every DVCR dimension.

Tier-2 accuracy audit

accuracy_meta_extractors.extract_hermeneutic_lenses() produces 6 claims from this content (one per lens entry), ready for accuracy_auditor.py --tier 2 to verify in CI. I can't run the tier-2 audit locally (no API key in the dev container — that's CI's job).

If any claim fails the tier-2 verifier, the auditor will flag it as a position the named hermeneutical school doesn't actually advocate, and the entry needs revision before merge.

The exegesis I think is most likely to draw an audit flag (transparency note): the typological claim that light/dark separation foreshadows Revelation 21 is defensible (Beale, The Temple and the Church's Mission; many patristic readings) but it's the closest to a stretch in this batch. The other two anchors in that entry — Spirit hovering → new creation, seventh-day rest → Hebrews 4 — are rock-solid. If the audit pushes back on the Rev 21 anchor, I'll trim it.

Scorer fix included

lens_quality_scorer.py had a regex bug that this content surfaced: VERSE_REF_RE was matching numbers after a colon, so Psalm 33:6 got parsed as a verse reference and flagged 33-6 as out-of-range for Genesis 1 (which has 31 verses).

Fixed by tightening VERSE_REF_RE to require an explicit v / vv prefix for in-chapter validation, and adding a separate looser ANY_VERSE_PATTERN for the Density "has any verse anchor?" check. Verified across 10 representative cases:

Input	Should match?	Did match?
v 5	yes	✓
vv 5-7	yes	✓
v. 31	yes	✓
(vv 26-28)	yes	✓
Psalm 33:6	no	✓
Heb 4:9-11	no	✓
Genesis 1:3	no	✓
Col 1:15	no	✓
verses 26 and 27	no	✓ (no v/vv prefix)
vv. 5, 8, 13	yes (first match)	✓

This regex fix is part of the same PR because the lens content immediately depends on it — no point shipping content that would fail the scorer for the wrong reason.

What this PR does NOT do

• The remaining 49 Genesis chapters (still tracked under [#820 / 2] Pilot batch — Genesis (50 chapters) #1781)
• The 8 genre batches ([#820 / 3] Pentateuch rest — Exodus, Leviticus, Numbers, Deuteronomy (137 ch) #1782 through [#820 / 10] Revelation (22 ch) #1789) — they were always going to wait for [#820 / 2] Pilot batch — Genesis (50 chapters) #1781 to ship clean
• Any tier-2 audit results (CI runs them)

Rollback plan

Pure additive content + one-line scorer regex tightening. git revert is clean — nothing references the new content row except the app's lens-toggle UI, which currently shows a guidance string and (per #1790) optionally filters/reorders panels. Reverting just makes Genesis 1's lens toggle show no guidance, same as the rest of the corpus.

Refs #820, #1781.

[github-actions]

Content Pipeline Results

✅ All pipeline checks passed

Step	Status	Details
Schema Validation	✅	138138 passed, 0 failed
Build DB	✅	—
DB Integrity	✅	101 passed, 0 failed

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3973	❌ 0	3973
Suites	✅ 531	❌ 0	531

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 95.8s

[CraigBuckmaster]

Genesis 1 — proof-of-concept for Epic #820 / #1781

Single-chapter slice of the Genesis pilot. Ships 6 lens entries for Genesis 1 to validate the entire Phase 0 + Phase 1 + quality-gate pipeline against real content before scaling.

The remaining 49 Genesis chapters stay under #1781 — this PR is the smoke test that proves the pipeline works end-to-end. The regex bug it caught (see below) is exactly the kind of thing a single-chapter PoC is supposed to surface.

Coverage decisions

Curated sparse: 6 of 8 lenses applied. Skipped:

• devotional — Genesis 1 is theological foundation; devotional response is more naturally drawn from later texts. Forcing it here would invite filler.
• mission — the missional thread of Scripture is rooted in Genesis 12 (Abrahamic call), not Genesis 1. The accuracy auditor would flag a missional Genesis-1 entry as fabrication.

The 6 included:

Lens	What it surfaces
grammatical	Sevenfold formula, opening/closing markers, bara bookending vv 1 and 27
literary	Forming/filling structure across days 1-3 paired with 4-6
typological	Spirit hovering, seventh-day rest, light/dark separation
redemptive	Image-of-God commission grounds the whole arc to Christ
christocentric	John 1:1-3, Col 1:16, divine speech as the same Word
canonical	Creation-by-speech echoes; Rev 21-22 recapitulation

Each entry has a panel_filter and panel_order selected to surface the existing chapter panels that that lens most illuminates, in the order that best supports the reading.

Quality gate results (local)

$ python3 _tools/schema_validator.py
RESULTS: 138138 passed, 0 failed, 19 warnings
$ python3 _tools/build_sqlite.py
[OK] hermeneutic_lenses: 8 lenses, 6 contents
$ python3 _tools/validate_sqlite.py
RESULTS: 101 passed, 0 failed, 2 warnings
$ python3 _tools/lens_quality_scorer.py --chapter gen1
LENS QUALITY REPORT — 6 entries
Passing (>= 90): 6
Failing (< 90): 0
Average: 100.0/100

All entries 257–274 chars (within the 80–280 schema range). All score 25/25 on every DVCR dimension.

Tier-2 accuracy audit

accuracy_meta_extractors.extract_hermeneutic_lenses() produces 6 claims from this content (one per lens entry), ready for accuracy_auditor.py --tier 2 to verify in CI. I can't run the tier-2 audit locally (no API key in the dev container — that's CI's job).

If any claim fails the tier-2 verifier, the auditor will flag it as a position the named hermeneutical school doesn't actually advocate, and the entry needs revision before merge.

The exegesis I think is most likely to draw an audit flag (transparency note): the typological claim that light/dark separation foreshadows Revelation 21 is defensible (Beale, The Temple and the Church's Mission; many patristic readings) but it's the closest to a stretch in this batch. The other two anchors in that entry — Spirit hovering → new creation, seventh-day rest → Hebrews 4 — are rock-solid. If the audit pushes back on the Rev 21 anchor, I'll trim it.

Scorer fix included

lens_quality_scorer.py had a regex bug that this content surfaced: VERSE_REF_RE was matching numbers after a colon, so Psalm 33:6 got parsed as a verse reference and flagged 33-6 as out-of-range for Genesis 1 (which has 31 verses).

Fixed by tightening VERSE_REF_RE to require an explicit v / vv prefix for in-chapter validation, and adding a separate looser ANY_VERSE_PATTERN for the Density "has any verse anchor?" check. Verified across 10 representative cases:

Input	Should match?	Did match?
v 5	yes	✓
vv 5-7	yes	✓
v. 31	yes	✓
(vv 26-28)	yes	✓
Psalm 33:6	no	✓
Heb 4:9-11	no	✓
Genesis 1:3	no	✓
Col 1:15	no	✓
verses 26 and 27	no	✓ (no v/vv prefix)
vv. 5, 8, 13	yes (first match)	✓

This regex fix is part of the same PR because the lens content immediately depends on it — no point shipping content that would fail the scorer for the wrong reason.

What this PR does NOT do

• The remaining 49 Genesis chapters (still tracked under [#820 / 2] Pilot batch — Genesis (50 chapters) #1781)
• The 8 genre batches ([#820 / 3] Pentateuch rest — Exodus, Leviticus, Numbers, Deuteronomy (137 ch) #1782 through [#820 / 10] Revelation (22 ch) #1789) — they were always going to wait for [#820 / 2] Pilot batch — Genesis (50 chapters) #1781 to ship clean
• Any tier-2 audit results (CI runs them)

Rollback plan

Pure additive content + one-line scorer regex tightening. git revert is clean — nothing references the new content row except the app's lens-toggle UI, which currently shows a guidance string and (per #1790) optionally filters/reorders panels. Reverting just makes Genesis 1's lens toggle show no guidance, same as the rest of the corpus.

Refs #820, #1781.

[github-actions]

Content Pipeline Results

✅ All pipeline checks passed

Step	Status	Details
Schema Validation	✅	138138 passed, 0 failed
Build DB	✅	—
DB Integrity	✅	101 passed, 0 failed

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3973	❌ 0	3973
Suites	✅ 531	❌ 0	531

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 95.8s

[CraigBuckmaster]

Genesis 1 — proof-of-concept for Epic #820 / #1781

Single-chapter slice of the Genesis pilot. Ships 6 lens entries for Genesis 1 to validate the entire Phase 0 + Phase 1 + quality-gate pipeline against real content before scaling.

The remaining 49 Genesis chapters stay under #1781 — this PR is the smoke test that proves the pipeline works end-to-end. The regex bug it caught (see below) is exactly the kind of thing a single-chapter PoC is supposed to surface.

Coverage decisions

Curated sparse: 6 of 8 lenses applied. Skipped:

• devotional — Genesis 1 is theological foundation; devotional response is more naturally drawn from later texts. Forcing it here would invite filler.
• mission — the missional thread of Scripture is rooted in Genesis 12 (Abrahamic call), not Genesis 1. The accuracy auditor would flag a missional Genesis-1 entry as fabrication.

The 6 included:

Lens	What it surfaces
grammatical	Sevenfold formula, opening/closing markers, bara bookending vv 1 and 27
literary	Forming/filling structure across days 1-3 paired with 4-6
typological	Spirit hovering, seventh-day rest, light/dark separation
redemptive	Image-of-God commission grounds the whole arc to Christ
christocentric	John 1:1-3, Col 1:16, divine speech as the same Word
canonical	Creation-by-speech echoes; Rev 21-22 recapitulation

Each entry has a panel_filter and panel_order selected to surface the existing chapter panels that that lens most illuminates, in the order that best supports the reading.

Quality gate results (local)

$ python3 _tools/schema_validator.py
RESULTS: 138138 passed, 0 failed, 19 warnings
$ python3 _tools/build_sqlite.py
[OK] hermeneutic_lenses: 8 lenses, 6 contents
$ python3 _tools/validate_sqlite.py
RESULTS: 101 passed, 0 failed, 2 warnings
$ python3 _tools/lens_quality_scorer.py --chapter gen1
LENS QUALITY REPORT — 6 entries
Passing (>= 90): 6
Failing (< 90): 0
Average: 100.0/100

All entries 257–274 chars (within the 80–280 schema range). All score 25/25 on every DVCR dimension.

Tier-2 accuracy audit

accuracy_meta_extractors.extract_hermeneutic_lenses() produces 6 claims from this content (one per lens entry), ready for accuracy_auditor.py --tier 2 to verify in CI. I can't run the tier-2 audit locally (no API key in the dev container — that's CI's job).

If any claim fails the tier-2 verifier, the auditor will flag it as a position the named hermeneutical school doesn't actually advocate, and the entry needs revision before merge.

The exegesis I think is most likely to draw an audit flag (transparency note): the typological claim that light/dark separation foreshadows Revelation 21 is defensible (Beale, The Temple and the Church's Mission; many patristic readings) but it's the closest to a stretch in this batch. The other two anchors in that entry — Spirit hovering → new creation, seventh-day rest → Hebrews 4 — are rock-solid. If the audit pushes back on the Rev 21 anchor, I'll trim it.

Scorer fix included

lens_quality_scorer.py had a regex bug that this content surfaced: VERSE_REF_RE was matching numbers after a colon, so Psalm 33:6 got parsed as a verse reference and flagged 33-6 as out-of-range for Genesis 1 (which has 31 verses).

Fixed by tightening VERSE_REF_RE to require an explicit v / vv prefix for in-chapter validation, and adding a separate looser ANY_VERSE_PATTERN for the Density "has any verse anchor?" check. Verified across 10 representative cases:

Input	Should match?	Did match?
v 5	yes	✓
vv 5-7	yes	✓
v. 31	yes	✓
(vv 26-28)	yes	✓
Psalm 33:6	no	✓
Heb 4:9-11	no	✓
Genesis 1:3	no	✓
Col 1:15	no	✓
verses 26 and 27	no	✓ (no v/vv prefix)
vv. 5, 8, 13	yes (first match)	✓

This regex fix is part of the same PR because the lens content immediately depends on it — no point shipping content that would fail the scorer for the wrong reason.

What this PR does NOT do

• The remaining 49 Genesis chapters (still tracked under [#820 / 2] Pilot batch — Genesis (50 chapters) #1781)
• The 8 genre batches ([#820 / 3] Pentateuch rest — Exodus, Leviticus, Numbers, Deuteronomy (137 ch) #1782 through [#820 / 10] Revelation (22 ch) #1789) — they were always going to wait for [#820 / 2] Pilot batch — Genesis (50 chapters) #1781 to ship clean
• Any tier-2 audit results (CI runs them)

Rollback plan

Pure additive content + one-line scorer regex tightening. git revert is clean — nothing references the new content row except the app's lens-toggle UI, which currently shows a guidance string and (per #1790) optionally filters/reorders panels. Reverting just makes Genesis 1's lens toggle show no guidance, same as the rest of the corpus.

Refs #820, #1781.

[github-actions]

Content Pipeline Results

✅ All pipeline checks passed

Step	Status	Details
Schema Validation	✅	138138 passed, 0 failed
Build DB	✅	—
DB Integrity	✅	101 passed, 0 failed

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3973	❌ 0	3973
Suites	✅ 531	❌ 0	531

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 95.8s

[CraigBuckmaster]

Genesis 1 — proof-of-concept for Epic #820 / #1781

Single-chapter slice of the Genesis pilot. Ships 6 lens entries for Genesis 1 to validate the entire Phase 0 + Phase 1 + quality-gate pipeline against real content before scaling.

The remaining 49 Genesis chapters stay under #1781 — this PR is the smoke test that proves the pipeline works end-to-end. The regex bug it caught (see below) is exactly the kind of thing a single-chapter PoC is supposed to surface.

Coverage decisions

Curated sparse: 6 of 8 lenses applied. Skipped:

• devotional — Genesis 1 is theological foundation; devotional response is more naturally drawn from later texts. Forcing it here would invite filler.
• mission — the missional thread of Scripture is rooted in Genesis 12 (Abrahamic call), not Genesis 1. The accuracy auditor would flag a missional Genesis-1 entry as fabrication.

The 6 included:

Lens	What it surfaces
grammatical	Sevenfold formula, opening/closing markers, bara bookending vv 1 and 27
literary	Forming/filling structure across days 1-3 paired with 4-6
typological	Spirit hovering, seventh-day rest, light/dark separation
redemptive	Image-of-God commission grounds the whole arc to Christ
christocentric	John 1:1-3, Col 1:16, divine speech as the same Word
canonical	Creation-by-speech echoes; Rev 21-22 recapitulation

Each entry has a panel_filter and panel_order selected to surface the existing chapter panels that that lens most illuminates, in the order that best supports the reading.

Quality gate results (local)

$ python3 _tools/schema_validator.py
RESULTS: 138138 passed, 0 failed, 19 warnings
$ python3 _tools/build_sqlite.py
[OK] hermeneutic_lenses: 8 lenses, 6 contents
$ python3 _tools/validate_sqlite.py
RESULTS: 101 passed, 0 failed, 2 warnings
$ python3 _tools/lens_quality_scorer.py --chapter gen1
LENS QUALITY REPORT — 6 entries
Passing (>= 90): 6
Failing (< 90): 0
Average: 100.0/100

All entries 257–274 chars (within the 80–280 schema range). All score 25/25 on every DVCR dimension.

Tier-2 accuracy audit

accuracy_meta_extractors.extract_hermeneutic_lenses() produces 6 claims from this content (one per lens entry), ready for accuracy_auditor.py --tier 2 to verify in CI. I can't run the tier-2 audit locally (no API key in the dev container — that's CI's job).

If any claim fails the tier-2 verifier, the auditor will flag it as a position the named hermeneutical school doesn't actually advocate, and the entry needs revision before merge.

The exegesis I think is most likely to draw an audit flag (transparency note): the typological claim that light/dark separation foreshadows Revelation 21 is defensible (Beale, The Temple and the Church's Mission; many patristic readings) but it's the closest to a stretch in this batch. The other two anchors in that entry — Spirit hovering → new creation, seventh-day rest → Hebrews 4 — are rock-solid. If the audit pushes back on the Rev 21 anchor, I'll trim it.

Scorer fix included

lens_quality_scorer.py had a regex bug that this content surfaced: VERSE_REF_RE was matching numbers after a colon, so Psalm 33:6 got parsed as a verse reference and flagged 33-6 as out-of-range for Genesis 1 (which has 31 verses).

Fixed by tightening VERSE_REF_RE to require an explicit v / vv prefix for in-chapter validation, and adding a separate looser ANY_VERSE_PATTERN for the Density "has any verse anchor?" check. Verified across 10 representative cases:

Input	Should match?	Did match?
v 5	yes	✓
vv 5-7	yes	✓
v. 31	yes	✓
(vv 26-28)	yes	✓
Psalm 33:6	no	✓
Heb 4:9-11	no	✓
Genesis 1:3	no	✓
Col 1:15	no	✓
verses 26 and 27	no	✓ (no v/vv prefix)
vv. 5, 8, 13	yes (first match)	✓

This regex fix is part of the same PR because the lens content immediately depends on it — no point shipping content that would fail the scorer for the wrong reason.

What this PR does NOT do

• The remaining 49 Genesis chapters (still tracked under [#820 / 2] Pilot batch — Genesis (50 chapters) #1781)
• The 8 genre batches ([#820 / 3] Pentateuch rest — Exodus, Leviticus, Numbers, Deuteronomy (137 ch) #1782 through [#820 / 10] Revelation (22 ch) #1789) — they were always going to wait for [#820 / 2] Pilot batch — Genesis (50 chapters) #1781 to ship clean
• Any tier-2 audit results (CI runs them)

Rollback plan

Pure additive content + one-line scorer regex tightening. git revert is clean — nothing references the new content row except the app's lens-toggle UI, which currently shows a guidance string and (per #1790) optionally filters/reorders panels. Reverting just makes Genesis 1's lens toggle show no guidance, same as the rest of the corpus.

Refs #820, #1781.

[github-actions]

Content Pipeline Results

✅ All pipeline checks passed

Step	Status	Details
Schema Validation	✅	138138 passed, 0 failed
Build DB	✅	—
DB Integrity	✅	101 passed, 0 failed

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3973	❌ 0	3973
Suites	✅ 531	❌ 0	531

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 95.8s

[CraigBuckmaster]

Genesis 1 — proof-of-concept for Epic #820 / #1781

Single-chapter slice of the Genesis pilot. Ships 6 lens entries for Genesis 1 to validate the entire Phase 0 + Phase 1 + quality-gate pipeline against real content before scaling.

The remaining 49 Genesis chapters stay under #1781 — this PR is the smoke test that proves the pipeline works end-to-end. The regex bug it caught (see below) is exactly the kind of thing a single-chapter PoC is supposed to surface.

Coverage decisions

Curated sparse: 6 of 8 lenses applied. Skipped:

• devotional — Genesis 1 is theological foundation; devotional response is more naturally drawn from later texts. Forcing it here would invite filler.
• mission — the missional thread of Scripture is rooted in Genesis 12 (Abrahamic call), not Genesis 1. The accuracy auditor would flag a missional Genesis-1 entry as fabrication.

The 6 included:

Lens	What it surfaces
grammatical	Sevenfold formula, opening/closing markers, bara bookending vv 1 and 27
literary	Forming/filling structure across days 1-3 paired with 4-6
typological	Spirit hovering, seventh-day rest, light/dark separation
redemptive	Image-of-God commission grounds the whole arc to Christ
christocentric	John 1:1-3, Col 1:16, divine speech as the same Word
canonical	Creation-by-speech echoes; Rev 21-22 recapitulation

Each entry has a panel_filter and panel_order selected to surface the existing chapter panels that that lens most illuminates, in the order that best supports the reading.

Quality gate results (local)

$ python3 _tools/schema_validator.py
RESULTS: 138138 passed, 0 failed, 19 warnings
$ python3 _tools/build_sqlite.py
[OK] hermeneutic_lenses: 8 lenses, 6 contents
$ python3 _tools/validate_sqlite.py
RESULTS: 101 passed, 0 failed, 2 warnings
$ python3 _tools/lens_quality_scorer.py --chapter gen1
LENS QUALITY REPORT — 6 entries
Passing (>= 90): 6
Failing (< 90): 0
Average: 100.0/100

All entries 257–274 chars (within the 80–280 schema range). All score 25/25 on every DVCR dimension.

Tier-2 accuracy audit

accuracy_meta_extractors.extract_hermeneutic_lenses() produces 6 claims from this content (one per lens entry), ready for accuracy_auditor.py --tier 2 to verify in CI. I can't run the tier-2 audit locally (no API key in the dev container — that's CI's job).

If any claim fails the tier-2 verifier, the auditor will flag it as a position the named hermeneutical school doesn't actually advocate, and the entry needs revision before merge.

The exegesis I think is most likely to draw an audit flag (transparency note): the typological claim that light/dark separation foreshadows Revelation 21 is defensible (Beale, The Temple and the Church's Mission; many patristic readings) but it's the closest to a stretch in this batch. The other two anchors in that entry — Spirit hovering → new creation, seventh-day rest → Hebrews 4 — are rock-solid. If the audit pushes back on the Rev 21 anchor, I'll trim it.

Scorer fix included

lens_quality_scorer.py had a regex bug that this content surfaced: VERSE_REF_RE was matching numbers after a colon, so Psalm 33:6 got parsed as a verse reference and flagged 33-6 as out-of-range for Genesis 1 (which has 31 verses).

Fixed by tightening VERSE_REF_RE to require an explicit v / vv prefix for in-chapter validation, and adding a separate looser ANY_VERSE_PATTERN for the Density "has any verse anchor?" check. Verified across 10 representative cases:

Input	Should match?	Did match?
v 5	yes	✓
vv 5-7	yes	✓
v. 31	yes	✓
(vv 26-28)	yes	✓
Psalm 33:6	no	✓
Heb 4:9-11	no	✓
Genesis 1:3	no	✓
Col 1:15	no	✓
verses 26 and 27	no	✓ (no v/vv prefix)
vv. 5, 8, 13	yes (first match)	✓

This regex fix is part of the same PR because the lens content immediately depends on it — no point shipping content that would fail the scorer for the wrong reason.

What this PR does NOT do

• The remaining 49 Genesis chapters (still tracked under [#820 / 2] Pilot batch — Genesis (50 chapters) #1781)
• The 8 genre batches ([#820 / 3] Pentateuch rest — Exodus, Leviticus, Numbers, Deuteronomy (137 ch) #1782 through [#820 / 10] Revelation (22 ch) #1789) — they were always going to wait for [#820 / 2] Pilot batch — Genesis (50 chapters) #1781 to ship clean
• Any tier-2 audit results (CI runs them)

Rollback plan

Pure additive content + one-line scorer regex tightening. git revert is clean — nothing references the new content row except the app's lens-toggle UI, which currently shows a guidance string and (per #1790) optionally filters/reorders panels. Reverting just makes Genesis 1's lens toggle show no guidance, same as the rest of the corpus.

Refs #820, #1781.

[github-actions]

Content Pipeline Results

✅ All pipeline checks passed

Step	Status	Details
Schema Validation	✅	138138 passed, 0 failed
Build DB	✅	—
DB Integrity	✅	101 passed, 0 failed

[github-actions]

Test Results

✅ All tests passed

	Passed	Failed	Total
Tests	✅ 3973	❌ 0	3973
Suites	✅ 531	❌ 0	531

Coverage

Statements	Branches	Functions	Lines
—	—	—	—

⏱️ Duration: 95.8s