Publish dataset version 6.0.0-embedding-lift-controlled

.gitattributes

@@ -61,3 +61,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 evaluation/evaluation_pairs.csv filter=lfs diff=lfs merge=lfs -text
 huggingface/evaluation_pairs/train.csv filter=lfs diff=lfs merge=lfs -text
 hub/evaluation_pairs/train.csv filter=lfs diff=lfs merge=lfs -text
+evaluation/balanced_evaluation_pairs.csv filter=lfs diff=lfs merge=lfs -text
+evaluation/production_like_evaluation_pairs.csv filter=lfs diff=lfs merge=lfs -text

evaluation/balanced_evaluation_pairs.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c861c4756f86cf36d81a7235ffab52e4abd67e11972991f4af0de806cc818b64
+size 10710116

evaluation/production_like_evaluation_pairs.csv

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:560230b89918fa850b9b01bef8da3b2b5a9d055c3f46f0e09c8aacf53387a5c5
+size 16993015

metadata/data_validation_report.json

@@ -103,6 +103,30 @@
     "passed": true,
     "critical": true
   },
+  {
+    "check": "production_like_validation_match_share",
+    "value": 0.05669335979023457,
+    "lower_bound": 0.035,
+    "upper_bound": 0.065,
+    "passed": true,
+    "critical": true
+  },
+  {
+    "check": "production_like_has_both_candidate_routes",
+    "value": 3.0,
+    "lower_bound": 2.0,
+    "upper_bound": NaN,
+    "passed": true,
+    "critical": true
+  },
+  {
+    "check": "production_like_sampling_probability_valid",
+    "value": 1.0,
+    "lower_bound": 1.0,
+    "upper_bound": 1.0,
+    "passed": true,
+    "critical": true
+  },
   {
     "check": "main_negative_hard_share",
     "value": 0.45,

metadata/dataset_manifest.json

@@ -22,6 +22,10 @@
       "sha256": "fdfb6b4b8935329f628855e068243ad550ca15072ed38f8cce644e716e2256af",
       "row_count": 36000
     },
+    "evaluation/balanced_evaluation_pairs.csv": {
+      "sha256": "c861c4756f86cf36d81a7235ffab52e4abd67e11972991f4af0de806cc818b64",
+      "row_count": 57600
+    },
     "evaluation/challenge_evaluation_pairs.csv": {
       "sha256": "2bc9ca74257685335eac570911d8442f5033978262badb19f6a5dc27ac084f14",
       "row_count": 21000
@@ -34,16 +38,20 @@
       "sha256": "3650474c763ab812bcafb28a6f85f8f65cdcc5abe8dd5d6f9c9dea2411410412",
       "row_count": 36000
     },
+    "evaluation/production_like_evaluation_pairs.csv": {
+      "sha256": "560230b89918fa850b9b01bef8da3b2b5a9d055c3f46f0e09c8aacf53387a5c5",
+      "row_count": 68222
+    },
     "evaluation/record_generation_metadata.csv": {
       "sha256": "990c63827e8018c5a5af3b90ade5665080aeaaae79efcd878933e481559aa1c0",
       "row_count": 36000
     },
     "metadata/data_validation_report.json": {
-      "sha256": "6032630cb72344050fa11c7cb5ea2fec3b08cf12d9bb8a92c5b41a6326a2ee1c",
+      "sha256": "a73c292ed0895f0e84650f78b75ae36ba082dfd653e22389408fce88b14ccdd9",
       "row_count": null
     },
     "metadata/dataset_summary.json": {
-      "sha256": "e0a8b2cfa876a89ccdaea50c72863a9f7dfdfba19c90ee545ee68e2a05f253e7",
+      "sha256": "e2969fee2683f5c29c1e77b463c7f087468fb2c7b730d2c7f71ff9b6a1d7d457",
       "row_count": null
     },
     "metadata/em_session_support_report.csv": {
@@ -55,7 +63,7 @@
       "row_count": null
     },
     "metadata/generation_config.json": {
-      "sha256": "026cbc992de32321f19a032afe0bf96e1a774de5503e0058d02206a08fed6a51",
+      "sha256": "6c90affffe509fe16b551eb034b13ba0f98557c2486185880544bbdfd2965093",
       "row_count": null
     },
     "metadata/pipeline_column_contract.json": {

metadata/dataset_summary.json

@@ -30,5 +30,13 @@
   "designed_lexical_positive_share": 0.85,
   "heuristic_lexical_proxy_recall": 0.8436388888888889,
   "hard_negative_count": 16740,
+  "balanced_evaluation_pair_count": 57600,
+  "challenge_evaluation_pair_count": 21000,
+  "production_like_evaluation_pair_count": 68222,
+  "production_like_match_share_by_split": {
+    "test": 0.05669335979023457,
+    "train": 0.08,
+    "validation": 0.05669335979023457
+  },
   "validation_passed": true
 }

metadata/generation_config.json

@@ -9,7 +9,7 @@
   },
   "semantic_entity_share": 0.225,
   "target_lexical_positive_recall": 0.85,
-  "evaluation_strategy": "entity-level 70/15/15; balanced main benchmark; sealed test; challenge set separate",
+  "evaluation_strategy": "entity-level 70/15/15; all-positive blocking truth; balanced model benchmark; challenge stress test; production-like threshold benchmark; sealed test",
   "um_training_strategy": "three sources/entity; stratified lexical variants; controlled tax/name/address/representative collisions; random, lexical-hard and semantic-hard nonmatches",
   "semantic_design": "inspection aliases paraphrase or translate brand, modifier, region and industry; semantic hard negatives share only industry and region"
 }