Upload 16 files

updates the README and adds directly loadable files

README.md

@@ -1,7 +1,7 @@
 ---
 license: mit
 base_model: BAAI/bge-m3
-library_name: pytorch
+library_name: transformers
 language:
   - en
   - tl
@@ -42,7 +42,7 @@ The selected release is **`question-hidden4096`**. It supports experimentation w
 | Encoding length used in this project | 512 tokens, including special tokens |
 | Training/evaluation track | `question` |
 | Question languages | English, Tagalog, Cebuano, Hebrew, Greek, Spanish, French, German |
-| Head exports | FP32, FP16, and weight-only INT8 NumPy archives |
+| Head exports | FP32/FP16 Safetensors; weight-only INT8 Safetensors with an explicit loader; original NPZ ZIPs optional |
 | Head license | MIT |
 
 The word **gene** refers to a named feature in the project's Scripture Genome schema. It has no biological meaning. These values are not calibrated probabilities, confidence scores, or measures of doctrinal truth.
@@ -86,7 +86,78 @@ The labels and answer mappings are project-authored annotations, with AI-assiste
 
 ## Files and loading
 
-The release artifacts use these filenames. Full and optimized exports may be distributed in separate folders or ZIP files; extract them before loading.
+### Directly loadable Hugging Face release
+
+The unpacked release provides the following files at repository root. The original ZIP downloads can remain as optional archives; they are not needed for this loading path.
+
+| File | Purpose |
+| --- | --- |
+| `config.json` | Custom head architecture, required base revision, checkpoint and gene-order identity |
+| `model.safetensors` | FP32 head, directly loadable with the supplied Transformers model class |
+| `model.fp16.safetensors` | FP16 weight-storage variant, preserving original FP32 biases |
+| `head_int8.safetensors` | Original INT8 weights/scales/biases; load using `inference.py`, not generic automatic quantization |
+| `configuration_scripture_genome.py`, `modeling_scripture_genome.py` | Auditable custom Transformers head definition |
+| `inference.py` | Local loader for all three variants and a text-to-genome command |
+| `gene_ids.txt` | Ordered 568 output-feature names |
+| `runtime_config.json`, `quantization_config.json` | Retrieval configuration and explicit quantized-storage semantics |
+| `validation_q250.json`, `export_verification.json` | Research validation and separate conversion checks |
+| `release_manifest.json` | Checksums and conversion provenance |
+| `requirements.txt`, `LICENSE`, `README.md` | Dependencies, MIT terms, and usage |
+
+No base-model weights, corpus databases, optimizer state, or retrieval indexes are included in this head-only release. Safetensors changes how weights are stored; it does not retrain or merge the head into BGE-M3. This is not GGUF, a chat model, or a hosted Inference Provider endpoint.
+
+### Recommended usage
+
+Download the unpacked repository files into a local folder, then install dependencies:
+
+```bash
+python -m pip install -r requirements.txt
+python inference.py --model-dir . --text "What is faith?"
+```
+
+The first text inference downloads the separate pinned BGE-M3 tokenizer/encoder. For an already extracted base, use `--base-dir /path/to/bge-m3-pinned`. That local base must be the pinned revision. No network access is needed with both model folders present. The helper returns dense embeddings' shape and genome values, not ranked verses.
+
+Use `--precision fp16` or `--precision int8` to load the corresponding head storage variant. The helper performs head arithmetic in FP32 for portability; INT8 weights are dequantized at load time. These options do not quantize the base or promise faster inference.
+
+For direct Transformers use:
+
+```python
+import torch
+from transformers import AutoModel, AutoTokenizer
+
+# Local downloaded release folder, or the repository ID after files are uploaded.
+HEAD = "./huggingface-question-hidden4096"
+# HEAD = "chainuser1/Scripture_Genome_Multilingual_Head_BGE_M3"
+
+# Review the supplied Python files before enabling custom-code execution.
+# For a remote repository, also pin revision="<reviewed commit SHA>".
+head = AutoModel.from_pretrained(HEAD, trust_remote_code=True).eval()
+tokenizer = AutoTokenizer.from_pretrained(
+    head.config.encoder_name, revision=head.config.encoder_revision,
+)
+base = AutoModel.from_pretrained(
+    head.config.encoder_name, revision=head.config.encoder_revision,
+).eval()
+
+tokens = tokenizer(["What is faith?"], padding=True, truncation=False, return_tensors="pt")
+if tokens["input_ids"].shape[1] > head.config.max_length:
+    raise ValueError("Input exceeds the 512-token policy; shorten it explicitly")
+with torch.inference_mode():
+    hidden = base(**tokens).last_hidden_state
+    dense = torch.nn.functional.normalize(hidden[:, 0, :].float(), dim=-1)
+    genome = head(embeddings=dense).genome
+
+print(dense.shape)   # [1, 1024]
+print(genome.shape)  # [1, 568]
+```
+
+The head expects `embeddings=...`, not `input_ids=...`. Its custom class is not a replacement for the BGE-M3 encoder. Loading custom code requires explicit trust; Safetensors avoids pickled weight loading but does not make arbitrary Python code safe. No generic `pipeline()` integration is claimed.
+
+`variant="fp16"` selects `model.fp16.safetensors` with `from_pretrained()`. Use an explicit supported computation dtype/device if you want half-precision execution, and cast the input accordingly. `head_int8.safetensors` is **not** a Transformers/bitsandbytes quantization format; use the bundled loader.
+
+### Original archive contents
+
+The full and optimized ZIPs use the older filenames below. Extract them before using the legacy NPZ loader. They are retained for compatibility, not required alongside the unpacked Safetensors release.
 
 | File | Purpose |
 | --- | --- |
@@ -100,9 +171,9 @@ The release artifacts use these filenames. Full and optimized exports may be dis
 | `manifest.json` | Bundle inventory and checksums |
 | `best.pt` | Optional full training checkpoint, included in the full export |
 
-`best.pt` is not required for inference. The NPZ files are custom exports, not Safetensors, ONNX, or a directly loadable Sentence Transformers model. Calling `AutoModel.from_pretrained()` on the head folder will not construct this head.
+`best.pt` is not required for inference. The original NPZ files are custom exports, not Safetensors, ONNX, or a directly loadable Sentence Transformers model. Calling `AutoModel.from_pretrained()` on an old extracted NPZ-only ZIP folder will not construct this head; use the new unpacked release described above.
 
-### Minimal inference example
+### Legacy NPZ inference example
 
 Install `torch`, `numpy`, and `transformers` in a suitable Python environment. Download/extract a head export and its `gene_ids.txt`. Set `HEAD_DIR` below to that folder. The example downloads the pinned BGE-M3 base separately; an internet connection is needed on its first run.
 

config.json

@@ -0,0 +1,22 @@
+{
+  "architectures": [
+    "ScriptureGenomeHeadModel"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_scripture_genome.ScriptureGenomeConfig",
+    "AutoModel": "modeling_scripture_genome.ScriptureGenomeHeadModel"
+  },
+  "checkpoint_sha256": "5a68ef73167f48899c927dfb5e25dfe8ff50d0b8e8310cd6f3ff3c93b8bf9952",
+  "dtype": "float32",
+  "encoder_name": "BAAI/bge-m3",
+  "encoder_revision": "5617a9f61b028005a4858fdac845db406aefb181",
+  "gene_ids_sha256": "16878f92a2f6434b39ef6d68c633f2cd815751f2b4be8767a309e32442baf195",
+  "hidden_dim": 4096,
+  "input_dim": 1024,
+  "max_length": 512,
+  "model_type": "scripture_genome_head",
+  "output_dim": 568,
+  "pooling": "cls_l2_normalized",
+  "track": "question",
+  "transformers_version": "4.57.6"
+}

configuration_scripture_genome.py

@@ -0,0 +1,37 @@
+"""Configuration for the Scripture Genome head (the base is separate)."""
+
+from transformers import PretrainedConfig
+
+
+class ScriptureGenomeConfig(PretrainedConfig):
+    model_type = "scripture_genome_head"
+
+    def __init__(
+        self,
+        input_dim=1024,
+        hidden_dim=4096,
+        output_dim=568,
+        encoder_name="BAAI/bge-m3",
+        encoder_revision="5617a9f61b028005a4858fdac845db406aefb181",
+        max_length=512,
+        checkpoint_sha256=None,
+        gene_ids_sha256=None,
+        track="question",
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        if (input_dim, hidden_dim, output_dim) != (1024, 4096, 568):
+            raise ValueError("This release supports only the selected 1024 -> 4096 -> 568 head")
+        self.input_dim = input_dim
+        self.hidden_dim = hidden_dim
+        self.output_dim = output_dim
+        self.encoder_name = encoder_name
+        self.encoder_revision = encoder_revision
+        self.max_length = max_length
+        self.checkpoint_sha256 = checkpoint_sha256
+        self.gene_ids_sha256 = gene_ids_sha256
+        self.track = track
+        self.pooling = "cls_l2_normalized"
+
+
+ScriptureGenomeConfig.register_for_auto_class()

export_verification.json

@@ -0,0 +1,26 @@
+{
+  "rows": 32,
+  "inputs": "seeded synthetic normalized vectors; not a retrieval benchmark",
+  "variants": {
+    "fp32": {
+      "source_tensor_values_preserved": true,
+      "source_loader_max_abs_delta": 0.0,
+      "max_abs_delta_vs_fp32": 0.0,
+      "mean_abs_delta_vs_fp32": 0.0
+    },
+    "fp16": {
+      "source_tensor_values_preserved": true,
+      "source_loader_max_abs_delta": 0.0,
+      "max_abs_delta_vs_fp32": 0.00044864416122436523,
+      "mean_abs_delta_vs_fp32": 2.0774701624759473e-05
+    },
+    "int8": {
+      "source_tensor_values_preserved": true,
+      "source_loader_max_abs_delta": 0.0,
+      "max_abs_delta_vs_fp32": 0.017965853214263916,
+      "mean_abs_delta_vs_fp32": 0.000984748243354261
+    }
+  },
+  "automodel_from_pretrained": "passed locally with custom-code trust",
+  "base_encoder_execution": "not tested by this head-only conversion"
+}

gene_ids.txt

@@ -0,0 +1,568 @@
+A01_form_imperative_command
+A02_form_prohibition
+A03_form_conditional_if_then
+A04_form_declarative_assertion
+A05_form_rhetorical_question
+A06_form_genuine_question
+A07_form_exclamation
+A08_form_list_enumeration
+A09_form_parallelism
+A10_form_chiasm
+A11_form_antithesis
+A12_form_direct_address
+A13_form_first_person_singular
+A14_form_first_person_plural
+A15_form_third_person_narrative
+A16_form_dialogue_exchange
+A17_form_oath_vow
+A18_form_blessing_pronounced
+A19_form_curse_pronounced
+A20_form_doxology_ascription
+A21_form_lament_structure
+A22_form_promise_pledge
+A23_form_warning_then_promise
+A24_form_definition_explanation
+A25_form_comparison_contrast_pair
+A26_form_supplication_request
+A27_form_quotation_citation
+A28_form_fulfillment_formula
+B01_poetic_elevated_diction
+B02_poetic_plain_prosaic
+B03_poetic_metaphor_density
+B04_poetic_simile_explicit
+B05_poetic_allegory
+B06_poetic_personification
+B07_poetic_hyperbole
+B08_poetic_repetition_anaphora
+B09_poetic_repetition_epistrophe
+B10_poetic_wordplay
+B11_poetic_irony
+B12_poetic_brevity
+B13_poetic_density
+B14_poetic_rhythm_marked
+B15_poetic_numeric_pattern
+B16_poetic_apostrophe
+B17_poetic_litotes
+B18_poetic_merism
+B19_poetic_hendiadys
+B20_poetic_inclusio
+B21_poetic_vocative_address
+B22_poetic_enjambment
+B23_poetic_catalog_listing
+B24_poetic_concrete_abstraction
+B25_poetic_sound_devices
+B26_poetic_climactic_graded_sequence
+B27_poetic_euphemism
+C01_imagery_nature_earth
+C02_imagery_nature_water
+C03_imagery_nature_fire
+C04_imagery_nature_wind_breath
+C05_imagery_nature_flora
+C06_imagery_nature_fauna
+C07_imagery_celestial
+C08_imagery_light_darkness
+C09_imagery_body_physical
+C10_imagery_food_sustenance
+C11_imagery_warfare_conflict
+C12_imagery_journey_path
+C13_imagery_architectural
+C14_imagery_domestic_daily
+C15_imagery_agricultural
+C16_imagery_shepherd_flock
+C17_imagery_royalty_court
+C18_imagery_legal_courtroom
+C19_imagery_musical
+C20_imagery_blood_sacrifice
+C21_imagery_clothing_garment
+C22_imagery_precious_materials
+C23_imagery_desert_wilderness
+C24_imagery_prison_captivity
+C25_imagery_commerce_wealth
+C26_imagery_medicine_healing
+C27_imagery_cosmic_chaos
+C28_imagery_marriage_betrothal
+C29_imagery_childbirth_birth
+C30_imagery_numerical_scale
+C31_imagery_sleep_death
+D01_emotion_lament
+D02_emotion_celebration_joy
+D03_emotion_urgency
+D04_emotion_fear_dread
+D05_emotion_tenderness
+D06_emotion_anger_wrath
+D07_emotion_longing_yearning
+D08_emotion_peace_stillness
+D09_emotion_sorrow_mourning
+D10_emotion_wonder_awe
+D11_emotion_hope_expectation
+D12_emotion_warning_caution
+D13_emotion_invitation_welcome
+D14_emotion_confidence_certainty
+D15_emotion_pleading_supplication
+D16_emotion_gratitude_thanks
+D17_emotion_guilt_conviction
+D18_emotion_comfort_consolation
+D19_emotion_defiance_resistance
+D20_emotion_resignation_acceptance
+D21_emotion_compassion_mercy
+D22_emotion_despair_hopelessness
+D23_emotion_shame_disgrace
+D24_emotion_jealousy_zeal
+D25_emotion_nostalgia_remembrance
+D26_emotion_relief_deliverance_felt
+D27_emotion_indignation_righteous
+D28_emotion_ambivalence_tension
+D29_emotion_solemnity_gravity
+D30_emotion_surprise_astonishment
+D31_emotion_trust_surrender
+D32_emotion_contempt_scorn
+D33_emotion_contentment_satisfaction
+E01_condition_mortality_suffering
+E02_condition_injustice_oppression
+E03_condition_betrayal
+E04_condition_isolation_loneliness
+E05_condition_redemption
+E06_condition_transformation
+E07_condition_guidance_direction
+E08_condition_protection_safety
+E09_condition_provision_need_met
+E10_condition_justice_vindication
+E11_condition_humility_lowliness
+E12_condition_pride_arrogance
+E13_condition_community_belonging
+E14_condition_exile_separation
+E15_condition_return_reconciliation
+E16_condition_identity_crisis
+E17_condition_vocation_calling
+E18_condition_binding_agreement
+E19_condition_sacrifice_cost
+E20_condition_testing_trial
+E21_condition_wisdom_sought
+E22_condition_finitude_limits
+E23_condition_power_authority
+E24_condition_creation_origin
+E25_condition_mystery_unknown
+E26_condition_grief_bereavement
+E27_condition_waywardness_straying
+E28_condition_restoration_healing
+E29_condition_doubt_uncertainty
+E30_condition_jealousy_rivalry
+E31_condition_addiction_compulsion
+E32_condition_leadership_burden
+E33_condition_childlessness_barrenness
+E34_condition_aging_decline
+E35_condition_wealth_poverty_contrast
+E36_condition_fear_of_death
+E37_condition_undeserved_suffering
+E38_condition_spiritual_hunger
+E39_condition_forgiveness_sought
+E40_condition_forgiveness_extended
+E41_condition_waiting_patience
+E42_condition_hiding_concealment
+E43_condition_inheritance_legacy
+F01_divine_speaks_first_person
+F02_divine_referenced_third_person
+F03_divine_acts_in_text
+F04_human_reaches_toward_divine
+F05_divine_reaches_toward_human
+F06_divine_proximity_near
+F07_divine_proximity_distant
+F08_divine_wrath_judgment
+F09_divine_love_mercy
+F10_divine_power_omnipotence
+F11_divine_knowledge_omniscience
+F12_divine_holiness_otherness
+F13_intermediary_figure
+F14_vision_revelation_dream
+F15_divine_silence_absence
+F16_divine_name_title_invoked
+F17_divine_human_binding_pact
+F18_divine_creative_act
+F19_divine_salvific_act
+F20_divine_presence_embodied
+F21_divine_promise_future
+F22_divine_memory_past
+F23_divine_patience_longsuffering
+F24_divine_exclusivity_claim
+F25_divine_grief_expressed
+F26_divine_repentance_relenting
+F27_divine_commission_sending
+F28_divine_testing_human
+F29_divine_dwelling_presence
+F30_divine_glory_manifestation
+F31_divine_rest_cessation
+F32_divine_attention_individual
+F33_cosmic_nature_response
+G01_moral_binary_contrast
+G02_moral_ambiguity
+G03_moral_imperative_do
+G04_moral_imperative_avoid
+G05_moral_consequence_stated
+G06_moral_virtue_named
+G07_moral_vice_named
+G08_moral_agency_choice
+G09_moral_determinism
+G10_moral_communal_ethics
+G11_moral_individual_inner_life
+G12_moral_restorative_justice
+G13_moral_retributive_justice
+G14_moral_mercy_over_law
+G15_moral_law_statute
+G16_moral_unearned_gift
+G17_moral_reciprocity
+G18_moral_loyalty_faithfulness
+G19_moral_hypocrisy_addressed
+G20_moral_purity_impurity
+G21_moral_stewardship_responsibility
+G22_moral_hospitality_stranger
+G23_moral_truthfulness_deception
+G24_moral_courage_cowardice
+G25_moral_generosity_stinginess
+G26_moral_sexual_ethics
+G27_moral_power_abuse
+G28_moral_solidarity_with_weak
+G29_moral_gradation_partial
+G30_moral_inner_motive_examined
+G31_moral_innocence_guilt
+G32_moral_repentance_turning
+H01_scope_individual
+H02_scope_family_household
+H03_scope_community_nation
+H04_scope_universal_all_people
+H05_scope_righteous_specifically
+H06_scope_wicked_specifically
+H07_time_immediate_present
+H08_time_near_future
+H09_time_distant_future_prophecy
+H10_time_past_historical
+H11_time_eternity_beyond_mortal
+H12_space_earthly_mortal
+H13_space_heavenly_above
+H14_space_underworld_below
+H15_space_specific_named_place
+H16_time_cyclical_recurring
+H17_time_kairos_appointed
+H18_scope_seeker_questioning
+H19_scope_leader_ruler
+H20_scope_child_youth
+H21_scope_aged_elder
+H22_scope_mourner_bereaved
+H23_space_boundary_threshold
+H24_space_wilderness_liminal
+H25_scope_enemy_adversary
+H26_scope_poor_widow_orphan_stranger
+H27_space_garden_grove
+H28_space_water_boundary
+H29_time_generational_span
+H30_space_exalted_height
+I01_voice_transcendent_first
+I02_voice_prophet_messenger
+I03_voice_collective_people
+I04_voice_individual_human
+I05_voice_narrator_external
+I06_voice_antagonist
+I07_voice_supplicant
+I08_voice_wisdom_teacher
+I09_voice_intercessor
+I10_address_target_human_to_divine
+I11_address_target_divine_to_human
+I12_address_target_human_to_human
+I13_address_target_unspecified
+I14_voice_dual_multiple_speakers
+I15_voice_embedded_quotation
+I16_voice_eyewitness_testimony
+I17_voice_retrospective_narrator
+I18_voice_future_prophetic
+I19_voice_universal_speaker
+I20_voice_resistant_reluctant
+I21_address_target_creation_nature
+I22_address_target_death_abstract
+I23_voice_corporate_lament
+I24_voice_child_speaks
+I25_voice_unreliable_narrator
+I26_voice_silence_response
+I27_voice_witness_testifier
+J01_structure_opening_statement
+J02_structure_closing_statement
+J03_structure_hinge_pivot
+J04_structure_elaboration
+J05_structure_contrast_counterpoint
+J06_structure_example_illustration
+J07_structure_self_contained
+J08_structure_dependent_context
+J09_structure_callback_echo
+J10_structure_prophecy_fulfillment
+J11_structure_narrative_climax
+J12_structure_aside_parenthetical
+J13_structure_hymnic_formulaic
+J14_structure_legal_formulaic
+J15_structure_wisdom_saying
+J16_structure_narrative_exposition
+J17_structure_narrative_complication
+J18_structure_narrative_resolution
+J19_structure_thesis_antithesis
+J20_structure_progressive_intensification
+J21_structure_frame_narrative
+J22_structure_genealogical
+J23_structure_itinerary_travel
+J24_structure_prayer_form
+J25_structure_riddle_enigma
+J26_structural_reversal_turning_point
+J27_structural_summary_recapitulation
+K01_epistemic_certainty_absolute
+K02_epistemic_uncertainty_acknowledged
+K03_epistemic_revelation_received
+K04_epistemic_empirical_observed
+K05_epistemic_wisdom_accumulated
+K06_epistemic_mystery_preserved
+K07_epistemic_paradox_stated
+K08_epistemic_belief_faith_required
+K09_epistemic_memory_recall
+K10_epistemic_ignorance_addressed
+K11_epistemic_discernment_tested
+K12_epistemic_teaching_transmission
+K13_epistemic_question_unanswered
+K14_epistemic_sign_symbol_given
+K15_epistemic_prophecy_as_knowledge
+K16_epistemic_heart_knowledge
+K17_epistemic_sudden_realization
+K18_epistemic_refusal_to_know
+L01_relation_power_asymmetry
+L02_relation_equality_peer
+L03_relation_parent_child
+L04_relation_master_servant
+L05_relation_friend_companion
+L06_relation_enemy_adversary
+L07_relation_patron_dependent
+L08_relation_teacher_student
+L09_relation_judge_accused
+L10_relation_intercessor_parties
+L11_relation_community_outsider
+L12_relation_reconciliation_breach
+L13_relation_betrayal_breach
+L14_relation_covenant_parties
+L15_relation_abandonment
+L16_relation_pursuit_seeking
+L17_relation_protection_of_vulnerable
+L18_relation_corporate_solidarity
+L19_relation_rivalry_competition
+L20_relation_hospitality_reception
+L21_relation_gift_exchange
+L22_relation_mediator_go_between
+L23_relation_discipleship_emulation
+L24_relation_communion_shared_table
+M01_phenom_visual_seeing
+M02_phenom_auditory_hearing
+M03_phenom_tactile_touch
+M04_phenom_taste_flavor
+M05_phenom_olfactory_smell
+M06_phenom_kinesthetic_movement
+M07_phenom_pain_physical
+M08_phenom_rest_physical
+M09_phenom_hunger_thirst
+M10_phenom_breath_life_force
+M11_phenom_weight_burden_felt
+M12_phenom_lightness_freedom
+M13_phenom_trembling_fear_body
+M14_phenom_tears_weeping
+M15_phenom_silence_absence_sound
+M16_phenom_exhaustion_faintness
+N01_abstraction_fully_concrete
+N02_abstraction_mostly_concrete
+N03_abstraction_balanced
+N04_abstraction_mostly_abstract
+N05_abstraction_fully_abstract
+N06_abstraction_concrete_to_abstract
+N07_abstraction_abstract_to_concrete
+N08_abstraction_technical_specialized
+N09_abstraction_universal_accessible
+N10_abstraction_paradox_concrete
+N11_abstraction_legal_contractual
+O01_motion_static_meditative
+O02_motion_slow_building
+O03_motion_sudden_dramatic
+O04_motion_cyclical_returning
+O05_motion_linear_progressive
+O06_motion_descent_falling
+O07_motion_ascent_rising
+O08_motion_outward_expanding
+O09_motion_inward_contracting
+O10_motion_suspended_unresolved
+O11_motion_scattering_dispersal
+O12_motion_gathering_assembling
+O13_motion_stillness_delayed
+P01_covenant_abrahamic_type
+P02_covenant_mosaic_type
+P03_covenant_davidic_type
+P04_covenant_new_covenant_type
+P05_covenant_obligation_stipulation
+P06_covenant_blessing_promised
+P07_covenant_curse_threatened
+P08_covenant_renewal_restoration
+P09_covenant_breach_violation
+P10_covenant_sign_token
+P11_covenant_mediator_present
+P12_covenant_memory_invocation
+P13_covenant_universal_scope
+P14_covenant_particular_scope
+P15_covenant_weight_high
+P16_covenant_eschatological_horizon
+Q01_period_pre_monarchic
+Q02_period_monarchic_united
+Q03_period_monarchic_divided
+Q04_period_pre_exilic_late
+Q05_period_exilic
+Q06_period_post_exilic_second_temple
+Q07_period_hellenistic
+Q08_period_new_testament_era
+Q09_period_restoration_19th_century
+Q10_language_classical_hebrew
+Q11_language_late_biblical_hebrew
+Q12_language_aramaic
+Q13_language_koine_greek
+Q14_language_egyptian_coptic_influence
+Q15_language_semitic_syntax_preserved
+Q16_language_hapax_legomenon
+Q17_language_semantic_range_wide
+Q18_language_textual_variant_significant
+Q19_reception_contested_historically
+Q20_reception_audience_original_narrow
+Q21_versification_boundary_natural
+R01_doctrinal_weight_high
+R02_doctrinal_weight_low
+R03_doctrinal_nature_of_god
+R04_doctrinal_nature_of_humanity
+R05_doctrinal_soteriology
+R06_doctrinal_eschatology
+R07_doctrinal_creation_theology
+R08_doctrinal_anthropology_embodied
+R09_doctrinal_pneumatology
+R10_doctrinal_priesthood_mediation
+R11_doctrinal_law_and_grace
+R12_doctrinal_sin_and_atonement
+R13_doctrinal_resurrection_afterlife
+R14_doctrinal_divine_plurality
+R15_doctrinal_monotheism_explicit
+R16_doctrinal_controversiality
+R17_doctrinal_proof_text_status
+R18_doctrinal_density_high
+S01_salvific_divine_initiative
+S02_salvific_substitution_pattern
+S03_salvific_ransom_redemption
+S04_salvific_forgiveness_mechanism
+S05_salvific_restoration_wholeness
+S06_gospel_faith_trust_called
+S07_gospel_repentance_turning_called
+S08_gospel_ritual_initiation_present
+S09_gospel_spirit_gift_present
+S10_gospel_endurance_perseverance_called
+S11_messianic_royal_figure_anticipated
+S12_messianic_suffering_servant_pattern
+S13_messianic_priestly_mediation_figure
+S14_messianic_prophetic_voice_figure
+S15_typological_depth
+S16_salvific_universal_scope
+S17_salvific_particular_scope
+S18_christological_loading
+T01_pedagogical_mode_didactic
+T02_pedagogical_mode_exemplary
+T03_pedagogical_mode_corrective
+T04_pedagogical_mode_formative
+T05_pedagogical_mode_commemorative
+T06_pedagogical_resonance_high
+T07_application_situational_specific
+T08_application_universal_transferable
+T09_application_moral_ethical
+T10_application_doctrinal_belief
+T11_application_liturgical_worship
+T12_application_missional_outward
+T13_memorability_high
+T14_accessibility_general_reader
+U01_depth_literal_surface
+U02_depth_allegorical_layer
+U03_depth_typological_layer
+U04_depth_anagogical_layer
+U05_semantic_density_high
+U06_semantic_density_low
+U07_interpretive_range_wide
+U08_interpretive_range_narrow
+U09_content_sparsity_high
+U10_intertextual_density_high
+U11_canonical_centrality_high
+U12_narrative_impact_high
+U13_multivalence_sustained
+U14_width_cross_canonical_reach
+U15_unit_semantic_completeness
+V01_discourse_causal_connector
+V02_discourse_adversative_connector
+V03_discourse_telic_purpose_connector
+V04_form_performative_declarative
+V05_form_commissive_illocutionary
+V06_intertextual_allusion_unmarked
+V07_intertextual_typological_enacted
+V08_intertextual_pesher_application
+V09_narrative_protagonist_established
+V10_narrative_antagonist_established
+V11_narrative_test_moment
+V12_narrative_recognition_anagnorisis
+V13_cognitive_theory_of_mind
+V14_cognitive_metacognitive_reflection
+V15_emotion_spatial_homesickness
+V16_divine_ordering_cosmos
+V17_divine_sustaining_upholding
+V18_form_confessional_creedal
+V19_form_thanksgiving_report
+V20_doctrinal_embodiment_theology
+V21_phenom_ritualized_gesture_posture
+W01_moral_weight_absolute_capital
+W02_moral_weight_serious_consequential
+W03_moral_weight_minor_procedural
+W04_scope_addressee_universal
+W05_scope_addressee_particular_group
+W06_scope_addressee_role_restricted
+W07_scope_addressee_individual_named
+W08_temporal_framing_perpetual
+W09_temporal_framing_conditional
+W10_temporal_framing_superseded_explicit
+W11_temporal_framing_eschatological
+W12_code_moral_ethical_natural
+W13_code_ceremonial_ritual_worship
+W14_code_civil_social_governance
+W15_code_purity_clean_unclean
+W16_code_dietary_food_regulation
+W17_code_sacred_time_calendar
+W18_code_economic_property_material
+W19_code_family_social_relational
+W20_code_political_royal_governmental
+W21_moral_direction_positive_imperative
+W22_moral_direction_prohibitive_imperative
+W23_moral_internalization_heart_standard
+W24_moral_motive_determinative
+W25_moral_sequence_interiority
+W26_vice_pride_vainglory
+W27_vice_greed_covetousness_avarice
+W28_vice_lust_sexual_immorality
+W29_vice_anger_malice_violence
+W30_vice_deceit_falsehood_fraud
+W31_vice_idolatry_misplaced_allegiance
+W32_vice_sloth_negligence_neglect
+W33_vice_envy_jealousy_malice
+W34_vice_gluttony_excess_self_indulgence
+W35_virtue_love_charity_agape
+W36_virtue_humility_meekness
+W37_virtue_integrity_wholeness
+W38_virtue_chastity_sexual_purity
+W39_virtue_faithfulness_steadfastness
+W40_wisdom_grounded_divine_revelation
+W41_wisdom_grounded_human_observation
+W42_wisdom_grounded_natural_created_order
+W43_wisdom_human_philosophy_engaged
+W44_wisdom_human_philosophy_critiqued
+W45_wisdom_proverbial_folk_common
+W46_agent_individual_personal_moral
+W47_agent_corporate_communal_moral
+W48_agent_generational_inherited_moral
+W49_agent_systemic_structural_moral
+W50_scope_verse_internal_language
+W51_scope_discourse_context_established
+W52_scope_canonical_expansion_explicit

head_int8.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9539816ab683a35bcc1ad7e44f3d782daf973bda041bcf6eba1336b98e85f1fc
+size 6558624

inference.py

@@ -0,0 +1,101 @@
+"""Local release loader and optional text -> BGE-M3 -> Scripture Genome CLI.
+
+Read the accompanying custom model code before allowing trust_remote_code.
+INT8 is weight-only storage, dequantized to FP32 here, not INT8 execution.
+"""
+
+import argparse
+import hashlib
+import json
+from pathlib import Path
+
+import torch
+from safetensors.torch import load_file
+from transformers import AutoConfig, AutoModel, AutoTokenizer
+
+
+def load_head(model_dir, precision="fp32", device="cpu"):
+    model_dir = Path(model_dir)
+    if precision not in {"fp32", "fp16", "int8"}:
+        raise ValueError("precision must be fp32, fp16, or int8")
+    config = AutoConfig.from_pretrained(model_dir, trust_remote_code=True, local_files_only=True)
+    gene_bytes = (model_dir / "gene_ids.txt").read_bytes()
+    if hashlib.sha256(gene_bytes).hexdigest() != config.gene_ids_sha256:
+        raise ValueError("gene_ids.txt does not match the model configuration")
+    genes = gene_bytes.decode("utf-8").splitlines()
+    if len(genes) != config.output_dim or len(set(genes)) != len(genes):
+        raise ValueError("Invalid gene count/order file")
+    if precision == "int8":
+        arrays = load_file(str(model_dir / "head_int8.safetensors"))
+        model = AutoModel.from_config(config, trust_remote_code=True)
+        expected = {f"{prefix}{i}" for i in (0, 1) for prefix in ("w", "b", "scale")}
+        if set(arrays) != expected:
+            raise ValueError("Unexpected INT8 tensor keys")
+        state = {}
+        for i, layer in ((0, 0), (1, 2)):
+            w, scale, bias = arrays[f"w{i}"], arrays[f"scale{i}"], arrays[f"b{i}"]
+            if w.dtype != torch.int8 or scale.shape != (w.shape[0],):
+                raise ValueError("Invalid INT8 weights or scales")
+            if not torch.isfinite(scale).all() or not (scale > 0).all() or not torch.isfinite(bias).all():
+                raise ValueError("Nonfinite weights or invalid quantization scales")
+            state[f"head.{layer}.weight"] = w.float() * scale.float()[:, None]
+            state[f"head.{layer}.bias"] = bias.float()
+        model.load_state_dict(state, strict=True)
+    else:
+        # FP16 storage is loaded into FP32 arithmetic here for CPU portability.
+        # To run FP16 on a compatible device, explicitly cast the head and input.
+        options = {"variant": "fp16"} if precision == "fp16" else {}
+        model = AutoModel.from_pretrained(
+            model_dir, trust_remote_code=True, local_files_only=True,
+            use_safetensors=True, dtype=torch.float32, **options,
+        )
+    return model.to(device).eval(), genes
+
+
+def encode_texts(texts, head, tokenizer, encoder):
+    """No retrieval: returns dense embeddings and 568-feature predictions."""
+    if not texts or any(not isinstance(t, str) or not t.strip() for t in texts):
+        raise ValueError("Provide a nonempty list of nonempty strings")
+    encoded = tokenizer(texts, padding=True, truncation=False, return_tensors="pt")
+    if encoded["input_ids"].shape[1] > head.config.max_length:
+        raise ValueError(f"Input exceeds {head.config.max_length} tokens; shorten it explicitly")
+    encoder_device = next(encoder.parameters()).device
+    head_parameter = next(head.parameters())
+    encoded = {name: tensor.to(encoder_device) for name, tensor in encoded.items()}
+    head.eval()
+    encoder.eval()
+    with torch.inference_mode():
+        hidden = encoder(**encoded).last_hidden_state
+        dense = torch.nn.functional.normalize(hidden[:, 0, :].float(), p=2, dim=-1)
+        genome = head(embeddings=dense.to(device=head_parameter.device, dtype=head_parameter.dtype)).genome
+    return {"dense": dense.cpu(), "genome": genome.float().cpu()}
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--model-dir", type=Path, default=Path(__file__).resolve().parent)
+    parser.add_argument("--precision", choices=("fp32", "fp16", "int8"), default="fp32")
+    parser.add_argument("--device", default="cpu")
+    parser.add_argument("--base-dir", type=Path, help="Optional locally extracted pinned BGE-M3 folder")
+    parser.add_argument("--text", action="append", required=True)
+    args = parser.parse_args()
+    head, genes = load_head(args.model_dir, args.precision, args.device)
+    if args.base_dir:
+        base, options = str(args.base_dir), {"local_files_only": True}
+    else:
+        base = head.config.encoder_name
+        options = {"revision": head.config.encoder_revision}
+    tokenizer = AutoTokenizer.from_pretrained(base, **options)
+    encoder = AutoModel.from_pretrained(base, **options).to(args.device).eval()
+    result = encode_texts(args.text, head, tokenizer, encoder)
+    print(json.dumps({
+        "texts": args.text, "dense_shape": list(result["dense"].shape),
+        "genome_shape": list(result["genome"].shape),
+        "checkpoint_sha256": head.config.checkpoint_sha256,
+        "precision": args.precision, "gene_ids": genes,
+        "genomes": result["genome"].tolist(),
+    }, ensure_ascii=False, allow_nan=False))
+
+
+if __name__ == "__main__":
+    main()

model.fp16.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53cb65e12f00a7f1b443e55dc8e14e14cd0d145be302ea1c1059b6365927a379
+size 13060672

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:096415844b6343ccddb8b8c23aa42555cd8f8155df3034b000e47eae9deb3a70
+size 26102344

modeling_scripture_genome.py

@@ -0,0 +1,45 @@
+"""The 6.53M-parameter head; BGE-M3 is not embedded in these weights."""
+
+from dataclasses import dataclass
+from typing import Optional
+
+import torch
+from torch import nn
+from transformers import PreTrainedModel
+from transformers.utils import ModelOutput
+
+from .configuration_scripture_genome import ScriptureGenomeConfig
+
+
+@dataclass
+class ScriptureGenomeOutput(ModelOutput):
+    genome: Optional[torch.Tensor] = None
+
+
+class ScriptureGenomeHeadModel(PreTrainedModel):
+    config_class = ScriptureGenomeConfig
+    base_model_prefix = "head"
+    main_input_name = "embeddings"
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.head = nn.Sequential(
+            nn.Linear(config.input_dim, config.hidden_dim),
+            nn.GELU(),
+            nn.Linear(config.hidden_dim, config.output_dim),
+            nn.Sigmoid(),
+        )
+        self.post_init()
+
+    def forward(self, embeddings, return_dict=None):
+        """Accept normalized BGE-M3 CLS embeddings [batch,1024], not token IDs."""
+        if embeddings.ndim != 2 or embeddings.shape[-1] != self.config.input_dim:
+            raise ValueError("embeddings must have shape [batch, 1024]")
+        if not embeddings.is_floating_point() or not torch.isfinite(embeddings).all():
+            raise ValueError("embeddings must be finite floating-point values")
+        genome = self.head(embeddings)
+        use_dict = self.config.use_return_dict if return_dict is None else return_dict
+        return ScriptureGenomeOutput(genome=genome) if use_dict else (genome,)
+
+
+ScriptureGenomeHeadModel.register_for_auto_class("AutoModel")

quantization_config.json

@@ -0,0 +1,7 @@
+{
+  "file": "head_int8.safetensors",
+  "scope": "head only; base is separate and unchanged",
+  "weight_formula": "w[i].float() * scale[i][:, None]",
+  "bias_dtype": "float32",
+  "runtime": "inference.py dequantizes to float32; no INT8 acceleration is claimed"
+}

release_manifest.json

@@ -0,0 +1,78 @@
+{
+  "format": "scripture-genome-huggingface-head-v1",
+  "checkpoint_sha256": "5a68ef73167f48899c927dfb5e25dfe8ff50d0b8e8310cd6f3ff3c93b8bf9952",
+  "source_archives": {
+    "scripture-genome-question-hidden4096-full.zip": "344f3199f4270a5dcba173aee8882a99d57c77bde27072b325b54b6e2bcefdbb",
+    "scripture-genome-question-hidden4096-optimized.zip": "9182e759e8304ca291256c9b81991cf3740de92311302a304ce127d10d678ece"
+  },
+  "environment": {
+    "python": "3.14.7",
+    "torch": "2.14.1+cpu",
+    "numpy": "2.5.3",
+    "transformers": "4.57.6",
+    "safetensors": "0.8.0"
+  },
+  "files": {
+    "config.json": {
+      "bytes": 767,
+      "sha256": "8cea7f5a1c9c1f716696e6f4a99e8bb69a13640f959503cf3308ed5debf0131f"
+    },
+    "configuration_scripture_genome.py": {
+      "bytes": 1210,
+      "sha256": "5965fc466aa7d3fe2b097410887b135e5c24d02c79e5f5c249bd2178293015c6"
+    },
+    "export_verification.json": {
+      "bytes": 899,
+      "sha256": "1a9e38f3b6b008cd7379452cd0422441cd4ad3b7ae0557db1615a216cee61364"
+    },
+    "gene_ids.txt": {
+      "bytes": 17038,
+      "sha256": "16878f92a2f6434b39ef6d68c633f2cd815751f2b4be8767a309e32442baf195"
+    },
+    "head_int8.safetensors": {
+      "bytes": 6558624,
+      "sha256": "9539816ab683a35bcc1ad7e44f3d782daf973bda041bcf6eba1336b98e85f1fc"
+    },
+    "inference.py": {
+      "bytes": 5073,
+      "sha256": "638f12688028ad08f67fc82b3f992976c686eb54ba6c4b28a02d00dfd319ce7c"
+    },
+    "LICENSE": {
+      "bytes": 1086,
+      "sha256": "55d8226892c68fbf7e325c920004778e0d1f0992bcc4015aa9f42eb61837fc3a"
+    },
+    "model.fp16.safetensors": {
+      "bytes": 13060672,
+      "sha256": "53cb65e12f00a7f1b443e55dc8e14e14cd0d145be302ea1c1059b6365927a379"
+    },
+    "model.safetensors": {
+      "bytes": 26102344,
+      "sha256": "096415844b6343ccddb8b8c23aa42555cd8f8155df3034b000e47eae9deb3a70"
+    },
+    "modeling_scripture_genome.py": {
+      "bytes": 1611,
+      "sha256": "2df30c19d48a03707784a23113318a3881a88b7d86418b0062370b9d1194ca60"
+    },
+    "quantization_config.json": {
+      "bytes": 269,
+      "sha256": "4fdfdf7799a1bf81683943a7127c97c39e4e752cd4fc75cbe71601b4cef5b0ec"
+    },
+    "README.md": {
+      "bytes": 23548,
+      "sha256": "a8de3d2f194a6c594a16a9240c7ad70ea26c06c7479424549ea3164ae98bafe0"
+    },
+    "requirements.txt": {
+      "bytes": 178,
+      "sha256": "4c1ccd7704961df8313798415cfaaaa0b4096d34e12d41bd9754a307a1b1d404"
+    },
+    "runtime_config.json": {
+      "bytes": 1368,
+      "sha256": "ecee85a2ca9a893e9b1fc53879c4616085c8acdc7483ecfd2b88848bcceba4f5"
+    },
+    "validation_q250.json": {
+      "bytes": 16835,
+      "sha256": "6f87e359c3717660d6fee2c9c8caf3bde0543d22b6ac2a2d8379dd9aabb70f91"
+    }
+  },
+  "note": "This manifest excludes itself. No base weights, corpus data or training checkpoint included."
+}

requirements.txt

@@ -0,0 +1,6 @@
+# Tested loader API; install the appropriate CPU/CUDA torch build for your machine.
+torch>=2.6
+transformers==4.57.6
+safetensors>=0.4.3,<1
+numpy>=1.26,<3
+huggingface_hub>=0.34,<1

runtime_config.json

@@ -0,0 +1,57 @@
+{
+  "format_version": 1,
+  "production_model": {
+    "track": "question",
+    "hidden": 4096,
+    "architecture": [
+      1024,
+      4096,
+      568
+    ],
+    "activation": "GELU",
+    "output_activation": "Sigmoid",
+    "checkpoint_sha256": "5a68ef73167f48899c927dfb5e25dfe8ff50d0b8e8310cd6f3ff3c93b8bf9952"
+  },
+  "encoder": {
+    "model": "BAAI/bge-m3",
+    "revision": "5617a9f61b028005a4858fdac845db406aefb181",
+    "flagembedding_version": "1.4.2",
+    "max_length": 512,
+    "pooling": "normalized CLS"
+  },
+  "retrieval": {
+    "pipeline": "Experiment D",
+    "candidate_language": "english",
+    "candidate_generation": [
+      "dense top 200",
+      "sparse top 200",
+      "genome top 200"
+    ],
+    "candidate_union": true,
+    "colbert": "BGE-M3 MaxSim over the candidate union",
+    "weights": {
+      "dense": 0.28,
+      "sparse": 0.14,
+      "colbert": 0.28,
+      "genome": 0.3
+    },
+    "score_normalization": "sample z-score over candidate union",
+    "family_retrieval_weights": {
+      "B": 0.75,
+      "I": 0.5,
+      "N": 0.5,
+      "Q": 0.0,
+      "T": 0.75,
+      "U": 0.75,
+      "default": 1.0
+    },
+    "family_q_excluded": true,
+    "default_top_k": 20
+  },
+  "validation": {
+    "split": "val",
+    "question_ids": 250,
+    "rows": 2000,
+    "report_sha256": "6f87e359c3717660d6fee2c9c8caf3bde0543d22b6ac2a2d8379dd9aabb70f91"
+  }
+}

validation_q250.json

@@ -0,0 +1,678 @@
+{
+  "manifest": {
+    "experiment": "five_method_four_signal_validation_question_hidden4096_v1",
+    "split": "val",
+    "question_ids": [
+      "Q025",
+      "Q032",
+      "Q051",
+      "Q10001",
+      "Q10388",
+      "Q10516",
+      "Q10608",
+      "Q10619",
+      "Q10876",
+      "Q10893",
+      "Q11004",
+      "Q1163",
+      "Q11829",
+      "Q12118",
+      "Q12184",
+      "Q12211",
+      "Q12262",
+      "Q12334",
+      "Q12573",
+      "Q12608",
+      "Q12645",
+      "Q12657",
+      "Q12884",
+      "Q12885",
+      "Q1322",
+      "Q13463",
+      "Q14082",
+      "Q14476",
+      "Q14534",
+      "Q14758",
+      "Q14784",
+      "Q14815",
+      "Q14847",
+      "Q14881",
+      "Q15096",
+      "Q15278",
+      "Q15367",
+      "Q15493",
+      "Q155",
+      "Q15573",
+      "Q15631",
+      "Q15951",
+      "Q16060",
+      "Q16175",
+      "Q16212",
+      "Q16424",
+      "Q16719",
+      "Q1714",
+      "Q17201",
+      "Q17213",
+      "Q17254",
+      "Q17683",
+      "Q17750",
+      "Q17828",
+      "Q18030",
+      "Q18071",
+      "Q18080",
+      "Q18151",
+      "Q18157",
+      "Q18176",
+      "Q18233",
+      "Q18277",
+      "Q18313",
+      "Q18719",
+      "Q1880",
+      "Q1917",
+      "Q193",
+      "Q19701",
+      "Q19769",
+      "Q19819",
+      "Q20025",
+      "Q20285",
+      "Q20334",
+      "Q20353",
+      "Q20594",
+      "Q20823",
+      "Q21168",
+      "Q21170",
+      "Q21183",
+      "Q21393",
+      "Q21447",
+      "Q21583",
+      "Q21697",
+      "Q21775",
+      "Q21951",
+      "Q2213",
+      "Q22212",
+      "Q22224",
+      "Q22727",
+      "Q22956",
+      "Q23096",
+      "Q23135",
+      "Q23174",
+      "Q23405",
+      "Q24424",
+      "Q24523",
+      "Q24781",
+      "Q25098",
+      "Q25345",
+      "Q25494",
+      "Q25618",
+      "Q25656",
+      "Q2580",
+      "Q26048",
+      "Q26433",
+      "Q2668",
+      "Q26781",
+      "Q26812",
+      "Q26818",
+      "Q26913",
+      "Q27082",
+      "Q27174",
+      "Q27217",
+      "Q2733",
+      "Q27645",
+      "Q27841",
+      "Q27882",
+      "Q28076",
+      "Q28175",
+      "Q28896",
+      "Q29160",
+      "Q29512",
+      "Q29663",
+      "Q29794",
+      "Q30022",
+      "Q30178",
+      "Q30383",
+      "Q31025",
+      "Q31068",
+      "Q31236",
+      "Q31407",
+      "Q31574",
+      "Q31660",
+      "Q31922",
+      "Q32625",
+      "Q33243",
+      "Q33250",
+      "Q33445",
+      "Q33672",
+      "Q34292",
+      "Q3487",
+      "Q35883",
+      "Q36375",
+      "Q36667",
+      "Q36729",
+      "Q36865",
+      "Q37269",
+      "Q3732",
+      "Q37562",
+      "Q37814",
+      "Q38116",
+      "Q38320",
+      "Q38886",
+      "Q39179",
+      "Q39680",
+      "Q39866",
+      "Q39939",
+      "Q40123",
+      "Q40343",
+      "Q40758",
+      "Q40794",
+      "Q40975",
+      "Q41061",
+      "Q41123",
+      "Q41162",
+      "Q41229",
+      "Q41633",
+      "Q41808",
+      "Q41967",
+      "Q41980",
+      "Q42002",
+      "Q42062",
+      "Q42285",
+      "Q42288",
+      "Q42321",
+      "Q42344",
+      "Q42618",
+      "Q42806",
+      "Q42825",
+      "Q43120",
+      "Q43276",
+      "Q43290",
+      "Q43768",
+      "Q43962",
+      "Q44001",
+      "Q44151",
+      "Q44206",
+      "Q44230",
+      "Q44399",
+      "Q44509",
+      "Q44544",
+      "Q4474",
+      "Q4479",
+      "Q45231",
+      "Q45258",
+      "Q45522",
+      "Q4591",
+      "Q45918",
+      "Q4594",
+      "Q46003",
+      "Q46006",
+      "Q46190",
+      "Q46246",
+      "Q4646",
+      "Q46627",
+      "Q46733",
+      "Q46736",
+      "Q46853",
+      "Q46868",
+      "Q46927",
+      "Q4701",
+      "Q47120",
+      "Q47246",
+      "Q4748",
+      "Q4765",
+      "Q4900",
+      "Q499",
+      "Q5086",
+      "Q5119",
+      "Q5591",
+      "Q5909",
+      "Q5980",
+      "Q6083",
+      "Q6182",
+      "Q6199",
+      "Q6360",
+      "Q6397",
+      "Q6455",
+      "Q6502",
+      "Q6671",
+      "Q6745",
+      "Q6821",
+      "Q6942",
+      "Q7290",
+      "Q7619",
+      "Q788",
+      "Q7887",
+      "Q797",
+      "Q8269",
+      "Q8308",
+      "Q8347",
+      "Q8391",
+      "Q856",
+      "Q8933",
+      "Q9462",
+      "Q9577",
+      "Q9610",
+      "Q9668",
+      "Q9929",
+      "Q9943"
+    ],
+    "languages": [
+      "cebuano",
+      "english",
+      "french",
+      "german",
+      "greek",
+      "hebrew",
+      "spanish",
+      "tagalog"
+    ],
+    "sample_seed": 20261006,
+    "candidate_language": "english",
+    "candidate_verses": 41995,
+    "candidates_per_lane": 200,
+    "methods": [
+      "dense",
+      "weighted_genome",
+      "weighted_90_10",
+      "bge_dense_sparse_colbert",
+      "experiment_d"
+    ],
+    "four_signal_weights": {
+      "dense": 0.28,
+      "sparse": 0.14,
+      "colbert": 0.28,
+      "genome": 0.3
+    },
+    "weighted_two_signal": {
+      "genome": 0.9,
+      "dense": 0.1
+    },
+    "checkpoint_sha256": "5a68ef73167f48899c927dfb5e25dfe8ff50d0b8e8310cd6f3ff3c93b8bf9952",
+    "checkpoint_track": "question",
+    "checkpoint_hidden": 4096,
+    "fp16_head_sha256": "55e910550ccff8e463729379c6c08c3ebb72f152ba4df0c3217d143aa91b9cf3",
+    "fp16_head_parity_max_abs": 0.0008486509323120117,
+    "fp16_head_parity_mean_cosine": 1.0,
+    "bge_m3_revision": "5617a9f61b028005a4858fdac845db406aefb181",
+    "flagembedding_version": "1.4.2",
+    "family_q_weight": 0.0
+  },
+  "completed_question_ids": 250,
+  "completed_rows": 2000,
+  "expected_rows": 2000,
+  "overall_cluster_bootstrap": {
+    "dense": {
+      "recall@10": {
+        "mean": 0.024358333333333336,
+        "low": 0.017833333333333333,
+        "high": 0.03143354166666666
+      },
+      "recall@20": {
+        "mean": 0.037825000000000004,
+        "low": 0.028966458333333334,
+        "high": 0.04721687499999999
+      },
+      "mrr@20": {
+        "mean": 0.051011985363707504,
+        "low": 0.03494716504776828,
+        "high": 0.06874623539703117
+      },
+      "weighted_ndcg@20": {
+        "mean": 0.03106096248570316,
+        "low": 0.021967145016875016,
+        "high": 0.040995203099106564
+      }
+    },
+    "weighted_genome": {
+      "recall@10": {
+        "mean": 0.1350833333333333,
+        "low": 0.11531645833333333,
+        "high": 0.15438583333333328
+      },
+      "recall@20": {
+        "mean": 0.17568333333333333,
+        "low": 0.15284145833333332,
+        "high": 0.19755854166666664
+      },
+      "mrr@20": {
+        "mean": 0.24174455661297764,
+        "low": 0.20410701687615146,
+        "high": 0.2787487501964732
+      },
+      "weighted_ndcg@20": {
+        "mean": 0.15626929941091317,
+        "low": 0.13224160393033846,
+        "high": 0.17942760874826716
+      }
+    },
+    "weighted_90_10": {
+      "recall@10": {
+        "mean": 0.14486666666666667,
+        "low": 0.124631875,
+        "high": 0.16446937499999997
+      },
+      "recall@20": {
+        "mean": 0.18672499999999997,
+        "low": 0.162731875,
+        "high": 0.20931812499999997
+      },
+      "mrr@20": {
+        "mean": 0.26032609940584006,
+        "low": 0.22205060619382555,
+        "high": 0.29715744103715125
+      },
+      "weighted_ndcg@20": {
+        "mean": 0.1676779937164905,
+        "low": 0.14310348312589943,
+        "high": 0.19142050183568896
+      }
+    },
+    "bge_dense_sparse_colbert": {
+      "recall@10": {
+        "mean": 0.03149999999999999,
+        "low": 0.024,
+        "high": 0.039525000000000005
+      },
+      "recall@20": {
+        "mean": 0.048958333333333326,
+        "low": 0.03919104166666667,
+        "high": 0.05955124999999998
+      },
+      "mrr@20": {
+        "mean": 0.06272950293600447,
+        "low": 0.046572760353251835,
+        "high": 0.08041405770930778
+      },
+      "weighted_ndcg@20": {
+        "mean": 0.038395102460161826,
+        "low": 0.028874713968453012,
+        "high": 0.048275712884382205
+      }
+    },
+    "experiment_d": {
+      "recall@10": {
+        "mean": 0.06696666666666667,
+        "low": 0.055533124999999996,
+        "high": 0.07885104166666664
+      },
+      "recall@20": {
+        "mean": 0.09768333333333332,
+        "low": 0.08335729166666668,
+        "high": 0.11235875
+      },
+      "mrr@20": {
+        "mean": 0.1367895560944903,
+        "low": 0.11040850932429927,
+        "high": 0.16394163302936954
+      },
+      "weighted_ndcg@20": {
+        "mean": 0.08129438742965414,
+        "low": 0.06645920942862378,
+        "high": 0.09618033839610174
+      }
+    }
+  },
+  "candidate_union_recall": {
+    "mean": 0.44202500000000006,
+    "low": 0.41277375,
+    "high": 0.4705010416666666
+  },
+  "by_language": {
+    "cebuano": {
+      "dense": {
+        "recall@10": 0.009599999999999997,
+        "recall@20": 0.01653333333333333,
+        "mrr@20": 0.013236263736263736,
+        "weighted_ndcg@20": 0.00883686361129843
+      },
+      "weighted_genome": {
+        "recall@10": 0.06559999999999999,
+        "recall@20": 0.084,
+        "mrr@20": 0.11596143790849674,
+        "weighted_ndcg@20": 0.06762174970799319
+      },
+      "weighted_90_10": {
+        "recall@10": 0.07019999999999998,
+        "recall@20": 0.09526666666666667,
+        "mrr@20": 0.13499191159717475,
+        "weighted_ndcg@20": 0.07917209758621281
+      },
+      "bge_dense_sparse_colbert": {
+        "recall@10": 0.010399999999999998,
+        "recall@20": 0.019,
+        "mrr@20": 0.016534271284271283,
+        "weighted_ndcg@20": 0.011714408546470418
+      },
+      "experiment_d": {
+        "recall@10": 0.028866666666666665,
+        "recall@20": 0.049466666666666666,
+        "mrr@20": 0.03887013272245471,
+        "weighted_ndcg@20": 0.028267045882839566
+      }
+    },
+    "english": {
+      "dense": {
+        "recall@10": 0.04333333333333333,
+        "recall@20": 0.05893333333333334,
+        "mrr@20": 0.09459431913116124,
+        "weighted_ndcg@20": 0.05378794583919604
+      },
+      "weighted_genome": {
+        "recall@10": 0.18180000000000002,
+        "recall@20": 0.21993333333333334,
+        "mrr@20": 0.3214196276719497,
+        "weighted_ndcg@20": 0.20533247541321145
+      },
+      "weighted_90_10": {
+        "recall@10": 0.1878,
+        "recall@20": 0.23073333333333332,
+        "mrr@20": 0.3551371843842432,
+        "weighted_ndcg@20": 0.2215837109636164
+      },
+      "bge_dense_sparse_colbert": {
+        "recall@10": 0.061599999999999995,
+        "recall@20": 0.08519999999999998,
+        "mrr@20": 0.13645095956674902,
+        "weighted_ndcg@20": 0.07713335517154105
+      },
+      "experiment_d": {
+        "recall@10": 0.08893333333333334,
+        "recall@20": 0.1368,
+        "mrr@20": 0.21050320298896147,
+        "weighted_ndcg@20": 0.12229188387989554
+      }
+    },
+    "french": {
+      "dense": {
+        "recall@10": 0.030866666666666667,
+        "recall@20": 0.04440000000000001,
+        "mrr@20": 0.05858647259173575,
+        "weighted_ndcg@20": 0.03766874713675634
+      },
+      "weighted_genome": {
+        "recall@10": 0.16853333333333334,
+        "recall@20": 0.20986666666666667,
+        "mrr@20": 0.2913620264496426,
+        "weighted_ndcg@20": 0.18962705545276934
+      },
+      "weighted_90_10": {
+        "recall@10": 0.17726666666666666,
+        "recall@20": 0.22393333333333335,
+        "mrr@20": 0.30028972465443055,
+        "weighted_ndcg@20": 0.2008721636839681
+      },
+      "bge_dense_sparse_colbert": {
+        "recall@10": 0.042,
+        "recall@20": 0.05859999999999999,
+        "mrr@20": 0.09079896214896215,
+        "weighted_ndcg@20": 0.05207448089235733
+      },
+      "experiment_d": {
+        "recall@10": 0.07899999999999999,
+        "recall@20": 0.10733333333333332,
+        "mrr@20": 0.17117551768836597,
+        "weighted_ndcg@20": 0.09800224140884863
+      }
+    },
+    "german": {
+      "dense": {
+        "recall@10": 0.020933333333333335,
+        "recall@20": 0.03793333333333333,
+        "mrr@20": 0.046312064646739565,
+        "weighted_ndcg@20": 0.028948449436592406
+      },
+      "weighted_genome": {
+        "recall@10": 0.14746666666666666,
+        "recall@20": 0.192,
+        "mrr@20": 0.26682582359161305,
+        "weighted_ndcg@20": 0.16959824611897464
+      },
+      "weighted_90_10": {
+        "recall@10": 0.15986666666666668,
+        "recall@20": 0.20506666666666667,
+        "mrr@20": 0.28080794310608553,
+        "weighted_ndcg@20": 0.18153347436058184
+      },
+      "bge_dense_sparse_colbert": {
+        "recall@10": 0.028599999999999997,
+        "recall@20": 0.04853333333333334,
+        "mrr@20": 0.0557501837859423,
+        "weighted_ndcg@20": 0.03629092247905656
+      },
+      "experiment_d": {
+        "recall@10": 0.07593333333333334,
+        "recall@20": 0.11279999999999998,
+        "mrr@20": 0.13878295160292067,
+        "weighted_ndcg@20": 0.08594360676351548
+      }
+    },
+    "greek": {
+      "dense": {
+        "recall@10": 0.023599999999999996,
+        "recall@20": 0.03373333333333334,
+        "mrr@20": 0.04868164016925627,
+        "weighted_ndcg@20": 0.030088462723027752
+      },
+      "weighted_genome": {
+        "recall@10": 0.11533333333333334,
+        "recall@20": 0.16853333333333334,
+        "mrr@20": 0.20299260820019333,
+        "weighted_ndcg@20": 0.14016082410501918
+      },
+      "weighted_90_10": {
+        "recall@10": 0.13266666666666665,
+        "recall@20": 0.1795333333333333,
+        "mrr@20": 0.21963679710382497,
+        "weighted_ndcg@20": 0.15146138165604775
+      },
+      "bge_dense_sparse_colbert": {
+        "recall@10": 0.023666666666666666,
+        "recall@20": 0.03886666666666667,
+        "mrr@20": 0.05120649905649906,
+        "weighted_ndcg@20": 0.030345255970291946
+      },
+      "experiment_d": {
+        "recall@10": 0.056666666666666664,
+        "recall@20": 0.08706666666666667,
+        "mrr@20": 0.12885387495201736,
+        "weighted_ndcg@20": 0.07435042042467555
+      }
+    },
+    "hebrew": {
+      "dense": {
+        "recall@10": 0.02033333333333333,
+        "recall@20": 0.0316,
+        "mrr@20": 0.048446094016063056,
+        "weighted_ndcg@20": 0.029004258165204178
+      },
+      "weighted_genome": {
+        "recall@10": 0.12179999999999999,
+        "recall@20": 0.15546666666666667,
+        "mrr@20": 0.2236293274492965,
+        "weighted_ndcg@20": 0.139133328819003
+      },
+      "weighted_90_10": {
+        "recall@10": 0.12506666666666666,
+        "recall@20": 0.17573333333333335,
+        "mrr@20": 0.22922287605755098,
+        "weighted_ndcg@20": 0.15109517077939089
+      },
+      "bge_dense_sparse_colbert": {
+        "recall@10": 0.025733333333333334,
+        "recall@20": 0.04293333333333334,
+        "mrr@20": 0.05265930490218415,
+        "weighted_ndcg@20": 0.03377922256511489
+      },
+      "experiment_d": {
+        "recall@10": 0.06453333333333333,
+        "recall@20": 0.08693333333333333,
+        "mrr@20": 0.13843008943241142,
+        "weighted_ndcg@20": 0.07898237278508083
+      }
+    },
+    "spanish": {
+      "dense": {
+        "recall@10": 0.025799999999999997,
+        "recall@20": 0.048,
+        "mrr@20": 0.06944719659425541,
+        "weighted_ndcg@20": 0.03861702120570765
+      },
+      "weighted_genome": {
+        "recall@10": 0.16713333333333333,
+        "recall@20": 0.2165333333333333,
+        "mrr@20": 0.3031002419149478,
+        "weighted_ndcg@20": 0.199683629211233
+      },
+      "weighted_90_10": {
+        "recall@10": 0.1795333333333333,
+        "recall@20": 0.22513333333333332,
+        "mrr@20": 0.32094485105537734,
+        "weighted_ndcg@20": 0.20800683530362363
+      },
+      "bge_dense_sparse_colbert": {
+        "recall@10": 0.034866666666666664,
+        "recall@20": 0.0596,
+        "mrr@20": 0.07037315029311933,
+        "weighted_ndcg@20": 0.042085814650551334
+      },
+      "experiment_d": {
+        "recall@10": 0.08233333333333333,
+        "recall@20": 0.1146,
+        "mrr@20": 0.17114001471524692,
+        "weighted_ndcg@20": 0.09664569797242636
+      }
+    },
+    "tagalog": {
+      "dense": {
+        "recall@10": 0.020399999999999998,
+        "recall@20": 0.03146666666666667,
+        "mrr@20": 0.028791832024184966,
+        "weighted_ndcg@20": 0.021535951767842468
+      },
+      "weighted_genome": {
+        "recall@10": 0.113,
+        "recall@20": 0.15913333333333332,
+        "mrr@20": 0.20866535971768166,
+        "weighted_ndcg@20": 0.13899708645910142
+      },
+      "weighted_90_10": {
+        "recall@10": 0.12653333333333333,
+        "recall@20": 0.15839999999999999,
+        "mrr@20": 0.2415775072880336,
+        "weighted_ndcg@20": 0.1476991153984826
+      },
+      "bge_dense_sparse_colbert": {
+        "recall@10": 0.025133333333333334,
+        "recall@20": 0.038933333333333334,
+        "mrr@20": 0.02806269245030855,
+        "weighted_ndcg@20": 0.023737359405911092
+      },
+      "experiment_d": {
+        "recall@10": 0.05946666666666667,
+        "recall@20": 0.08646666666666666,
+        "mrr@20": 0.0965606646535439,
+        "weighted_ndcg@20": 0.06587183031995107
+      }
+    }
+  },
+  "mean_seconds_per_query": 2.1415414981744654,
+  "note": "Exploratory validation sample only. Question IDs are the independent units; translations are averaged within q_id."
+}