Publish rebalanced BBB assay transfer V21

README.md

@@ -0,0 +1,26 @@
+---
+configs:
+- config_name: default
+  data_files:
+  - split: train
+    path: train/data.parquet
+  - split: validation_ranking
+    path: validation_ranking/data.parquet
+  - split: test_ranking
+    path: test_ranking/data.parquet
+---
+# BBB all-record assay transfer V21
+
+V21 freezes V20 records, splits, source-native prompts, target geometry, and 20-parent
+Morgan ranking panels. Training buckets are ranked by train-record count and assigned
+independent per-record query/retrieval degree caps of 96, 48, 24, 12, or 6 by size
+quintile. Direct-BBB records always use cap 6, and deterministic bucket-balanced
+downsampling limits direct pairs to 50% of training rows. Parent roles remain capped at 576.
+
+- train rows: 480,974
+- validation ranking rows: 2,800
+- test ranking rows: 5,060
+- achieved direct-BBB train share: 0.500000
+- continuous target: unchanged V20 train-SD sigmoid centered at 0.4 SD
+- prompts and evidence: original source values and units
+- log10 geometry: machine-readable metric geometry only, never prompt text

manifest.json

@@ -0,0 +1,71 @@
+{
+  "arrow_schema": "prompt: string\ncompletion: string\nsplit: string\npair_id: string\nranking_query_id: string\nranking_member_index: int64\nranking_candidate_row_count: int64\nranking_candidate_parent_count: int64\nretrieval_parent_rank: int64\nretrieval_parent_context_index: int64\nretrieval_parent_context_count: int64\ntask_id: string\nsource_id: string\nretrieval_strategy: string\nquery_record_id: string\nretrieval_record_id: string\nquery_benchmark_row_id: string\nretrieval_benchmark_row_id: string\nquery_molecule_identity_key: string\nretrieval_molecule_identity_key: string\nquery_smiles: string\nretrieval_smiles: string\nquery_condition_group: string\nretrieval_condition_group: string\nquery_condition_scope: string\nretrieval_condition_scope: string\nquery_condition_atoms: list<element: string>\n  child 0, element: string\nretrieval_condition_atoms: list<element: string>\n  child 0, element: string\ncandidate_context_relation: string\nquery_source_record_count: int64\nretrieval_source_record_count: int64\nquery_agreement_fraction: double\nretrieval_agreement_fraction: double\nquery_gold_Y: int64\nretrieval_gold_Y: int64\nranking_balance_label: string\nquery_geometry_value: double\nretrieval_geometry_value: double\nabsolute_geometry_difference: double\ncalibration_sample_standard_deviation: double\ntarget_distribution: struct<nontransfer: double, transfer: double>\n  child 0, nontransfer: double\n  child 1, transfer: double\ncanonical_endpoint_key: string\nassay_concept: string\nmeasurement_kind: string\nmetadata: struct<calibration_sha256: string, groups: struct<assay_concept: string, measurement_kind: string, q (... 230 chars omitted)\n  child 0, calibration_sha256: string\n  child 1, groups: struct<assay_concept: string, measurement_kind: string, query_condition_group: string, query_conditi (... 81 chars omitted)\n      child 0, assay_concept: string\n      child 1, measurement_kind: string\n      child 2, query_condition_group: string\n      child 3, query_condition_scope: string\n      child 4, retrieval_strategy: string\n      child 5, source_id: string\n      child 6, task_id: string\n  child 2, soft_targets: struct<(A): double, (B): double>\n      child 0, (A): double\n      child 1, (B): double\n  child 3, source_snapshot_sha256: string\n  child 4, target_contract: string\nabsolute_canonical_value_difference: double\nstandardized_canonical_value_difference: double\nsd_logistic_similarity: double\ntarget_train_sample_standard_deviation: double\ntarget_a: double\ntarget_b: double\nlabel_mode: string\nbinary_label: int64\nis_decisive: bool\nmorgan_tanimoto_similarity: double\ngraph_a_position: int64\ngraph_b_position: int64\ngraph_offset: int64\nrepresentation: string\npair_bucket_key: string\nquery_parent_id: string\nretrieval_parent_id: string\nquery_scaffold: string\nretrieval_scaffold: string\nquery_evidence: struct<aggregation_method: string, display_measurement_text: string, display_unit_text: string, disp (... 184 chars omitted)\n  child 0, aggregation_method: string\n  child 1, display_measurement_text: string\n  child 2, display_unit_text: string\n  child 3, display_transform_id: string\n  child 4, source_record_count: int64\n  child 5, deduplicated_source_record_count: int64\n  child 6, aggregate_min: double\n  child 7, aggregate_max: double\n  child 8, aggregate_q1: double\n  child 9, aggregate_q3: double\nretrieval_evidence: struct<aggregation_method: string, display_measurement_text: string, display_unit_text: string, disp (... 184 chars omitted)\n  child 0, aggregation_method: string\n  child 1, display_measurement_text: string\n  child 2, display_unit_text: string\n  child 3, display_transform_id: string\n  child 4, source_record_count: int64\n  child 5, deduplicated_source_record_count: int64\n  child 6, aggregate_min: double\n  child 7, aggregate_max: double\n  child 8, aggregate_q1: double\n  child 9, aggregate_q3: double",
+  "artifacts": {
+    "README.md": "8decbd3e4fa3f47472d226c79ee426d2084ed85bef54ae7a3d2d2a7125eb1dd4",
+    "calibration.json": "5997cb99660910b1a132fba7fc0333f3e124fbfff392ede828a0d85fd1ccf7f1",
+    "test_ranking/data.parquet": "6083e9a98af4e9af27d292f3efb65f68ed74673a8a7cba48befe59c339850d5e",
+    "train/data.parquet": "0e1d8f8d8ecf5db5f7411ad71326091f1fee187a153025268d005f7113eb69d1",
+    "validation_ranking/data.parquet": "9ebac30d46558e8ffdeeab2e60593cbf968c03ec5ea3556e01f755ed413a9b65"
+  },
+  "build": {
+    "degree_tiers": [
+      96,
+      48,
+      24,
+      12,
+      6
+    ],
+    "direct_degree_cap": 6,
+    "direct_target_share": 0.5,
+    "global_parent_role_cap": 576,
+    "implementation_sha256": {
+      "assay_transfer/record_level/v20/geometry.py": "a53c448cfa2fd0278b281b61c01b979e21ce7b6a3e68812606643d7aec42cdf9",
+      "assay_transfer/record_level/v20/hf.py": "24e3a7d51e3593fa8b615366ff47bced94a362e3ab8d1690dd1213e1db899e9d",
+      "assay_transfer/record_level/v20/logit_geometry_interventions.json": "4efd77a959295c8fe9db5c8907773ccc5188a12ac18a915a92a9c6433ee1c583",
+      "assay_transfer/record_level/v20/projection.py": "86acf4233baf8cd7eaf5f0ec2f2cd16b8374ae00e6f0736dc7c9eb7eb44e6135",
+      "assay_transfer/record_level/v20/prompt.jinja": "435d70f963d9b55ac9b13da6f8c32af5414ee402d2af3af08fc5be01c4c7203d",
+      "assay_transfer/record_level/v21/__init__.py": "5a676e9e51e71f0f33f348a7b4e5e5b55008ebd722869efec89bbf217a250ae4",
+      "assay_transfer/record_level/v21/build.py": "57116d05b65d8f93a81331e946f734056fa17b19e9a4b5561c0df03b8fb7fcb6",
+      "assay_transfer/record_level/v21/contract.py": "2a30b15f6162cfe76d8ea4606f97c94629424db5e675c71ecaef94dfdd4c26cd",
+      "assay_transfer/record_level/v21/hf.py": "af5cdebc79db04c845ea9a68a91d6f61e06c9c25b2fa208d95cd3c185c868a4a",
+      "assay_transfer/record_level/v21/pairs.py": "2df23d34c5d634de190ab7b87a14fda5a055960a606a0c81892d1ca0fb46fa66",
+      "assay_transfer/record_level/v21/publish.py": "1fb9c0c2ce83c9d33425b028eadf1923b1039d01941f2c674de1b3fc4e1824b6"
+    },
+    "packages": {
+      "huggingface_hub": "0.36.2",
+      "pyarrow": "23.0.0",
+      "rdkit": "2025.9.4"
+    },
+    "python": "3.11.14",
+    "ranking_cap": 20,
+    "repository_commit": "7ae235a1573a01dc281266ff781bda01e80d8ae0",
+    "repository_dirty": true
+  },
+  "calibration_sha256": "5997cb99660910b1a132fba7fc0333f3e124fbfff392ede828a0d85fd1ccf7f1",
+  "hub_dataset_id": "jiosephlee/assay-transfer-record-level-v21-bbb-martins-mixed-intern",
+  "label_mode": "mixed_v9_record_level",
+  "membership": {
+    "test_ranking": "575c6a55ebeea6b0d5681672559e0190693665670591b9bd16b1fb61287af879",
+    "train": "203583adb0c7a898c2178db0ce71b19ed4c04f465f2e9af00e164bc3f63a97fd",
+    "validation_ranking": "0d1eea286771798d8d442a1d2e75f13822cd1bcc3ddd08cfc43ab6090b208693"
+  },
+  "prompt_contract": {
+    "canonical_fallback": false,
+    "known_result_fields": "all",
+    "query_result_fields": "hidden",
+    "source_native_columns": true,
+    "template_sha256": "435d70f963d9b55ac9b13da6f8c32af5414ee402d2af3af08fc5be01c4c7203d"
+  },
+  "rows": {
+    "test_ranking": 5060,
+    "train": 480974,
+    "validation_ranking": 2800
+  },
+  "schema_version": "assay_transfer_all_record_bbb.v21",
+  "source_pair_manifest_sha256": "afe236163ae4b7ffc529343035ec47388254b6c11824a0db688dc739a90387ea",
+  "source_split_manifest_sha256": "91a5e67f23a882bafeba1b0ae88f5583ff9ba30ec522bd96abaacb89bd792a45",
+  "stage": "hf",
+  "target_contract": "train_sd_sigmoid_logit_geometry_intervention_empirical_binary.v20",
+  "task_id": "bbb_martins",
+  "version": "v21"
+}

test_ranking/data.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6083e9a98af4e9af27d292f3efb65f68ed74673a8a7cba48befe59c339850d5e
+size 1863646

train/data.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e1d8f8d8ecf5db5f7411ad71326091f1fee187a153025268d005f7113eb69d1
+size 175181744

validation_ranking/data.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ebac30d46558e8ffdeeab2e60593cbf968c03ec5ea3556e01f755ed413a9b65
+size 1036970