Add CODE dataset: 5,233 image-text drawing samples and reference DXFs

Dataset-only upload. Preserve frozen training/validation/test records and all existing GUI files. No construction, training, evaluation scripts or model weights included.

.gitattributes

@@ -58,3 +58,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+CODE/sft_train.jsonl filter=lfs diff=lfs merge=lfs -text

CODE/README.txt

@@ -0,0 +1,24 @@
+CODE DATASET
+
+5,233 unique mechanical parts, with isometric images, text instructions, target drawing programs, and reference DXF drawings.
+
+SPLITS
+sft_train.jsonl: 3,929 training samples.
+sft_val.jsonl: all 256 original validation samples.
+sft_val_effective254.jsonl: the existing formal 254-sample validation subset, not additional data.
+test.jsonl: all 1,048 original test samples; 1,045 are eligible under the frozen GT-only mask.
+
+ASSETS
+Extract images.zip and reference_dxf.zip into this CODE directory. Each ZIP contains its named subdirectory; JSONL image paths then resolve without edits.
+images/: 5,233 PNG input images. reference_dxf/: 5,233 existing DXF ground-truth drawings.
+
+METADATA
+sample_index.csv gives actual SFT splits, original source splits, categories, historical tiers, asset paths and exclusion flags.
+labels.csv is preserved byte-for-byte from the source and labels the combined 4,185 train/validation pool as train.
+evaluation_exclusions.json preserves the established validation/test masks; no original samples have been removed.
+asset_manifest.jsonl gives byte counts and SHA-256 for every archived asset. SHA256SUMS verifies uploaded files.
+dataset_manifest.json records counts, source versions and unchanged source-file hashes.
+
+SCOPE
+Only the dataset is uploaded. Assistant Python/ezdxf programs remain inside JSONL as supervision targets. No construction, training or evaluation scripts, model weights, credentials, or experiment logs are included.
+Training/validation target programs use the final compact representation; test targets retain their original representation, matching the existing experiment data.

CODE/SHA256SUMS

@@ -0,0 +1,12 @@
+2b085780d75da28d76b6a4c30b0a0372f82c61158beb603b2d6544b1663bfc27  README.txt
+de004145a4cdded83dc5cb45fd335c544a095a70628cbafea858836524f30239  asset_manifest.jsonl
+f477ccb7c7af86f995f13fa76dcae3c5fb67b61bfe44019ba5df454551516f4a  dataset_manifest.json
+9acc75711ceedfa93ba8d21f217aa072163e2d6d4390cae6fc0036a28f849e68  evaluation_exclusions.json
+82d52040bce43fa5c154f9e59e71bdc35e1ba2b0daec39ec6c18fa44ced45f2b  images.zip
+76769b9e25d5be6056b6f14584984cbed9bb14b4909c0f0cb1bd8ac4109fcccf  labels.csv
+0c0acf4a565a87716642c3ce646c3f368b65954f8e6755938c0647c8cca51c2f  reference_dxf.zip
+44ea9079a2a7007629f21c40e9eb3104a7d7cbae6ca9b1e62695df78e04c54c8  sample_index.csv
+47623d3e972e1367f8050ccd1f97996482d4660d40bc8659b0f399a47577fc4b  sft_train.jsonl
+c21acdb39197b68a8066b8b5354f2a1c28645eb1bba9bbebd95be06942cd2a58  sft_val.jsonl
+4876a34eb54213ff27e9eb7f4c4e7e66ce76adc6db24b75c149031be72e41020  sft_val_effective254.jsonl
+8e606cf133e8a2c1850873c0880c48b9e1c9815b00537c8cc898bf62f607567f  test.jsonl

CODE/dataset_manifest.json

@@ -0,0 +1,67 @@
+{
+  "dataset": "CODE",
+  "release_date": "2026-11-05",
+  "unique_samples": 5233,
+  "source_version": "physical-v1-direct-compact-v1 for train/validation; frozen original test1048",
+  "source_origin": "GenCAD-derived CODE engineering-drawing dataset",
+  "splits": {
+    "train": 3929,
+    "validation_original": 256,
+    "validation_effective": 254,
+    "test_original": 1048,
+    "test_effective": 1045
+  },
+  "categories": {
+    "DISK": 560,
+    "CONVEX_PLATE": 1801,
+    "POLY_WITH_HOLES": 427,
+    "POLY_PRISM": 277,
+    "DISK_WITH_HOLES": 504,
+    "CURVED_PLATE": 644,
+    "ARC_COMPLEX": 381,
+    "COAXIAL_STACK": 235,
+    "RECTILINEAR_PLATE": 404
+  },
+  "sorted_unique_ids_sha256": "45a92de4e27f13d3ee5aa93610eb7fb61333295f1242b87813a17320b4134689",
+  "records": {
+    "sft_train.jsonl": {
+      "rows": 3929,
+      "sha256": "47623d3e972e1367f8050ccd1f97996482d4660d40bc8659b0f399a47577fc4b",
+      "source_relative_path": "code_datasets/physical_v1_direct_compact_v1/sft_train.jsonl"
+    },
+    "sft_val.jsonl": {
+      "rows": 256,
+      "sha256": "c21acdb39197b68a8066b8b5354f2a1c28645eb1bba9bbebd95be06942cd2a58",
+      "source_relative_path": "code_datasets/physical_v1_direct_compact_v1/sft_val.jsonl"
+    },
+    "sft_val_effective254.jsonl": {
+      "rows": 254,
+      "sha256": "4876a34eb54213ff27e9eb7f4c4e7e66ce76adc6db24b75c149031be72e41020",
+      "source_relative_path": "code_datasets/physical_v1_direct_compact_v1/sft_val_effective254.jsonl"
+    },
+    "test.jsonl": {
+      "rows": 1048,
+      "sha256": "8e606cf133e8a2c1850873c0880c48b9e1c9815b00537c8cc898bf62f607567f",
+      "source_relative_path": "code_datasets/test.jsonl"
+    }
+  },
+  "assets": {
+    "images.zip": {
+      "members": 5233,
+      "extract_under": "CODE/"
+    },
+    "reference_dxf.zip": {
+      "members": 5233,
+      "extract_under": "CODE/"
+    }
+  },
+  "schema": {
+    "id": "Stable part identifier",
+    "images": "Image paths relative to CODE/ after extracting images.zip",
+    "messages": "User image/text instruction and assistant ground-truth drawing program"
+  },
+  "code_scope": "Assistant drawing programs are supervised data labels. No pipeline, training, evaluation, inference scripts or model checkpoints are included.",
+  "no_execution": "No dataset target program was executed during packaging.",
+  "labels_csv_note": "Unmodified legacy labels.csv uses train for the original 4185 training pool; sample_index.csv gives the actual SFT train/validation/test split.",
+  "license_note": "This upload does not introduce or change a license; source-data and repository license terms remain applicable."
+}

CODE/evaluation_exclusions.json

@@ -0,0 +1,38 @@
+{
+  "validation": {
+    "raw_count": 256,
+    "effective_count": 254,
+    "excluded": [
+      {
+        "id": "0092_00927567",
+        "reasons": [
+          "gt_contains_two_outline_views_only",
+          "third_outline_view_missing",
+          "orphan_native_dimension",
+          "sidecar_reference_view_count_mismatch",
+          "prompt_negative_extrusion_depth"
+        ]
+      },
+      {
+        "id": "0093_00937357",
+        "reasons": [
+          "gt_contains_two_outline_views_only",
+          "third_outline_view_missing",
+          "orphan_native_dimension",
+          "sidecar_reference_view_count_mismatch"
+        ]
+      }
+    ]
+  },
+  "test": {
+    "raw_count": 1048,
+    "effective_count": 1045,
+    "excluded_ids": [
+      "0042_00422612",
+      "0051_00511376",
+      "0086_00862775"
+    ],
+    "reason": "view_split_not_found_or_target_semantic_partition_ambiguous"
+  },
+  "policy": "Preserve all original records and assets. Exclusions are evaluation masks, not deleted samples. Model failures are not dataset exclusions."
+}

CODE/images.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:82d52040bce43fa5c154f9e59e71bdc35e1ba2b0daec39ec6c18fa44ced45f2b
+size 20814984

CODE/reference_dxf.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c0acf4a565a87716642c3ce646c3f368b65954f8e6755938c0647c8cca51c2f
+size 44941589

CODE/sft_train.jsonl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47623d3e972e1367f8050ccd1f97996482d4660d40bc8659b0f399a47577fc4b
+size 19472674