Croissant: place rai:hasSyntheticData near top + add unprefixed alias for validator compatibility

CROISSANT.json

@@ -49,6 +49,62 @@
     "syntheticDataGeneration": "rai:syntheticDataGeneration"
   },
   "@type": "sc:Dataset",
+  "name": "TSFMI",
+  "description": "\n\t\n\t\t\n\t\tTSFMI-Synthetic\n\t\n\nSynthetic time-series datasets with mathematically exact ground-truth labels\nfor the TSFMI baseline-controlled probing protocol.\n\nCompanion data for the NeurIPS 2026 Evaluations & Datasets Track submission\n\"TSFMI: A Baseline-Controlled Evaluation Protocol for Time-Series Foundation\nModel Representations.\" The code (anonymous) lives at\nhttps://anonymous.4open.science/r/TSFMI.\n\n\n\t\n\t\t\n\t\n\t\n\t\tWhy this dataset exists\n\t\n\nProbing time-series foundation models (TSFMs) is hard\u2026 See the full description on the dataset page: https://huggingface.co/datasets/EvalData/TSFMI.",
+  "license": "https://choosealicense.com/licenses/mit/",
+  "url": "https://huggingface.co/datasets/EvalData/TSFMI",
+  "creator": {
+    "@type": "Person",
+    "name": "Evaldataset",
+    "url": "https://huggingface.co/EvalData"
+  },
+  "datePublished": "2026-05-05",
+  "version": "1.0.0",
+  "isLiveDataset": false,
+  "conformsTo": "http://mlcommons.org/croissant/1.1",
+  "alternateName": [
+    "EvalData/TSFMI",
+    "TSFMI-Synthetic"
+  ],
+  "keywords": [
+    "time-series-forecasting",
+    "tabular-classification",
+    "tabular-regression",
+    "English",
+    "mit",
+    "10K - 100K",
+    "parquet",
+    "Tabular",
+    "Time-series",
+    "Datasets",
+    "pandas",
+    "Polars",
+    "Croissant",
+    "\ud83c\uddfa\ud83c\uddf8 Region: US",
+    "time-series",
+    "foundation-models",
+    "probing",
+    "benchmark",
+    "evaluation",
+    "Synthetic",
+    "baseline-controlled",
+    "neurips-2026"
+  ],
+  "citeAs": "Anonymous Authors, TSFMI: A Baseline-Controlled Evaluation Protocol for Time-Series Foundation Model Representations. NeurIPS 2026 Evaluations & Datasets Track. https://anonymous.4open.science/r/TSFMI/",
+  "rai:dataAnnotationProtocol": "Labels are derived analytically from generation parameters. No human annotation.",
+  "rai:dataBiases": "By construction the synthetic data has no human or demographic content and therefore no demographic bias. The canonical generators are deliberately simple; this motivates the baseline-controlled probing protocol that is the paper's main contribution.",
+  "rai:dataCollection": "Synthetic generation via deterministic Python procedures with NumPy random number generators. No human subjects, no scraping, no third-party data ingestion.",
+  "rai:dataCollectionType": "Synthetic / procedural",
+  "rai:dataLimitations": "Each generator instantiates ONE family of synthetic constructions and is therefore a statistically simple test bed. The kurtosis sufficiency of the canonical anomaly task is documented in the paper Appendix A.8; a 'realistic_anomaly' variant with mixed anomaly types is included to bound this limitation. Synthetic-to-real transfer is weak (paper \u00a74.2).",
+  "rai:dataMisuseCases": "Not intended for deployment-time decision making, anomaly detection in production, or as a substitute for any real-world benchmark. The canonical anomaly task should not be used to claim a model 'detects anomalies' in any operational sense (paper \u00a76 Misuse Risks).",
+  "rai:dataPreprocessingProtocol": "None. Generators emit final tensors (N x seq_len) with labels directly.",
+  "rai:dataSocialImpact": "The protocol is intended to raise the rigor bar on TSFM representation claims. Direct social impact is low (synthetic data, no human subjects); indirect impact is methodological -- discouraging over-claiming in time-series probing literature.",
+  "rai:dataUseCases": "Probing internal representations of pre-trained time-series foundation models; calibrating new probing protocols against simple non-model baselines; comparing intervention diagnostics (LEACE, LDA steering, CKA) under matched controls.",
+  "rai:hasSyntheticData": true,
+  "rai:personalSensitiveInformation": "None. Datasets contain no personal, demographic, medical, or otherwise sensitive information.",
+  "rai:provenanceActivities": "Procedural generation by src/datasets/synthetic.py (seed=42), 60/20/20 split (split_seed=0), Parquet serialization. See the GitHub-anonymous mirror at https://anonymous.4open.science/r/TSFMI/ for source code.",
+  "rai:sourceDatasets": "None. TSFMI-Synthetic is generated from scratch and does not incorporate or transform any pre-existing dataset.",
+  "rai:syntheticDataGeneration": "All TSFMI-Synthetic data is fully synthetic; no real-world or human-derived signals.",
   "distribution": [
     {
       "@type": "cr:FileObject",
@@ -1371,60 +1427,6 @@
       ]
     }
   ],
-  "conformsTo": "http://mlcommons.org/croissant/1.1",
-  "name": "TSFMI",
-  "description": "\n\t\n\t\t\n\t\tTSFMI-Synthetic\n\t\n\nSynthetic time-series datasets with mathematically exact ground-truth labels\nfor the TSFMI baseline-controlled probing protocol.\n\nCompanion data for the NeurIPS 2026 Evaluations & Datasets Track submission\n\"TSFMI: A Baseline-Controlled Evaluation Protocol for Time-Series Foundation\nModel Representations.\" The code (anonymous) lives at\nhttps://anonymous.4open.science/r/TSFMI.\n\n\n\t\n\t\t\n\t\n\t\n\t\tWhy this dataset exists\n\t\n\nProbing time-series foundation models (TSFMs) is hard\u2026 See the full description on the dataset page: https://huggingface.co/datasets/EvalData/TSFMI.",
-  "alternateName": [
-    "EvalData/TSFMI",
-    "TSFMI-Synthetic"
-  ],
-  "creator": {
-    "@type": "Person",
-    "name": "Evaldataset",
-    "url": "https://huggingface.co/EvalData"
-  },
-  "keywords": [
-    "time-series-forecasting",
-    "tabular-classification",
-    "tabular-regression",
-    "English",
-    "mit",
-    "10K - 100K",
-    "parquet",
-    "Tabular",
-    "Time-series",
-    "Datasets",
-    "pandas",
-    "Polars",
-    "Croissant",
-    "\ud83c\uddfa\ud83c\uddf8 Region: US",
-    "time-series",
-    "foundation-models",
-    "probing",
-    "benchmark",
-    "evaluation",
-    "Synthetic",
-    "baseline-controlled",
-    "neurips-2026"
-  ],
-  "license": "https://choosealicense.com/licenses/mit/",
-  "url": "https://huggingface.co/datasets/EvalData/TSFMI",
-  "rai:dataCollection": "Synthetic generation via deterministic Python procedures with NumPy random number generators. No human subjects, no scraping, no third-party data ingestion.",
-  "rai:dataCollectionType": "Synthetic / procedural",
-  "rai:dataPreprocessingProtocol": "None. Generators emit final tensors (N x seq_len) with labels directly.",
-  "rai:dataAnnotationProtocol": "Labels are derived analytically from generation parameters. No human annotation.",
-  "rai:dataLimitations": "Each generator instantiates ONE family of synthetic constructions and is therefore a statistically simple test bed. The kurtosis sufficiency of the canonical anomaly task is documented in the paper Appendix A.8; a 'realistic_anomaly' variant with mixed anomaly types is included to bound this limitation. Synthetic-to-real transfer is weak (paper \u00a74.2).",
-  "rai:dataBiases": "By construction the synthetic data has no human or demographic content and therefore no demographic bias. The canonical generators are deliberately simple; this motivates the baseline-controlled probing protocol that is the paper's main contribution.",
-  "rai:personalSensitiveInformation": "None. Datasets contain no personal, demographic, medical, or otherwise sensitive information.",
-  "rai:dataUseCases": "Probing internal representations of pre-trained time-series foundation models; calibrating new probing protocols against simple non-model baselines; comparing intervention diagnostics (LEACE, LDA steering, CKA) under matched controls.",
-  "rai:dataMisuseCases": "Not intended for deployment-time decision making, anomaly detection in production, or as a substitute for any real-world benchmark. The canonical anomaly task should not be used to claim a model 'detects anomalies' in any operational sense (paper \u00a76 Misuse Risks).",
-  "rai:dataSocialImpact": "The protocol is intended to raise the rigor bar on TSFM representation claims. Direct social impact is low (synthetic data, no human subjects); indirect impact is methodological -- discouraging over-claiming in time-series probing literature.",
-  "rai:provenanceActivities": "Procedural generation by src/datasets/synthetic.py (seed=42), 60/20/20 split (split_seed=0), Parquet serialization. See the GitHub-anonymous mirror at https://anonymous.4open.science/r/TSFMI/ for source code.",
-  "rai:sourceDatasets": "None. TSFMI-Synthetic is generated from scratch and does not incorporate or transform any pre-existing dataset.",
-  "citeAs": "Anonymous Authors, TSFMI: A Baseline-Controlled Evaluation Protocol for Time-Series Foundation Model Representations. NeurIPS 2026 Evaluations & Datasets Track. https://anonymous.4open.science/r/TSFMI/",
-  "isLiveDataset": false,
-  "version": "1.0.0",
-  "datePublished": "2026-05-05",
-  "rai:hasSyntheticData": true,
-  "rai:syntheticDataGeneration": "All TSFMI-Synthetic data is fully synthetic; no real-world or human-derived signals."
+  "hasSyntheticData": true,
+  "syntheticDataGeneration": "All TSFMI-Synthetic data is fully synthetic; no real-world or human-derived signals."
 }