Update croissant.json to RAI 1.1 spec (add hasSyntheticData, wasDerivedFrom, wasGeneratedBy, real md5)

croissant.json

@@ -7,6 +7,7 @@
     "conformsTo": "dct:conformsTo",
     "cr": "http://mlcommons.org/croissant/",
     "rai": "http://mlcommons.org/croissant/RAI/",
+    "prov": "http://www.w3.org/ns/prov#",
     "data": {"@id": "cr:data", "@type": "@json"},
     "dataType": {"@id": "cr:dataType", "@type": "@vocab"},
     "dct": "http://purl.org/dc/terms/",
@@ -40,8 +41,8 @@
   "@type": "sc:Dataset",
   "name": "BillSim",
   "conformsTo": [
-    "http://mlcommons.org/croissant/1.0",
-    "http://mlcommons.org/croissant/RAI/1.0"
+    "http://mlcommons.org/croissant/1.1",
+    "http://mlcommons.org/croissant/RAI/1.1"
   ],
   "description": "BillSim is a synthetic heterogeneous graph benchmark for non-adversarial anomaly detection in usage-based billing pipelines. The default dataset contains 63,155 nodes across 11 entity types (customer, contract, sku, pricing_rule, usage_record, rated_charge, invoice, line_item, payment, ar_record, gl_entry) and 108,930 edges across 14 relationship types (subscribes_to, has_sku, governs, applies_to, generates_usage, for_sku, rates_to, priced_by, billed_on, part_of, invoiced_to, settles, opens_ar, posts_to). Six anomaly types are organized into three categories: structural (Pricing Mismatch, Revenue Recognition, Premature AR Closure), relational (Missing Metering, Orphaned Invoice Lines), and peer-comparison (Stale Discount). Five baselines are evaluated (MLP, GCN, GAT, RGCN, HGT) across 10 random seeds and 4 injection rates (0.01, 0.03, 0.05, 0.10). Headline finding: a 0.374 AUROC oracle gap on peer-comparison anomalies between a hand-coded oracle (0.965 AUROC on Stale Discount) and the best message-passing baseline (RGCN at 0.591; HGT at 0.540 is statistically indistinguishable), localizing the limitation to message-passing aggregation rather than task ill-posedness. Distributed as a PyTorch Geometric HeteroData (.pt) tensor plus deterministic Python generation scripts.",
   "citeAs": "Anonymous Authors. BillSim: A Synthetic Heterogeneous Graph Benchmark for Non-Adversarial Anomaly Detection in Usage-Based Billing. Submitted to NeurIPS 2026 Evaluations & Datasets Track, 2026.",
@@ -71,7 +72,58 @@
   "rai:dataSocialImpact": "Low. The dataset contains no real-world subjects and cannot be used to identify, target, or impact any individual or group. The primary intended impact is methodological: it provides a reproducible benchmark for studying when and why message-passing graph neural networks fail on operational anomaly-detection tasks that require peer-distribution comparison.",
   "rai:dataLimitations": "BillSim covers six anomaly types representative of common operational failure modes in usage-based billing (missed metering, pricing mismatches, orphaned invoice lines, stale discounts, revenue recognition errors, premature AR closure) but is not exhaustive of real-world billing system failure modes. Synthetic distributions approximate but do not replicate the full complexity of production billing data (no tax tiers, no currency mismatches, no retroactive adjustments, no off-cycle events).",
   "rai:dataReleaseMaintenancePlan": "The dataset is released as a single default PyTorch Geometric HeteroData artifact plus the deterministic generation pipeline. Because the generator is deterministic given (seed, injection_rate), the dataset can be regenerated indefinitely without depending on continued hosting of the binary artifact. Issues and contributions are accepted via the anonymous code mirror referenced in the paper.",
-  "cr:syntheticData": true,
+  "rai:hasSyntheticData": true,
+  "prov:wasDerivedFrom": [
+    {
+      "@type": "prov:Entity",
+      "prov:label": "NumPy random seed (default: 42)",
+      "description": "BillSim is fully synthetic and has no real-world parent dataset. The 'derived-from' entity for synthetic data is the random seed plus the parameterized statistical distributions defined in billsim/config.py. The default release uses NumPy random seed 42 with injection_rate 0.05; the full benchmark protocol sweeps 10 seeds (42, 123, 456, 789, 1024, 2048, 4096, 8192, 16384, 32768) and 4 rates (0.01, 0.03, 0.05, 0.10). Distributions: Pareto (revenue), log-normal (usage volumes), Poisson (arrivals), uniform (pricing). No real-world data was sampled, scraped, purchased, or inherited."
+    }
+  ],
+  "prov:wasGeneratedBy": [
+    {
+      "@type": "prov:Activity",
+      "prov:type": "Data Collection (synthetic generation)",
+      "prov:label": "Heterogeneous billing-graph synthesis",
+      "description": "A deterministic Python pipeline (billsim.generate) instantiates a 24-month simulated horizon of an enterprise usage-based-billing operation, producing 63,155 nodes across 11 entity types (customer, contract, sku, pricing_rule, usage_record, rated_charge, invoice, line_item, payment, ar_record, gl_entry) and 108,930 edges across 14 relationship types (subscribes_to, has_sku, governs, applies_to, generates_usage, for_sku, rates_to, priced_by, billed_on, part_of, invoiced_to, settles, opens_ar, posts_to). Generation is bit-level reproducible given (seed, injection_rate). No real customer, account, transaction, or other financial data is collected, sampled, or scraped at any stage.",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:SoftwareAgent",
+          "@id": "billsim-generator",
+          "prov:label": "BillSim deterministic generator (billsim.generate)",
+          "description": "Python 3.9+ pipeline using NumPy and PyTorch Geometric. Source code released alongside the dataset under Apache-2.0."
+        }
+      ]
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": "Data Annotation (programmatic labeling)",
+      "prov:label": "Anomaly injection and label generation",
+      "description": "Anomaly labels are generated programmatically at injection time, not annotated by humans. The injector (billsim.anomalies.injector) marks each affected node with a binary anomaly indicator (y in {0, 1}) and an integer anomaly type (anomaly_type in {0..6}). Six anomaly types are organized into three categories: structural (Pricing Mismatch, Revenue Recognition, Premature AR Closure), relational (Missing Metering, Orphaned Invoice Lines), and peer-comparison (Stale Discount). Every node has at most one anomaly type, so multi-class labels are mutually exclusive. Anomaly placement is seeded for bit-level reproducibility given (seed, injection_rate). No human annotators participated.",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:SoftwareAgent",
+          "@id": "billsim-injector",
+          "prov:label": "BillSim deterministic anomaly injector (billsim.anomalies.injector)",
+          "description": "Python module that applies one of six parameterized anomaly templates to a configurable fraction of nodes per type."
+        }
+      ]
+    },
+    {
+      "@type": "prov:Activity",
+      "prov:type": "Data Preprocessing",
+      "prov:label": "Feature normalization and train/test split",
+      "description": "Per-node-type z-score normalization (mean 0, std 1) on each numeric feature column. Categorical features (customer segment, payment terms, SKU category, payment method) are integer-encoded with stable per-process hashing. Train/test split is deterministic per seed: static entity types (customer, contract, pricing_rule, sku) use a 75/25 random split; dynamic entity types (usage_record, rated_charge, invoice, line_item, payment, ar_record, gl_entry) use a temporal split (months 1-18 train versus 19-24 test) over the 24-month horizon. No missing values, NaN, Inf, or out-of-bounds edge indices remain after preprocessing; an automated check verifies integrity on every regeneration.",
+      "prov:wasAttributedTo": [
+        {
+          "@type": "prov:SoftwareAgent",
+          "@id": "billsim-preprocessor",
+          "prov:label": "BillSim preprocessor (billsim.graph.builder + billsim.config)",
+          "description": "Python modules invoked by the generator after graph synthesis and before tensor serialization."
+        }
+      ]
+    }
+  ],
   "distribution": [
     {
       "@type": "cr:FileObject",
@@ -80,7 +132,7 @@
       "description": "Default pre-generated dataset (seed=42, injection_rate=0.05). PyTorch Geometric HeteroData object with 11 node types, 14 edge types, z-score-normalized features, binary anomaly labels (y), integer anomaly type labels (anomaly_type in 0-6), and temporal train/test masks.",
       "contentUrl": "data/billsim_default.pt",
       "encodingFormat": "application/x-pytorch",
-      "md5": "00000000000000000000000000000000"
+      "md5": "1aff80c89fdf12d65b098f1a559de174"
     }
   ],
   "recordSet": [