Replace mislabeled v1 vectors with static shape embedding v0

README.md

@@ -1,10 +1,10 @@
 ---
 license: mit
 configs:
-- config_name: geometry-vector-v1
+- config_name: static-embedding-v0
   data_files:
   - split: train
-    path: "metadata/geometry-vector-v1.parquet"
+    path: "metadata/static-embedding-v0.parquet"
 tags:
 - quantum
 - superconducting
@@ -13,6 +13,7 @@ tags:
 - layout
 - embedding
 - geometry
+- computer-vision
 pretty_name: SQuADDS Layout Embeddings
 size_categories:
 - 1K<n<10K
@@ -24,32 +25,74 @@ size_categories:
 
 # SQuADDS Layout Embeddings
 
-Versioned, reproducible geometry vectors for layouts in
+Versioned layout representations for the 4,577 GDS artifacts in
 [SQuADDS/SQuADDS_Layouts](https://huggingface.co/datasets/SQuADDS/SQuADDS_Layouts).
 
-## `geometry-vector-v1`
+## Static embedding model `v0`
 
-This release provides 4,577 embedding vectors (generated through the v0 algorithm), one per layout in the
-paired layout registry.
+`static-embedding-v0` implements the original SQuADDS proof-of-concept model:
 
-The exact feature order, layer vocabulary, normalization rule, and source
-feature-schema version are frozen in `metadata/geometry-vector-v1.schema.json`. 
-Every record retains `layout_id`, `artifact_id`, `design_id`, `component_name`,
-and `source_id` to link it directly to the simulation database and checksum
-verified GDS artifact.
+```text
+v0 = parameter_sum + geometric_moments + flattened_shape_bitmap
+```
+
+Each unit-normalized vector has **9,227 dimensions**:
+
+| Block | Dimensions | Contents |
+| --- | ---: | --- |
+| Parameter sum | 1 | Permutation- and parameter-count-invariant sum of numerical design options, converted to micrometers where units are present |
+| Geometric moments | 10 | Functional area, perimeter, aspect ratio, occupancy, centroid, second central moments, and eccentricity |
+| Shape tensor | 9,216 | Row-major flattened 96×96 signed bitmap of functional GDS geometry |
+
+The bitmap uses `+1` for conductor, `-1` for etch, `+0.5` for explicit port
+geometry, and `0` for background. Geometry is cropped to its functional bounds,
+centered without distortion, supersampled at 4×, and reduced to 96×96. The
+large simulation-domain ground rectangle on layer `(1, 0)` is excluded so it
+does not hide the component shape.
+
+This is a deterministic static embedding, not a learned model. It is intended
+as a transparent baseline and as a stable input for similarity search. A future
+`v1` may use a compact learned raster or geometry-graph encoder, but will be
+published as a separate model rather than changing `v0` in place.
+
+The exact block offsets, moment order, raster semantics, normalization
+statistics, and source schema are frozen in
+`metadata/static-embedding-v0.schema.json`.
+
+## Coverage and links
+
+| Component | Embeddings |
+| --- | ---: |
+| `GeneralizedCapNInterdigital` | 3,683 |
+| `CapNInterdigitalTee` | 894 |
+
+Every row retains `layout_id`, `artifact_id`, `design_id`, `component_name`,
+and `source_id`, plus the raw parameter sum, geometric moments, functional
+bounds, and a SHA-256 hash of the 96×96 bitmap.
 
 ## Access
 
 ```python
-from squadds.layouts import GeometryEmbeddingClient
+from squadds.layouts import StaticEmbeddingClient
 
-embeddings = GeometryEmbeddingClient()
-record = embeddings.get("layout:sha256:<layout hash>")
-neighbors = embeddings.nearest(record["layout_id"], limit=10)
+client = StaticEmbeddingClient()
+record = client.get("layout:sha256:<layout hash>")
+bitmap = client.shape_bitmap(record["layout_id"])
+neighbors = client.nearest(record["layout_id"], limit=10)
 ```
 
-For raw geometry, simulation provenance, and numerical geometry features, use
-the [SQuADDS Layouts dataset](https://huggingface.co/datasets/SQuADDS/SQuADDS_Layouts).
+SQuADDS_DB rows can resolve the same vector directly with
+`SQuADDS_DB.get_layout_embedding(row)`. The SQuADDS MCP server also provides
+`get_layout_embedding` and `find_similar_layouts`.
+
+## Provenance
+
+Raw GDS artifacts, layer semantics, checksums, and geometry features live in
+[SQuADDS/SQuADDS_Layouts](https://huggingface.co/datasets/SQuADDS/SQuADDS_Layouts).
+Simulation results and design options live in
+[SQuADDS/SQuADDS_DB](https://huggingface.co/datasets/SQuADDS/SQuADDS_DB).
+The generalized-capacitor dataset was contributed by Saikat Das of the
+Levenson-Falk Lab at USC.
 
 ## Citation
 

metadata/geometry-vector-v1.schema.json

@@ -1,53 +0,0 @@
-{
-  "geometry_vector_schema_version": "1.0.0",
-  "source_geometry_feature_schema_version": "1.0.0",
-  "normalization": "L2 unit norm; raw dimensions use log1p except log aspect ratio and layer fractions.",
-  "layer_vocabulary": [
-    {
-      "layer": 1,
-      "datatype": 0
-    },
-    {
-      "layer": 1,
-      "datatype": 10
-    },
-    {
-      "layer": 1,
-      "datatype": 11
-    },
-    {
-      "layer": 2,
-      "datatype": 0
-    },
-    {
-      "layer": 3,
-      "datatype": 0
-    }
-  ],
-  "feature_names": [
-    "log1p_bbox_width_um",
-    "log1p_bbox_height_um",
-    "log1p_bbox_area_um2",
-    "log_bbox_aspect_ratio",
-    "log1p_total_area_um2",
-    "log1p_polygon_count",
-    "log1p_cell_count",
-    "log1p_layer_count",
-    "layer_1_datatype_0_present",
-    "layer_1_datatype_0_area_fraction",
-    "layer_1_datatype_0_polygon_fraction",
-    "layer_1_datatype_10_present",
-    "layer_1_datatype_10_area_fraction",
-    "layer_1_datatype_10_polygon_fraction",
-    "layer_1_datatype_11_present",
-    "layer_1_datatype_11_area_fraction",
-    "layer_1_datatype_11_polygon_fraction",
-    "layer_2_datatype_0_present",
-    "layer_2_datatype_0_area_fraction",
-    "layer_2_datatype_0_polygon_fraction",
-    "layer_3_datatype_0_present",
-    "layer_3_datatype_0_area_fraction",
-    "layer_3_datatype_0_polygon_fraction"
-  ],
-  "dimensions": 23
-}

metadata/{geometry-vector-v1.parquet → static-embedding-v0.parquet}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ff9fc02154da56a58095af4100f48c4557d9f17aa05d9bbd13cffeecb48907ab
-size 1614070
+oid sha256:eee64ef7382c0fbe6cb9d763f848e1d482790c746429570810a501b94dd86fb7
+size 27206661

metadata/static-embedding-v0.schema.json

@@ -0,0 +1,83 @@
+{
+  "model": "static-shape-v0",
+  "embedding_schema_version": "0.1.0",
+  "dimensions": 9227,
+  "blocks": {
+    "parameter_sum": {
+      "offset": 0,
+      "dimensions": 1
+    },
+    "geometric_moments": {
+      "offset": 1,
+      "dimensions": 10,
+      "names": [
+        "log1p_functional_area_um2",
+        "log1p_functional_perimeter_um",
+        "log_functional_bbox_aspect_ratio",
+        "bitmap_occupancy_fraction",
+        "bitmap_centroid_x",
+        "bitmap_centroid_y",
+        "bitmap_mu20",
+        "bitmap_mu02",
+        "bitmap_mu11",
+        "bitmap_eccentricity"
+      ]
+    },
+    "shape_bitmap": {
+      "offset": 11,
+      "dimensions": 9216,
+      "shape": [
+        96,
+        96
+      ],
+      "flatten_order": "row-major"
+    }
+  },
+  "shape_rasterization": {
+    "resolution": [
+      96,
+      96
+    ],
+    "crop": "functional geometry bounds, aspect preserving, centered with 4-pixel margin",
+    "supersampling": 4,
+    "values": {
+      "conductor": 1.0,
+      "etch": -1.0,
+      "port": 0.5,
+      "background": 0.0
+    },
+    "excluded": "simulation-domain ground layer (1, 0)"
+  },
+  "normalization": {
+    "parameter_mean": 119.12693904304129,
+    "parameter_std": 41.47948098628826,
+    "moment_mean": [
+      7.199095883714775,
+      6.038557291097784,
+      -0.06114899487646157,
+      0.3444617315897131,
+      0.027325720598655186,
+      0.001453481663011831,
+      0.04410981652657115,
+      0.03196827568718987,
+      -0.00023640558809321613,
+      0.7251485793088448
+    ],
+    "moment_std": [
+      0.9058000344121664,
+      0.6270697679224184,
+      0.7402350020741016,
+      0.1290950058579193,
+      0.05763537913270997,
+      0.0021810992553648666,
+      0.02303165066879144,
+      0.014210240903499162,
+      0.00040800322351881635,
+      0.2101244913274543
+    ],
+    "parameter": "tanh(z-score)",
+    "moments": "z-score, clipped to [-5, 5], then L2 normalized",
+    "shape": "signed bitmap, then L2 normalized",
+    "embedding": "concatenated blocks, then L2 normalized"
+  }
+}