Add prov:wasDerivedFrom + prov:wasGeneratedBy provenance fields

croissant.json

@@ -6,6 +6,7 @@
     "column": "cr:column",
     "conformsTo": "dct:conformsTo",
     "cr": "http://mlcommons.org/croissant/",
+    "prov": "http://www.w3.org/ns/prov#",
     "rai": "http://mlcommons.org/croissant/RAI/",
     "data": {
       "@id": "cr:data",
@@ -76,6 +77,19 @@
   "rai:dataBiases": "All cities are mid-to-large US urban centres with dense building layouts; receiver placements follow DeepMIMO v4 user grids and may oversample regular street patterns. Material RF properties come from a fixed per-material table with no spatial variation within a class.",
   "rai:dataUseCases": "Training and benchmarking ML models for environment-aware multi-path channel prediction, channel charting, site-specific radio resource management, digital-twin radio simulation, and evaluation of generative models for multi-path channels.",
   "rai:dataSocialImpact": "Open release of the dataset and the generation toolchain reduces reliance on proprietary or computationally expensive ray tracers in 5G/6G research and supports reproducibility in environment-aware wireless communications.",
+  "prov:wasDerivedFrom": [
+    {
+      "@id": "https://deepmimo.net/",
+      "name": "DeepMIMO v4",
+      "description": "Building geometry (OpenStreetMap-derived footprints extruded to uniform heights) and TX/RX user grids for 20 US city scenarios at 3.5 GHz."
+    }
+  ],
+  "prov:wasGeneratedBy": {
+    "@type": "prov:Activity",
+    "name": "CityMPC data generation pipeline",
+    "description": "Synthetic data, fully automated, no human annotators. Steps: (1) DeepMIMO v4 scenario files downloaded via scripts/fetch_deepmimo.py. (2) Converted to Mitsuba XML scenes + PLY mesh files via scripts/convert_scenarios.py (scenegen.deepmimo). (3) Per-link ray tracing with Sionna RT at 3.5 GHz computing up to 25 paths (complex coefficient, delay, AoD, AoA per path) for every TX/RX pair (scripts/gen_channels.py). (4) POV image stacks (12 channels: RGB + depth + normals + 5 RF material props) rendered from both TX and RX viewpoints with Mitsuba 3, plus a global building height map centred on each TX-RX midpoint (povgen). (5) Link-level power filter, path-level prune (relative_path_db threshold), and L_max=25 truncation applied (data/filters.py); per-city manifests written by scripts/make_city_manifests.py. (6) Deterministic 80/10/10 train/val/test split (seed 42); norm_stats.json recomputed from the filtered training set. (7) Final HDF5 splits assembled by scripts/bake_dataset.py. Mini bundle (train_2000 / val_500 / test_500): seed-42 random subset of links, baked through the same pipeline. Full code at the citympc repository (link withheld for double-blind).",
+    "prov:wasAssociatedWith": "Anonymous"
+  },
   "distribution": [
     {
       "@type": "cr:FileObject",