Update Croissant RAI metadata

croissant.json

@@ -21,6 +21,7 @@
       "@type": "@json"
     },
     "extract": "cr:extract",
+    "equivalentProperty": "cr:equivalentProperty",
     "field": "cr:field",
     "fileProperty": "cr:fileProperty",
     "fileObject": "cr:fileObject",
@@ -39,6 +40,7 @@
     "repeated": "cr:repeated",
     "replace": "cr:replace",
     "sc": "https://schema.org/",
+    "samplingRate": "cr:samplingRate",
     "separator": "cr:separator",
     "source": "cr:source",
     "subField": "cr:subField",
@@ -84,7 +86,7 @@
     "Released labels are produced by a brain-MRI segmenter trained on FreeSurfer 7.4.1 reference labels; they are not manually annotated ground truth. This dataset is not recommended for clinical decision support, individual-patient diagnosis, or fine-tuning of production diagnostic systems.",
     "The canonical pseudo-label segmenter inherits FreeSurfer 7.4.1 region conventions; users following alternative atlases (DKT-only, AAL, JHU) may observe boundary differences not captured by the released quality panel.",
     "The released quality panel is reference-free (no manual labels are consulted at scoring time) and source-local (axis values are normalised within each source dataset). It does not provide an absolute, manually anchored quality score.",
-    "Demographic metadata coverage is incomplete: age 62.8%, sex 71.5%, scanner manufacturer 73.1%, magnetic-field strength 69.5%, health status 27.6%. Missing values are reported as missing, not imputed.",
+    "Demographic metadata coverage is incomplete: age 63.8%, sex 72.9%, scanner manufacturer 74.9%, magnetic-field strength 71.0%, health status 29.0%. Missing values are reported as missing, not imputed.",
     "The Mindboggle-101 manually annotated cohort (n=101) is the only human-annotated cohort referenced in our paper; it is held out for a confirmatory evaluation and is not redistributed in this release.",
     "OpenBrain v1.0 is the CC0/public-domain-compatible subset of a larger 42,732-case provenance pool; non-CC0 sources are excluded and audited in metadata/excluded_cases_summary.csv."
   ],
@@ -142,7 +144,7 @@
       "@type": "cr:FileObject",
       "@id": "openbrain_v1_0_metadata.csv",
       "name": "openbrain_v1_0_metadata.csv",
-      "description": "Per-case core metadata: BIDS identifiers, source dataset DOI/snapshot, demographic fields (age, sex, race, handedness, weight, BMI), scanner fields (manufacturer, model, field strength, repetition time, echo time, phase encoding direction), and release-path fields. Coverage of demographic and scanner fields varies; see rai:dataBiases.",
+      "description": "Case-level core metadata for OpenBrain release candidates and released cases: BIDS identifiers, source dataset DOI/snapshot, demographic fields (age, sex, race, handedness, weight, BMI), scanner fields (manufacturer, model, field strength, repetition time, echo time, phase encoding direction), and release-path fields. The canonical 35,838-case hosted release is defined by metadata/openbrain_v1_0_qc_manifest.csv; this metadata table may include additional provenance-audited candidates retained for accounting. Coverage of demographic and scanner fields varies; see rai:dataBiases.",
       "contentUrl": "https://huggingface.co/datasets/openbrain-anon/openbrain_v1_0/resolve/main/metadata/openbrain_v1_0_metadata.csv",
       "encodingFormat": "text/csv",
       "sha256": "3ee84bc3e7daae4e29d910ff316658db1d33c79b492d020288d352f4c8974a5c"
@@ -151,7 +153,7 @@
       "@type": "cr:FileObject",
       "@id": "openbrain_v1_0_qc_manifest.csv",
       "name": "openbrain_v1_0_qc_manifest.csv",
-      "description": "Per-case Core-7 QC manifest: case_id, source_dataset, raw QC values Q1..Q7, within-source percentile risks r1..r7, combined risk R, and confidence tier (HighConf/Standard/Flagged). Tier counts in v1.0: HighConf 8,960 (25.0%) / Standard 17,918 (50.0%) / Flagged 8,960 (25.0%).",
+      "description": "Per-case QC-7 manifest: case_id, source_dataset, raw QC values Q1..Q7, within-source percentile risks r1..r7, combined risk R, and confidence tier (HighConf/Standard/Flagged). Tier counts in v1.0: HighConf 8,960 (25.0%) / Standard 17,918 (50.0%) / Flagged 8,960 (25.0%).",
       "contentUrl": "https://huggingface.co/datasets/openbrain-anon/openbrain_v1_0/resolve/main/metadata/openbrain_v1_0_qc_manifest.csv",
       "encodingFormat": "text/csv",
       "sha256": "3b073972b106ec085aaed52be1987db3206ab6b97dbb541bde68ec0b27b24b36"
@@ -169,7 +171,7 @@
       "@type": "cr:FileObject",
       "@id": "openneuro_dataset_audit.csv",
       "name": "openneuro_dataset_audit.csv",
-      "description": "Per-OpenNeuro-source provenance audit: dataset name, snapshot tag, DOI, license, and case count contributed.",
+      "description": "Per-OpenNeuro-source provenance audit: dataset name, snapshot tag, DOI, license, and case counts for release sources and retained provenance-accounting sources. The canonical 35,838-case hosted release is defined by metadata/openbrain_v1_0_qc_manifest.csv.",
       "contentUrl": "https://huggingface.co/datasets/openbrain-anon/openbrain_v1_0/resolve/main/metadata/openneuro_dataset_audit.csv",
       "encodingFormat": "text/csv",
       "sha256": "2f78ba8a728666cb76165a02827c2cac4973448b50f664c249c162aa6b40224c"
@@ -178,7 +180,7 @@
       "@type": "cr:FileObject",
       "@id": "openneuro_case_path_audit.csv",
       "name": "openneuro_case_path_audit.csv",
-      "description": "Per-case path audit: matched BIDS path within the OpenNeuro snapshot, original raw T1w image path, derived release paths.",
+      "description": "Case-path provenance audit for release candidates and released cases: matched BIDS path within the OpenNeuro snapshot, original raw T1w image path, match status, and notes. The canonical 35,838-case hosted release is defined by metadata/openbrain_v1_0_qc_manifest.csv.",
       "contentUrl": "https://huggingface.co/datasets/openbrain-anon/openbrain_v1_0/resolve/main/metadata/openneuro_case_path_audit.csv",
       "encodingFormat": "text/csv",
       "sha256": "e290704cb2431c33c32a7e142284ea554a62bee9f93aed4f1f782eda4f0fee9f"
@@ -231,7 +233,7 @@
       "@type": "cr:RecordSet",
       "@id": "cases",
       "name": "cases",
-      "description": "One record per released case. Joins core metadata + Core-7 QC manifest on case_id.",
+      "description": "One record per released case. Joins core metadata + QC-7 manifest on case_id.",
       "key": {
         "@id": "cases/case_id"
       },
@@ -518,4 +520,4 @@
     "Source-local percentile-risk normalisation: each raw quality-check value is converted to a within-source rank in [0, 1] before being averaged into the combined risk score."
   ],
   "rai:dataAnnotationProtocol": "Released labels are produced by trained segmenters rather than human annotators. The canonical released label per case comes from a 3D residual-encoder U-Net trained on 1,189 brain MRI scans with FreeSurfer 7.4.1 reference labels. Two committee segmenters (SwinUNETR and UNETR), trained against the same FreeSurfer-aligned reference space, are run in parallel; their outputs are not the canonical released label but feed three of the seven per-case quality checks (committee Dice, voxel-vote entropy, committee volume coefficient of variation). No per-case manual review is performed on the OpenBrain release. Human supervision enters upstream through the FreeSurfer-curated training data of all three segmenters and downstream through Mindboggle-101, a manually annotated external cohort that is held out for confirmatory evaluation rather than redistributed."
-}
+}