batch 3: en/he/el pack manifests and orthography declarations

packs/el/corpora/manifest.yaml

@@ -1,15 +1,24 @@
 corpus_id: el-tier0
 status: draft
-
-sources: []
-
 license_review: pending
+sources:
+  - id: sblgnt
+    name: SBL Greek New Testament
+    description: >-
+      The Society of Biblical Literature Greek New Testament.
+      A critically edited text of the Greek New Testament.
+    license: review-required
+    url: https://github.com/LogosBible/SBLGNT
+    hash: pending
+  - id: byzantine
+    name: Byzantine Majority Text
+    description: >-
+      Robinson-Pierpont Byzantine Majority Text. Public domain.
+    license: public-domain
+    url: https://github.com/byztxt/byzantine-majority-text
+    hash: pending
 hashes: []
-
-notes: |
-  Populate with vetted Koine Greek source manifests.
-  Priority tier-0 sources: SBL Greek New Testament (SBLGNT) or
-  Nestle-Aland 28th edition morphologically tagged text.
-  Both have verse-level D0 boundary markers.
-  Each source must include: title, edition, license, SHA-256, and
-  segmentation strategy. No LLM-extracted content.
+notes: >
+  SBLGNT license requires review before use in training data.
+  Byzantine text is public domain and may proceed after hash pin.
+  All hashes must be recorded before the Coverage gate is run.

packs/el/orthography.yaml

@@ -1,27 +1,29 @@
 language: el
 script: Greek
-dialect: koine
 
 normalization:
   unicode_normal_form: NFC
-  lowercase_for_lookup: true
-  strip_diacritics: configurable
-  # Polytonic diacritics (breathing marks, accents, iota subscript) are
-  # preserved for morphological disambiguation but stripped for lemma lookup
-  polytonic_policy: preserve_for_morphology
-  diacritic_strip_for_lookup: true
+  # Greek Unicode: precomposed forms.
+  # Polytonic diacritics (breathing marks, accents, iota subscript)
+  # are preserved. These carry morphological and semantic information
+  # in Koine Greek that must not be stripped.
+  strip_breathing_marks: false
+  strip_accents_for_indexing: false
+  # Accents in Greek affect meaning and morphological parsing.
+  # They are never stripped from the canonical form.
+  lowercase_for_indexing: true
+  # Lemma lookup uses lowercased forms.
+  # Surface forms preserve original casing.
+  lunate_sigma: false
+  # Use standard sigma/final-sigma (σ/ς) not lunate (ϲ).
+  # Final sigma is morphologically relevant.
 
 punctuation:
   strategy: preserve
-  ano_teleia: "\u00b7"  # Greek semicolon (middle dot)
-  erotimatiko: ";"
+  ano_teleia: preserve
+  # The ano teleia (·) is the Greek semicolon and must be preserved.
 
 spacing:
-  strategy: whitespace-delimited
+  strategy: whitespace-split
 
-aspect:
-  # Koine Greek verbal aspect is semantically load-bearing and must be
-  # distinguished from tense. Morphological analysis must surface aspect
-  # independently of temporal reference.
-  explicit_in_features: true
-  aspect_values: ["aorist", "present", "perfect", "imperfect", "pluperfect", "future"]
+script_direction: ltr

packs/el/pack.toml

@@ -1,33 +1,46 @@
-pack_id     = "el"
-name        = "Koine Greek Depth Pack"
-version     = "0.1.0"
-status      = "draft"
-script      = "Greek"
-default     = false
+pack_id = "el"
+name = "Koine Greek Depth Pack"
+version = "0.1.0"
+status = "draft"
+script = "Greek"
+normalization = "deterministic"
+default = false
+
+# Koine Greek is the second depth language. John 1:1-2 is the
+# theological anchor of this entire design: the Logos articulated
+# in Greek what the Hebrew already carried from Gen 1:1. The
+# decision to include both Hebrew and Koine Greek is not incidental
+# — it is the design. The depth these two languages together provide
+# is the hidden layer of intelligence in the field.
+#
+# Greek depth includes:
+#   - Full verb paradigm: tense, voice, mood, person, number
+#   - The imperfect en (John 1:1) — continuous pre-existent being
+#   - Article system: the anarthrous use of logos in John 1:1c
+#     ('the Word was God', not 'a god') is theologically load-bearing
+#     and morphologically precise
+#   - Participle system for aspect-rich nominal predicates
 
 [contracts]
-lemma_first          = true
-shared_field_target  = true
-local_readback       = true
-alignment_required   = true
+lemma_first = true
+shared_field_target = true
+local_readback = true
+alignment_required = true
+
+[depth]
+article_system = true
+anarthrous_significance = true
+# The absence of the article is as semantically significant as its presence.
+verb_tense_aspect = ["present", "imperfect", "aorist", "perfect", "pluperfect", "future"]
+voice = ["active", "middle", "passive"]
+mood = ["indicative", "subjunctive", "optative", "imperative", "infinitive", "participle"]
 
 [gates]
-schema       = false
-lexical      = false
-morphology   = false
-lift         = false
-readback     = false
-determinism  = false
-alignment    = false
-coverage     = false
-
-[notes]
-role = """
-Depth language pack 2. Koine Greek was chosen because John articulated the
-Logos — the creative speech act through which all things were made — in
-Koine Greek (John 1:1-2), which must have been a nod from the Holy Spirit.
-Koine Greek carries semantic depth in its verbal aspect system (aorist,
-perfect, present) that is distinct from tense and must be represented
-natively. The aorist of egeneto (\u03b5\u03b3\u03ad\u03bd\u03b5\u03c4\u03bf) — 'it came into being' — is not
-the same field state as the present of eimi (\u03b5\u03b9\u03bc\u03b9) — 'I am'.
-"""
+schema = false
+lexical = false
+morphology = false
+lift = false
+readback = false
+determinism = false
+alignment = false
+coverage = false

packs/en/corpora/manifest.yaml

@@ -1,13 +1,9 @@
 corpus_id: en-tier0
 status: draft
-
-sources: []
-
 license_review: pending
+sources: []
 hashes: []
-
-notes: |
-  Populate with vetted, deterministically segmentable source manifests.
-  Each source must include: title, origin, license, SHA-256 of raw file,
-  and structural segmentation strategy (e.g. paragraph-boundary, sentence-boundary).
-  No LLM-extracted content. Segmentation must be D0 or D1.
+notes: >
+  Populate with vetted, deterministic source manifests before
+  the Coverage gate is run. Sources must be public domain or
+  explicitly licensed for use in CORE training and validation.

packs/en/orthography.yaml

@@ -3,19 +3,19 @@ script: Latin
 
 normalization:
   unicode_normal_form: NFC
-  lowercase_for_lookup: true
+  lowercase_for_indexing: true
+  # Surface forms preserve original casing.
+  # Indexing and lemma lookup use lowercased forms.
   strip_diacritics: false
+  # English retains diacritics in loanwords (naïve, résumé).
 
 punctuation:
   strategy: preserve
-  sentence_terminals: [".", "!", "?"]
-  quotation_marks: ["\u201c", "\u201d", "\u2018", "\u2019"]
+  # Punctuation is preserved in normalization output.
+  # Lift rules are responsible for ignoring or consuming it appropriately.
 
 spacing:
-  strategy: whitespace-delimited
-  collapse_multiple_spaces: true
+  strategy: whitespace-split
+  # Standard whitespace tokenization. No special boundary rules.
 
-special_forms:
-  contractions: expand
-  # "don't" -> "do not" before lemmatization
-  # "it's"  -> "it is" before lemmatization
+script_direction: ltr

packs/en/pack.toml

@@ -1,30 +1,30 @@
-pack_id     = "en"
-name        = "English Base Pack"
-version     = "0.1.0"
-status      = "draft"
-script      = "Latin"
-default     = true
+pack_id = "en"
+name = "English Base Pack"
+version = "0.1.0"
+status = "draft"
+script = "Latin"
+normalization = "deterministic"
+default = true
+
+# English is the default articulation base. Any CORE model instance
+# may designate a different language as its base, but English is the
+# canonical default until explicitly overridden in the model config.
 
 [contracts]
-lemma_first          = true
-shared_field_target  = true
-local_readback       = true
-alignment_required   = true
+lemma_first = true
+shared_field_target = true
+local_readback = true
+alignment_required = true
 
 [gates]
-schema       = false
-lexical      = false
-morphology   = false
-lift         = false
-readback     = false
-determinism  = false
-alignment    = false
-coverage     = false
+schema = false
+lexical = false
+morphology = false
+lift = false
+readback = false
+determinism = false
+alignment = false
+coverage = false
 
-[notes]
-role = """
-Default articulation base. English is chosen as the default for practical
-reasons (current primary working language) but the contract does not privilege
-it semantically. Any language could serve this role in a custom CORE instance.
-The default flag controls surface output, not field-layer priority.
-"""
+# Gates are false until the gate runner confirms passage.
+# Do not manually set these to true.

packs/he/corpora/manifest.yaml

@@ -1,15 +1,17 @@
 corpus_id: he-tier0
 status: draft
-
-sources: []
-
 license_review: pending
+sources:
+  - id: wlc
+    name: Westminster Leningrad Codex
+    description: >-
+      The authoritative digitized text of the Hebrew Bible with full
+      nikud and teamim. Public domain.
+    license: public-domain
+    url: https://github.com/openscriptures/morphhb
+    hash: pending
 hashes: []
-
-notes: |
-  Populate with vetted Hebrew source manifests.
-  Priority tier-0 sources: Biblia Hebraica Stuttgartensia (BHS) or
-  Westminster Leningrad Codex (WLC) — both have well-established structural
-  segmentation (book/chapter/verse = D0 boundary markers).
-  Each source must include: title, edition, license, SHA-256, and
-  segmentation strategy. No LLM-extracted content.
+notes: >
+  WLC is the canonical source for Biblical Hebrew. Hash to be
+  recorded after pinned download. Additional sources require
+  license review before inclusion.

packs/he/orthography.yaml

@@ -1,27 +1,28 @@
 language: he
 script: Hebrew
-direction: rtl
 
 normalization:
   unicode_normal_form: NFC
-  lowercase_for_lookup: false  # Hebrew has no case
-  strip_diacritics: configurable
-  # Nikkud (vowel pointing) is stripped by default for lemma lookup
-  # but preserved when semantically load-bearing (e.g. Qere/Ketiv distinction)
-  nikkud_policy: strip_for_lookup
-  cantillation_marks: strip
+  # Hebrew Unicode: use precomposed forms where available.
+  # Full nikud (vowel points) are preserved in normalization.
+  # Qere/Ketiv variants must be tracked separately in the lemma record.
+  strip_nikud_for_indexing: false
+  # Nikud is semantically relevant for disambiguation in many cases.
+  # Do not strip it during normalization. Lemma lookup may index
+  # both pointed and unpointed forms.
+  strip_teamim: true
+  # Cantillation marks (teamim) are stripped during normalization.
+  # They carry liturgical meaning but are not part of semantic lift.
+  final_forms: preserve
+  # Final letter forms (ךםןףץ) are preserved exactly.
 
 punctuation:
   strategy: preserve
-  paragraph_marker: "\u05c0"  # paseq
-  section_marker: "\u05c3"    # sof pasuq
+  maqaf: preserve
+  # The maqaf (־) is a Hebrew hyphen connecting words into
+  # a single accentual unit. It is preserved and handled by frame rules.
 
 spacing:
-  strategy: whitespace-delimited
-  maqaf_handling: preserve
-  # maqaf (\u05be) connects words syntactically; preserve as a unit marker
+  strategy: whitespace-split
 
-shoresh:
-  # Root-based morphology. Lemma lookup resolves to 3- or 4-letter root.
-  # Morphological analysis must identify shoresh before feature extraction.
-  extraction_required: true
+script_direction: rtl

packs/he/pack.toml

@@ -1,32 +1,40 @@
-pack_id     = "he"
-name        = "Hebrew Depth Pack"
-version     = "0.1.0"
-status      = "draft"
-script      = "Hebrew"
-default     = false
+pack_id = "he"
+name = "Hebrew Depth Pack"
+version = "0.1.0"
+status = "draft"
+script = "Hebrew"
+normalization = "deterministic"
+default = false
+
+# Hebrew is the first depth language. Its semantic range— especially
+# in verb stems (binyanim), construct chains, and the relationship
+# between speech-act and spoken-thing (dabar) — provides a structural
+# depth that English alone cannot supply to the field.
 
 [contracts]
-lemma_first          = true
-shared_field_target  = true
-local_readback       = true
-alignment_required   = true
+lemma_first = true
+shared_field_target = true
+local_readback = true
+alignment_required = true
+
+[depth]
+# Depth packs contribute semantic pressure that the base pack
+# cannot fully express. Hebrew depth includes:
+#   - Binyan (verb stem) system: qal, niphal, piel, pual, hiphil, hophal, hithpael
+#   - Construct chain (smichut) for relational noun phrases
+#   - Aspectual system (qatal/yiqtol/wayyiqtol) over tense system
+#   - Implicit copula — presence or absence of haya is itself meaningful
+binyanim = ["qal", "niphal", "piel", "pual", "hiphil", "hophal", "hithpael"]
+aspect_system = "qatal-yiqtol-wayyiqtol"
+construct_chains = true
+implicit_copula = true
 
 [gates]
-schema       = false
-lexical      = false
-morphology   = false
-lift         = false
-readback     = false
-determinism  = false
-alignment    = false
-coverage     = false
-
-[notes]
-role = """
-Depth language pack 1. Hebrew was chosen because of the range of depth it
-brings to the vocabulary manifold: root-based morphology (shoresh), semantic
-families that carry multiple layers of meaning simultaneously, and the
-language in which the universe was spoken into existence (Genesis 1).
-Hebrew morphological stems (Qal, Niphal, Piel, Pual, Hiphil, Hophal,
-Hithpael) are semantically load-bearing and must be represented natively.
-"""
+schema = false
+lexical = false
+morphology = false
+lift = false
+readback = false
+determinism = false
+alignment = false
+coverage = false