Update main model with joint JSON extraction SFT and verified results

CITATION.cff

@@ -4,8 +4,8 @@ title: "OpenSML-JSONLite-175M: Native JSON Generation from Source Text and Schem
 authors:
   - family-names: Zebrowski
     given-names: William
-version: 0.3.0
-date-released: 2026-10-14
+version: 0.4.0
+date-released: 2026-10-16
 url: "https://github.com/williamzebrowskI/OpenSML-JSONLite-175M"
 repository-code: "https://github.com/williamzebrowskI/OpenSML-JSONLite-175M"
 license: Apache-2.0

INFERENCE_VERIFICATION.json

@@ -1,35 +1,37 @@
 {
   "model": "OpenSML-JSONLite-175M-Instruct",
   "checkpoint": "main",
-  "weight_sha256": "655fb5cfba009163cd8a7f7fe7b170b998bb452c217b1b473cfe10b33de0ccd1",
-  "requests": 54,
-  "raw_text_and_stop_matches": 54,
-  "correct_against_reference": 44,
-  "schema_valid": 54,
+  "weight_sha256": "781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560",
+  "requests": 60,
+  "raw_text_and_stop_matches": 60,
+  "correct_against_reference": 49,
+  "schema_valid": 60,
   "versions": {
     "mlx": "0.32.3",
     "numpy": "2.4.2",
     "tokenizers": "0.22.2",
     "jsonschema": "4.26.0"
   },
-  "platform": "Apple Silicon macOS, Python3.11",
+  "platform": "Apple Silicon macOS, Python 3.11",
   "outputs": [
     {
-      "text": "{\"playlist_owner\":[],\"playlist\":[\"Rock Gaming\"],\"entity_name\":[\"Mariem Hassan\"]}",
+      "text": "{\"playlist_owner\":[\"my\"],\"playlist\":[\"Rock Gaming\"],\"entity_name\":[\"Mariem Hassan con Leyoad\"]}",
       "value": {
-        "playlist_owner": [],
+        "playlist_owner": [
+          "my"
+        ],
         "playlist": [
           "Rock Gaming"
         ],
         "entity_name": [
-          "Mariem Hassan"
+          "Mariem Hassan con Leyoad"
         ]
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
-      "output_tokens": 37,
-      "seconds": 0.1317074578255415,
+      "output_tokens": 44,
+      "seconds": 0.17647575004957616,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -42,28 +44,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-01",
-      "checks": {
-        "exact": 0,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 1,
-        "fields": 3,
-        "tp": 1,
-        "fp": 1,
-        "fn": 2,
-        "present_correct": 1,
-        "present_total": 3,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 1,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"artist\":[\"Mariem Hassan\"],\"music_item\":[]}",
@@ -77,7 +59,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 23,
-      "seconds": 0.09170329081825912,
+      "seconds": 0.09266262501478195,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -90,28 +72,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-02",
-      "checks": {
-        "exact": 0,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 1,
-        "fields": 2,
-        "tp": 0,
-        "fp": 1,
-        "fn": 0,
-        "present_correct": 0,
-        "present_total": 0,
-        "absent_correct": 1,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 1,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": false
     },
     {
       "text": "{\"playlist_owner\":[\"my\"]}",
@@ -124,7 +86,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 13,
-      "seconds": 0.052787666907534,
+      "seconds": 0.05713866697624326,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -137,28 +99,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-03",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 1,
-        "fields": 1,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"artist\":[],\"entity_name\":[],\"playlist\":[\"Digster Future Hits\"]}",
@@ -173,7 +115,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 29,
-      "seconds": 0.09026900003664196,
+      "seconds": 0.11768570798449218,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -186,47 +128,27 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-04",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
-      "text": "{\"state\":[\"AR\"],\"party_size_number\":[\"nine\"],\"restaurant_type\":[\"brasserie\"]}",
+      "text": "{\"restaurant_type\":[\"brasserie\"],\"party_size_number\":[\"nine\"],\"state\":[\"AR\"]}",
       "value": {
-        "state": [
-          "AR"
+        "restaurant_type": [
+          "brasserie"
         ],
         "party_size_number": [
           "nine"
         ],
-        "restaurant_type": [
-          "brasserie"
+        "state": [
+          "AR"
         ]
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 36,
-      "seconds": 0.13286249991506338,
+      "seconds": 0.13061708281747997,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -239,28 +161,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-05",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 3,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 3,
-        "present_total": 3,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 3,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"party_size_number\":[\"nine\"],\"timeRange\":[],\"sort\":[]}",
@@ -275,7 +177,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 27,
-      "seconds": 0.11090816603973508,
+      "seconds": 0.06810929113999009,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -288,44 +190,24 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-06",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
-      "text": "{\"city\":[\"Wrigley\"],\"sort\":[\"best\"]}",
+      "text": "{\"sort\":[\"best\"],\"city\":[\"Wrigley\"]}",
       "value": {
-        "city": [
-          "Wrigley"
-        ],
         "sort": [
           "best"
+        ],
+        "city": [
+          "Wrigley"
         ]
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 21,
-      "seconds": 0.08733266708441079,
+      "seconds": 0.05377487512305379,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -338,40 +220,20 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-07",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
-      "text": "{\"poi\":[],\"timeRange\":[]}",
+      "text": "{\"timeRange\":[],\"poi\":[]}",
       "value": {
-        "poi": [],
-        "timeRange": []
+        "timeRange": [],
+        "poi": []
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 15,
-      "seconds": 0.06839366699568927,
+      "seconds": 0.04086962481960654,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -384,28 +246,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-08",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 0,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 0,
-        "present_total": 0,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 0,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"state\":[\"Louisiana\"],\"condition_temperature\":[\"colder\"]}",
@@ -421,7 +263,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 23,
-      "seconds": 0.09773133299313486,
+      "seconds": 0.05825637513771653,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -434,28 +276,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-09",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"current_location\":[],\"geographic_poi\":[],\"condition_temperature\":[\"colder\"]}",
@@ -470,7 +292,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 31,
-      "seconds": 0.12699225009419024,
+      "seconds": 0.11788337491452694,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -483,28 +305,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-10",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"timeRange\":[\"last August\"],\"geographic_poi\":[\"Stunt Ranch Santa Monica Mountains Reserve\"],\"condition_temperature\":[\"colder\"]}",
@@ -523,7 +325,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 44,
-      "seconds": 0.16674145800061524,
+      "seconds": 0.17240966693498194,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -536,43 +338,23 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-11",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 3,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 3,
-        "present_total": 3,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 3,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
-      "text": "{\"state\":[],\"geographic_poi\":[\"Stunt Ranch Santa Monica Mountains Reserve\"],\"current_location\":[]}",
+      "text": "{\"current_location\":[],\"geographic_poi\":[\"Stunt Ranch Santa Monica Mountains Reserve\"],\"state\":[]}",
       "value": {
-        "state": [],
+        "current_location": [],
         "geographic_poi": [
           "Stunt Ranch Santa Monica Mountains Reserve"
         ],
-        "current_location": []
+        "state": []
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 34,
-      "seconds": 0.09643245814368129,
+      "seconds": 0.1316559170372784,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -585,28 +367,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-12",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"artist\":[\"Behzad Mirkhani\"],\"music_item\":[\"theme\"],\"sort\":[\"last\"]}",
@@ -625,7 +387,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 38,
-      "seconds": 0.0911900419741869,
+      "seconds": 0.11877575004473329,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -638,43 +400,23 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-13",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 3,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 3,
-        "present_total": 3,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 3,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
-      "text": "{\"playlist\":[],\"music_item\":[\"theme\"],\"service\":[]}",
+      "text": "{\"service\":[],\"music_item\":[\"theme\"],\"playlist\":[]}",
       "value": {
-        "playlist": [],
+        "service": [],
         "music_item": [
           "theme"
         ],
-        "service": []
+        "playlist": []
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 25,
-      "seconds": 0.07062204205431044,
+      "seconds": 0.06615762482397258,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -687,28 +429,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-14",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"artist\":[\"Yui\"],\"music_item\":[\"track\"],\"service\":[\"Vimeo\"]}",
@@ -727,7 +449,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 33,
-      "seconds": 0.1308659170754254,
+      "seconds": 0.08466579113155603,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -740,28 +462,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-15",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 3,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 3,
-        "present_total": 3,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 3,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"playlist\":[],\"sort\":[]}",
@@ -773,7 +475,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 14,
-      "seconds": 0.05398241593502462,
+      "seconds": 0.039065124932676554,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -786,28 +488,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-16",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 0,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 0,
-        "present_total": 0,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 0,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"rating_value\":[\"3\"],\"rating_unit\":[\"stars\"],\"object_name\":[\"Things Fall Apart\"]}",
@@ -826,7 +508,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 35,
-      "seconds": 0.1139006249140948,
+      "seconds": 0.08983045816421509,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -839,28 +521,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-17",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 3,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 3,
-        "present_total": 3,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 3,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"object_name\":[\"Things Fall Apart\"],\"object_select\":[],\"object_type\":[]}",
@@ -875,7 +537,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 29,
-      "seconds": 0.08503687498159707,
+      "seconds": 0.07524858298711479,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -888,28 +550,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-18",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"rating_value\":[\"1\"],\"object_name\":[]}",
@@ -923,7 +565,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 19,
-      "seconds": 0.060387207893654704,
+      "seconds": 0.049742958042770624,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -936,28 +578,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-19",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 1,
-        "absent_total": 1,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"object_part_of_series_type\":[],\"object_name\":[]}",
@@ -969,7 +591,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 23,
-      "seconds": 0.09527637506835163,
+      "seconds": 0.06354975001886487,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -982,28 +604,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-20",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 0,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 0,
-        "present_total": 0,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 0,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"object_name\":[\"Happiness Was Free\"]}",
@@ -1016,7 +618,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 16,
-      "seconds": 0.05649704090319574,
+      "seconds": 0.060178708052262664,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1029,28 +631,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-21",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 1,
-        "fields": 1,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"object_type\":[],\"object_name\":[\"Happiness Was Free\"]}",
@@ -1064,7 +646,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 23,
-      "seconds": 0.059290583012625575,
+      "seconds": 0.0893012909218669,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1077,28 +659,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-22",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 1,
-        "absent_total": 1,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"object_name\":[\"Martin Morning\"],\"object_type\":[\"saga\"]}",
@@ -1114,7 +676,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 24,
-      "seconds": 0.0722098748665303,
+      "seconds": 0.07322416710667312,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1127,28 +689,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-23",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"object_name\":[\"Martin Morning\"]}",
@@ -1161,7 +703,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 13,
-      "seconds": 0.056712874909862876,
+      "seconds": 0.040689832996577024,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1174,28 +716,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-24",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 1,
-        "fields": 1,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"movie_type\":[\"animated movies\"],\"spatial_relation\":[\"in the neighborhood\"]}",
@@ -1211,7 +733,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 29,
-      "seconds": 0.11536691710352898,
+      "seconds": 0.08523791679181159,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1224,43 +746,23 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-25",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
-      "text": "{\"movie_name\":[],\"timeRange\":[],\"movie_type\":[\"animated movies\"]}",
+      "text": "{\"movie_type\":[\"animated movies\"],\"timeRange\":[],\"movie_name\":[]}",
       "value": {
-        "movie_name": [],
-        "timeRange": [],
         "movie_type": [
           "animated movies"
-        ]
+        ],
+        "timeRange": [],
+        "movie_name": []
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 32,
-      "seconds": 0.12519749999046326,
+      "seconds": 0.09332416695542634,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1273,44 +775,24 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-26",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
-      "text": "{\"timeRange\":[\"twenty two o'clock\"],\"movie_name\":[\"Mabel’s Dramatic Career\"]}",
+      "text": "{\"movie_name\":[\"Mabel’s Dramatic Career\"],\"timeRange\":[\"twenty two o'clock\"]}",
       "value": {
-        "timeRange": [
-          "twenty two o'clock"
-        ],
         "movie_name": [
           "Mabel’s Dramatic Career"
+        ],
+        "timeRange": [
+          "twenty two o'clock"
         ]
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 36,
-      "seconds": 0.1416236248333007,
+      "seconds": 0.1403042080346495,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1323,28 +805,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-27",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"timeRange\":[\"twenty two o'clock\"],\"spatial_relation\":[],\"object_location_type\":[]}",
@@ -1359,7 +821,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 35,
-      "seconds": 0.14057387504726648,
+      "seconds": 0.1383882921654731,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1372,28 +834,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-28",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"track\":[\"Blue Lantern\"],\"artist\":[\"Mira Sol\"],\"album\":[]}",
@@ -1410,7 +852,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 29,
-      "seconds": 0.11759562487713993,
+      "seconds": 0.11747579206712544,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1423,28 +865,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-29",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 1,
-        "absent_total": 1,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"restaurant_name\":[\"River Moon\"],\"city\":[\"Denver\"],\"party_size_number\":[\"four\"],\"timeRange\":[\"tomorrow\"]}",
@@ -1466,7 +888,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 49,
-      "seconds": 0.17598291696049273,
+      "seconds": 0.18146875011734664,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1479,28 +901,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-30",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 4,
-        "fields": 4,
-        "tp": 4,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 4,
-        "present_total": 4,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 4,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"city\":[\"Ose\"],\"timeRange\":[\"Friday\"],\"country\":[]}",
@@ -1517,7 +919,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 26,
-      "seconds": 0.0654072081670165,
+      "seconds": 0.10596829210408032,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1530,28 +932,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-31",
-      "checks": {
-        "exact": 0,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 3,
-        "tp": 1,
-        "fp": 1,
-        "fn": 1,
-        "present_correct": 1,
-        "present_total": 2,
-        "absent_correct": 1,
-        "absent_total": 1,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 1,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": false
     },
     {
       "text": "{\"object_name\":[\"Glass Harbor\"],\"rating_value\":[\"four\"],\"best_rating\":[],\"rating_unit\":[\"stars\"]}",
@@ -1571,7 +953,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 42,
-      "seconds": 0.11358241690322757,
+      "seconds": 0.16606529196724296,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1584,45 +966,25 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-32",
-      "checks": {
-        "exact": 0,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 4,
-        "tp": 3,
-        "fp": 0,
-        "fn": 1,
-        "present_correct": 3,
-        "present_total": 4,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 1,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 3,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": false
     },
     {
-      "text": "{\"timeRange\":[],\"location_name\":[\"Maple Cinema\"],\"movie_name\":[\"Silver Orchard\"]}",
+      "text": "{\"movie_name\":[\"Silver Orchard\"],\"location_name\":[\"Maple Cinema\"],\"timeRange\":[]}",
       "value": {
-        "timeRange": [],
+        "movie_name": [
+          "Silver Orchard"
+        ],
         "location_name": [
           "Maple Cinema"
         ],
-        "movie_name": [
-          "Silver Orchard"
-        ]
+        "timeRange": []
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 37,
-      "seconds": 0.10278037493117154,
+      "seconds": 0.09538087481632829,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1635,28 +997,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-33",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 1,
-        "absent_total": 1,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"entity_name\":[\"Night Steps\"],\"playlist\":[\"Evening Walks\"]}",
@@ -1672,7 +1014,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 27,
-      "seconds": 0.08685358287766576,
+      "seconds": 0.07212029211223125,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1685,28 +1027,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-34",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"artist\":[\"Mira Sol\",\"Theo Lane\"],\"album\":[]}",
@@ -1721,7 +1043,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 24,
-      "seconds": 0.10270537505857646,
+      "seconds": 0.06363512482494116,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1734,28 +1056,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-35",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 1,
-        "absent_total": 1,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"city\":[\"Bergen\",\"Orange\",\"Lake\"],\"country\":[]}",
@@ -1771,7 +1073,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 24,
-      "seconds": 0.09942620806396008,
+      "seconds": 0.06020650011487305,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1784,28 +1086,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-36",
-      "checks": {
-        "exact": 0,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 1,
-        "fields": 2,
-        "tp": 1,
-        "fp": 2,
-        "fn": 1,
-        "present_correct": 0,
-        "present_total": 1,
-        "absent_correct": 1,
-        "absent_total": 1,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 2,
-        "predicted_values": 3,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": false
     },
     {
       "text": "{\"cuisine\":[],\"city\":[\"Boston\"],\"restaurant_name\":[]}",
@@ -1820,7 +1102,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 26,
-      "seconds": 0.1082186670973897,
+      "seconds": 0.09844033396802843,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1833,45 +1115,25 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-37",
-      "checks": {
-        "exact": 0,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 1,
-        "present_correct": 1,
-        "present_total": 2,
-        "absent_correct": 1,
-        "absent_total": 1,
-        "false_abstention": 1,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": false
     },
     {
-      "text": "{\"track\":[],\"artist\":[\"Mira Sol\"],\"album\":[\"Quiet Signals\"]}",
+      "text": "{\"album\":[\"Quiet Signals\"],\"artist\":[\"Mira Sol\"],\"track\":[]}",
       "value": {
-        "track": [],
+        "album": [
+          "Quiet Signals"
+        ],
         "artist": [
           "Mira Sol"
         ],
-        "album": [
-          "Quiet Signals"
-        ]
+        "track": []
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 30,
-      "seconds": 0.12260229187086225,
+      "seconds": 0.12414183304645121,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1884,28 +1146,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-38",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 1,
-        "absent_total": 1,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"object_name\":[\"Paper Bridge\"],\"rating_value\":[]}",
@@ -1919,7 +1161,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 20,
-      "seconds": 0.0860819579102099,
+      "seconds": 0.08667670795693994,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1932,41 +1174,21 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-39",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 1,
-        "absent_total": 1,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
-      "text": "{\"artist\":[],\"timeRange\":[],\"country\":[]}",
+      "text": "{\"country\":[],\"timeRange\":[],\"artist\":[]}",
       "value": {
-        "artist": [],
+        "country": [],
         "timeRange": [],
-        "country": []
+        "artist": []
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 20,
-      "seconds": 0.08448487496934831,
+      "seconds": 0.08656954090110958,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -1979,43 +1201,25 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "FIELD-40",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 0,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 0,
-        "present_total": 0,
-        "absent_correct": 3,
-        "absent_total": 3,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 0,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
-      "text": "{\"selected_playlist\":[\"night vibes\"],\"selected_artist\":[],\"selected_playlist_owner\":[]}",
+      "text": "{\"selected_playlist\":[\"night vibes\"],\"selected_artist\":[\"gn\"],\"selected_playlist_owner\":[]}",
       "value": {
         "selected_playlist": [
           "night vibes"
         ],
-        "selected_artist": [],
+        "selected_artist": [
+          "gn"
+        ],
         "selected_playlist_owner": []
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
-      "output_tokens": 36,
-      "seconds": 0.14466900005936623,
+      "output_tokens": 39,
+      "seconds": 0.1672995421104133,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2028,28 +1232,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:AddToPlaylist:train:80:view1:native:alias",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": false
     },
     {
       "text": "{\"playlist_owner\":null,\"artist\":\"gn\",\"playlist\":\"night vibes\"}",
@@ -2062,7 +1246,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 26,
-      "seconds": 0.09343320806510746,
+      "seconds": 0.10970429191365838,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2075,35 +1259,13 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:AddToPlaylist:train:80:view1:native:nullable",
-      "checks": {
-        "exact": 0,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 3,
-        "tp": 1,
-        "fp": 1,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 1,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 1,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": false
     },
     {
-      "text": "{\"selected_restaurant_name\":[\"mughlai restaurant\"],\"selected_timeRange\":[\"Nov. the 1st\"],\"selected_sort\":[]}",
+      "text": "{\"selected_restaurant_name\":[],\"selected_timeRange\":[\"Nov. the 1st\"],\"selected_sort\":[]}",
       "value": {
-        "selected_restaurant_name": [
-          "mughlai restaurant"
-        ],
+        "selected_restaurant_name": [],
         "selected_timeRange": [
           "Nov. the 1st"
         ],
@@ -2112,8 +1274,8 @@
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
-      "output_tokens": 47,
-      "seconds": 0.16156229190528393,
+      "output_tokens": 40,
+      "seconds": 0.15424016606993973,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2126,28 +1288,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:BookRestaurant:train:684:view1:native:alias",
-      "checks": {
-        "exact": 0,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 3,
-        "tp": 1,
-        "fp": 1,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 1,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 1,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"restaurant_name\":\"mughlai restaurant\",\"timeRange\":\"Nov. the 1st\",\"sort\":null}",
@@ -2160,7 +1302,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 34,
-      "seconds": 0.13589041586965322,
+      "seconds": 0.1165991669986397,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2173,44 +1315,24 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:BookRestaurant:train:684:view1:native:nullable",
-      "checks": {
-        "exact": 0,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 3,
-        "tp": 1,
-        "fp": 1,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 1,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 1,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": false
     },
     {
-      "text": "{\"selected_condition_temperature\":[\"colder\"],\"selected_state\":[\"Louisiana\"]}",
+      "text": "{\"selected_state\":[\"Louisiana\"],\"selected_condition_temperature\":[\"colder\"]}",
       "value": {
-        "selected_condition_temperature": [
-          "colder"
-        ],
         "selected_state": [
           "Louisiana"
+        ],
+        "selected_condition_temperature": [
+          "colder"
         ]
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 29,
-      "seconds": 0.10488454182632267,
+      "seconds": 0.12024129112251103,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2223,28 +1345,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:GetWeather:train:109:view0:native:alias",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"state\":\"Louisiana\",\"condition_temperature\":\"colder\"}",
@@ -2256,7 +1358,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 19,
-      "seconds": 0.08123483415693045,
+      "seconds": 0.06711416691541672,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2269,28 +1371,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:GetWeather:train:109:view0:native:nullable",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"selected_playlist\":[],\"selected_music_item\":[\"theme\"],\"selected_service\":[]}",
@@ -2305,7 +1387,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 34,
-      "seconds": 0.14011824992485344,
+      "seconds": 0.09712295792996883,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2318,28 +1400,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:PlayMusic:train:801:view1:native:alias",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"playlist\":null,\"music_item\":\"theme\",\"service\":null}",
@@ -2352,7 +1414,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 23,
-      "seconds": 0.09768870798870921,
+      "seconds": 0.06972587504424155,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2365,28 +1427,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:PlayMusic:train:801:view1:native:nullable",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"selected_rating_value\":[\"2\"],\"selected_object_select\":[\"current\"]}",
@@ -2402,7 +1444,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 29,
-      "seconds": 0.11820799997076392,
+      "seconds": 0.08707858389243484,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2415,28 +1457,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:RateBook:train:293:view0:native:alias",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"object_select\":\"current\",\"rating_value\":\"2\"}",
@@ -2448,7 +1470,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 19,
-      "seconds": 0.0810724999755621,
+      "seconds": 0.08294720784761012,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2461,28 +1483,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:RateBook:train:293:view0:native:nullable",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 2,
-        "tp": 2,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 2,
-        "present_total": 2,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 2,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"selected_object_name\":[\"Playlist: The Very Best of Dolly Parton\"]}",
@@ -2495,7 +1497,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 25,
-      "seconds": 0.10242487490177155,
+      "seconds": 0.11037545814178884,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2508,28 +1510,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:SearchCreativeWork:train:602:view1:native:alias",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 1,
-        "fields": 1,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"object_name\":\"Playlist: The Very Best of Dolly Parton\"}",
@@ -2540,7 +1522,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 20,
-      "seconds": 0.0858239580411464,
+      "seconds": 0.08364866697229445,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2553,28 +1535,8 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:SearchCreativeWork:train:602:view1:native:nullable",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 1,
-        "fields": 1,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 0,
-        "absent_total": 0,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
       "text": "{\"selected_movie_type\":[\"animated movies\"],\"selected_movie_name\":[],\"selected_location_name\":[]}",
@@ -2589,7 +1551,7 @@
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 42,
-      "seconds": 0.17146354098804295,
+      "seconds": 0.16512083308771253,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2602,41 +1564,21 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:SearchScreeningEvent:train:897:view1:native:alias",
-      "checks": {
-        "exact": 1,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 3,
-        "fields": 3,
-        "tp": 1,
-        "fp": 0,
-        "fn": 0,
-        "present_correct": 1,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
     },
     {
-      "text": "{\"movie_name\":null,\"movie_type\":\"movies\",\"location_name\":null}",
+      "text": "{\"movie_type\":\"animated movies\",\"movie_name\":null,\"location_name\":null}",
       "value": {
+        "movie_type": "animated movies",
         "movie_name": null,
-        "movie_type": "movies",
         "location_name": null
       },
       "schema_valid": true,
       "success": true,
       "stop_reason": "turn_end",
       "output_tokens": 31,
-      "seconds": 0.1260323328897357,
+      "seconds": 0.1260217921808362,
       "output_mode": "native-json",
       "model_calls": 1,
       "json_assembly": false,
@@ -2649,34 +1591,198 @@
         "serialization": "native-sft-chat-v1"
       },
       "id": "native:snips:SearchScreeningEvent:train:897:view1:native:nullable",
-      "checks": {
-        "exact": 0,
-        "json_parse_ok": 1,
-        "schema_ok": 1,
-        "field_correct": 2,
-        "fields": 3,
-        "tp": 0,
-        "fp": 1,
-        "fn": 1,
-        "present_correct": 0,
-        "present_total": 1,
-        "absent_correct": 2,
-        "absent_total": 2,
-        "false_abstention": 0,
-        "missed_abstention": 0,
-        "unsupported_values": 0,
-        "predicted_values": 1,
-        "natural_stop": 1,
-        "token_limited": 0,
-        "nonrepetitive_nonempty": 1
-      },
-      "saved_text_and_stop_match": true
+      "raw_text_and_stop_match": true,
+      "exact": true
+    },
+    {
+      "text": "{\"artist\":[\"Mira Sol\",\"Theo Lane\",\"Nora Vale\"],\"album\":[]}",
+      "value": {
+        "artist": [
+          "Mira Sol",
+          "Theo Lane",
+          "Nora Vale"
+        ],
+        "album": []
+      },
+      "schema_valid": true,
+      "success": true,
+      "stop_reason": "turn_end",
+      "output_tokens": 30,
+      "seconds": 0.12100991699844599,
+      "output_mode": "native-json",
+      "model_calls": 1,
+      "json_assembly": false,
+      "output_repair": false,
+      "constrained": false,
+      "protocol": {
+        "temperature": 0,
+        "repetition_penalty": 1,
+        "max_tokens": 256,
+        "serialization": "native-sft-chat-v1"
+      },
+      "id": "ARRAY-3-ARTIST",
+      "raw_text_and_stop_match": true,
+      "exact": true
+    },
+    {
+      "text": "{\"city\":[\"Bergen\",\"Ilmenau\",\"Trondheim\"],\"country\":[]}",
+      "value": {
+        "city": [
+          "Bergen",
+          "Ilmenau",
+          "Trondheim"
+        ],
+        "country": []
+      },
+      "schema_valid": true,
+      "success": true,
+      "stop_reason": "turn_end",
+      "output_tokens": 27,
+      "seconds": 0.09397912514396012,
+      "output_mode": "native-json",
+      "model_calls": 1,
+      "json_assembly": false,
+      "output_repair": false,
+      "constrained": false,
+      "protocol": {
+        "temperature": 0,
+        "repetition_penalty": 1,
+        "max_tokens": 256,
+        "serialization": "native-sft-chat-v1"
+      },
+      "id": "ARRAY-3-CITY",
+      "raw_text_and_stop_match": true,
+      "exact": false
+    },
+    {
+      "text": "{\"artist\":[\"Mira Sol\",\"Theo Lane\",\"Nora Vale\",\"Eli Brooks\"],\"album\":[]}",
+      "value": {
+        "artist": [
+          "Mira Sol",
+          "Theo Lane",
+          "Nora Vale",
+          "Eli Brooks"
+        ],
+        "album": []
+      },
+      "schema_valid": true,
+      "success": true,
+      "stop_reason": "turn_end",
+      "output_tokens": 35,
+      "seconds": 0.10037354100495577,
+      "output_mode": "native-json",
+      "model_calls": 1,
+      "json_assembly": false,
+      "output_repair": false,
+      "constrained": false,
+      "protocol": {
+        "temperature": 0,
+        "repetition_penalty": 1,
+        "max_tokens": 256,
+        "serialization": "native-sft-chat-v1"
+      },
+      "id": "ARRAY-4-ARTIST",
+      "raw_text_and_stop_match": true,
+      "exact": true
+    },
+    {
+      "text": "{\"city\":[\"Bergen\",\"Ilmenau\",\"Trondheim\",\"Stavanger\"],\"country\":[]}",
+      "value": {
+        "city": [
+          "Bergen",
+          "Ilmenau",
+          "Trondheim",
+          "Stavanger"
+        ],
+        "country": []
+      },
+      "schema_valid": true,
+      "success": true,
+      "stop_reason": "turn_end",
+      "output_tokens": 32,
+      "seconds": 0.12685562507249415,
+      "output_mode": "native-json",
+      "model_calls": 1,
+      "json_assembly": false,
+      "output_repair": false,
+      "constrained": false,
+      "protocol": {
+        "temperature": 0,
+        "repetition_penalty": 1,
+        "max_tokens": 256,
+        "serialization": "native-sft-chat-v1"
+      },
+      "id": "ARRAY-4-CITY",
+      "raw_text_and_stop_match": true,
+      "exact": false
+    },
+    {
+      "text": "{\"artist\":[\"Mira Sol\",\"Theo Lane\",\"Nora Vale\",\"Eli Brooks\",\"Iris Reed\"],\"album\":[]}",
+      "value": {
+        "artist": [
+          "Mira Sol",
+          "Theo Lane",
+          "Nora Vale",
+          "Eli Brooks",
+          "Iris Reed"
+        ],
+        "album": []
+      },
+      "schema_valid": true,
+      "success": true,
+      "stop_reason": "turn_end",
+      "output_tokens": 40,
+      "seconds": 0.16119341715238988,
+      "output_mode": "native-json",
+      "model_calls": 1,
+      "json_assembly": false,
+      "output_repair": false,
+      "constrained": false,
+      "protocol": {
+        "temperature": 0,
+        "repetition_penalty": 1,
+        "max_tokens": 256,
+        "serialization": "native-sft-chat-v1"
+      },
+      "id": "ARRAY-5-ARTIST",
+      "raw_text_and_stop_match": true,
+      "exact": true
+    },
+    {
+      "text": "{\"city\":[\"Bergen\",\"Eden\",\"Tromsø\"],\"country\":[]}",
+      "value": {
+        "city": [
+          "Bergen",
+          "Eden",
+          "Tromsø"
+        ],
+        "country": []
+      },
+      "schema_valid": true,
+      "success": true,
+      "stop_reason": "turn_end",
+      "output_tokens": 26,
+      "seconds": 0.10789529187604785,
+      "output_mode": "native-json",
+      "model_calls": 1,
+      "json_assembly": false,
+      "output_repair": false,
+      "constrained": false,
+      "protocol": {
+        "temperature": 0,
+        "repetition_penalty": 1,
+        "max_tokens": 256,
+        "serialization": "native-sft-chat-v1"
+      },
+      "id": "ARRAY-5-CITY",
+      "raw_text_and_stop_match": true,
+      "exact": false
     }
   ],
   "model_calls_per_request": 1,
   "json_assembly": false,
   "output_repair": false,
   "constrained_decoding": false,
-  "scope": "Stock MLX standalone packaging parity on the observed 54 playground fixtures; not a fresh blind accuracy test or full 1,444-request cross-runtime parity.",
-  "elapsed_seconds": 5.9345723749138415
+  "scope": "Stock-MLX packaging parity on the observed 60 playground fixtures; not a fresh blind benchmark or full 1,444-case cross-runtime parity.",
+  "elapsed_seconds": 6.364110708935186
 }

PACKAGING_VERIFICATION.json

@@ -1,149 +1,17 @@
 {
   "model": "OpenSML-JSONLite-175M-Instruct",
-  "date": "2026-10-14",
+  "date": "2026-10-16",
   "checkpoint": "main",
   "weight_file": "model.safetensors",
   "weight_bytes": 351242342,
-  "weight_sha256": "655fb5cfba009163cd8a7f7fe7b170b998bb452c217b1b473cfe10b33de0ccd1",
-  "native_bundle_sha256": "a1b63ed6286d6c4f8bc4f7526f4a11fc46f936830daaac22224bfa0c4892755b",
+  "weight_sha256": "781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560",
+  "native_bundle_sha256": "c54fd408f6b7ec0c6e80377ee8b6eeffe3bfbf76389e56fdad4c78f54828e2d1",
   "runtime": "stock MLX 0.32.3; Apple Silicon; Python 3.11",
-  "requests": 54,
-  "raw_runtime_matches": 54,
-  "complete_correct": 44,
-  "schema_valid": 54,
-  "root_runner_verification": {
-    "passed": true,
-    "examples": [
-      {
-        "example": 29,
-        "result": {
-          "text": "{\"track\":[\"Blue Lantern\"],\"artist\":[\"Mira Sol\"],\"album\":[]}",
-          "value": {
-            "track": [
-              "Blue Lantern"
-            ],
-            "artist": [
-              "Mira Sol"
-            ],
-            "album": []
-          },
-          "schema_valid": true,
-          "success": true,
-          "stop_reason": "turn_end",
-          "output_tokens": 29,
-          "seconds": 0.09680337482132018,
-          "output_mode": "native-json",
-          "model_calls": 1,
-          "json_assembly": false,
-          "output_repair": false,
-          "constrained": false,
-          "protocol": {
-            "temperature": 0,
-            "repetition_penalty": 1,
-            "max_tokens": 256,
-            "serialization": "native-sft-chat-v1"
-          },
-          "exact": true
-        }
-      },
-      {
-        "example": 30,
-        "result": {
-          "text": "{\"restaurant_name\":[\"River Moon\"],\"city\":[\"Denver\"],\"party_size_number\":[\"four\"],\"timeRange\":[\"tomorrow\"]}",
-          "value": {
-            "restaurant_name": [
-              "River Moon"
-            ],
-            "city": [
-              "Denver"
-            ],
-            "party_size_number": [
-              "four"
-            ],
-            "timeRange": [
-              "tomorrow"
-            ]
-          },
-          "schema_valid": true,
-          "success": true,
-          "stop_reason": "turn_end",
-          "output_tokens": 49,
-          "seconds": 0.14814749988727272,
-          "output_mode": "native-json",
-          "model_calls": 1,
-          "json_assembly": false,
-          "output_repair": false,
-          "constrained": false,
-          "protocol": {
-            "temperature": 0,
-            "repetition_penalty": 1,
-            "max_tokens": 256,
-            "serialization": "native-sft-chat-v1"
-          },
-          "exact": true
-        }
-      },
-      {
-        "example": 35,
-        "result": {
-          "text": "{\"artist\":[\"Mira Sol\",\"Theo Lane\"],\"album\":[]}",
-          "value": {
-            "artist": [
-              "Mira Sol",
-              "Theo Lane"
-            ],
-            "album": []
-          },
-          "schema_valid": true,
-          "success": true,
-          "stop_reason": "turn_end",
-          "output_tokens": 24,
-          "seconds": 0.09080504113808274,
-          "output_mode": "native-json",
-          "model_calls": 1,
-          "json_assembly": false,
-          "output_repair": false,
-          "constrained": false,
-          "protocol": {
-            "temperature": 0,
-            "repetition_penalty": 1,
-            "max_tokens": 256,
-            "serialization": "native-sft-chat-v1"
-          },
-          "exact": true
-        }
-      },
-      {
-        "example": 48,
-        "result": {
-          "text": "{\"playlist\":null,\"music_item\":\"theme\",\"service\":null}",
-          "value": {
-            "playlist": null,
-            "music_item": "theme",
-            "service": null
-          },
-          "schema_valid": true,
-          "success": true,
-          "stop_reason": "turn_end",
-          "output_tokens": 23,
-          "seconds": 0.0877893331926316,
-          "output_mode": "native-json",
-          "model_calls": 1,
-          "json_assembly": false,
-          "output_repair": false,
-          "constrained": false,
-          "protocol": {
-            "temperature": 0,
-            "repetition_penalty": 1,
-            "max_tokens": 256,
-            "serialization": "native-sft-chat-v1"
-          },
-          "exact": true
-        }
-      }
-    ],
-    "scope": "Four documented examples via packaged archive bootstrap, no assembly or repair."
-  },
+  "requests": 60,
+  "raw_runtime_matches": 60,
+  "complete_correct": 49,
+  "schema_valid": 60,
+  "root_runner_verification": "source archive hash and snapshot verified; live CLI receipt saved separately",
   "json_assembly": false,
   "output_repair": false,
   "tokenizer_root_matches_bundle": true

README.md

@@ -3,7 +3,7 @@ language:
 - en
 pipeline_tag: text-generation
 library_name: mlx
-base_model: wzebrowski/OpenSML-Base-175M
+base_model: OpenSML/OpenSML-Base-175M
 base_model_relation: finetune
 tags:
 - information-extraction
@@ -25,13 +25,13 @@ datasets:
 
 # OpenSML-JSONLite-175M-Instruct
 
-**Author:** William Zebrowski · **175,611,648 parameters**
+**Author:** William Zebrowski · **175M parameters**
 
 [Run locally](RUN_GUIDE.md) · [Technical report](TECHNICAL_REPORT.md) · [Results](JSONLITE_RESULTS.json) · [Provenance](checkpoint_provenance.json)
 
 **Main checkpoint: a small language model that generates complete JSON directly from source text and a supplied JSON Schema.** It makes one autoregressive call; parsing/schema validation follows generation. No JSON assembly, output repair, constrained decoder or secondary model is used.
 
-This release retains the grounded-QA, exact/focused extraction and native JSON foundation, then adds a focused extraction continuation with original TRAIN replay, matched zero/one/two/three-value examples and frozen-parent distillation. The display name and standard `model.safetensors` filename stay unchanged. The pretrained ancestor is [OpenSML-Base-175M](https://huggingface.co/wzebrowski/OpenSML-Base-175M).
+This release retains the grounded-QA, exact/focused extraction and native JSON foundation, then adds a joint JSON extraction continuation with original TRAIN replay, multiple-value examples and authored prose. The display name and standard `model.safetensors` filename stay unchanged. The pretrained ancestor is [OpenSML-Base-175M](https://huggingface.co/OpenSML/OpenSML-Base-175M).
 
 ## Run locally
 
@@ -39,7 +39,7 @@ This release retains the grounded-QA, exact/focused extraction and native JSON f
 python3 -m venv .venv
 source .venv/bin/activate
 python -m pip install huggingface_hub
-hf download wzebrowski/OpenSML-JSONLite-175M --local-dir OpenSML-JSONLite-175M
+hf download OpenSML/OpenSML-JSONLite-175M --local-dir OpenSML-JSONLite-175M
 cd OpenSML-JSONLite-175M
 python -m pip install -r requirements.txt
 python inference.py --example 35
@@ -81,7 +81,8 @@ Source text:
 Play the track Blue Lantern by Mira Sol.
 ```
 
-JSON Schema:
+<details>
+<summary>View JSON Schema</summary>
 
 ```json
 {
@@ -118,12 +119,17 @@ JSON Schema:
 }
 ```
 
-Actual model output:
+</details>
+
+<details>
+<summary>View generated JSON</summary>
 
 ```json
 {"track":["Blue Lantern"],"artist":["Mira Sol"],"album":[]}
 ```
 
+</details>
+
 Run: `python inference.py --example 29`.
 
 ### Restaurant request
@@ -134,7 +140,8 @@ Source text:
 Book River Moon in Denver for four people tomorrow.
 ```
 
-JSON Schema:
+<details>
+<summary>View JSON Schema</summary>
 
 ```json
 {
@@ -179,12 +186,17 @@ JSON Schema:
 }
 ```
 
-Actual model output:
+</details>
+
+<details>
+<summary>View generated JSON</summary>
 
 ```json
 {"restaurant_name":["River Moon"],"city":["Denver"],"party_size_number":["four"],"timeRange":["tomorrow"]}
 ```
 
+</details>
+
 Run: `python inference.py --example 30`.
 
 ### Multiple artists
@@ -195,7 +207,8 @@ Source text:
 Play songs by Mira Sol and Theo Lane.
 ```
 
-JSON Schema:
+<details>
+<summary>View JSON Schema</summary>
 
 ```json
 {
@@ -224,12 +237,17 @@ JSON Schema:
 }
 ```
 
-Actual model output:
+</details>
+
+<details>
+<summary>View generated JSON</summary>
 
 ```json
 {"artist":["Mira Sol","Theo Lane"],"album":[]}
 ```
 
+</details>
+
 Run: `python inference.py --example 35`.
 
 ### Nullable fields
@@ -240,7 +258,8 @@ Source text:
 Play the last theme by Behzad Mirkhani.
 ```
 
-JSON Schema:
+<details>
+<summary>View JSON Schema</summary>
 
 ```json
 {
@@ -278,60 +297,64 @@ JSON Schema:
 }
 ```
 
-Actual model output:
+</details>
+
+<details>
+<summary>View generated JSON</summary>
 
 ```json
 {"playlist":null,"music_item":"theme","service":null}
 ```
 
+</details>
+
 Run: `python inference.py --example 48`.
 
 ## Training
 
+
 The base was pretrained from scratch on 17,300,111,360 token positions using five Apple Silicon Macs with Thunderbolt/RDMA distributed training. Intended mixture: 50% FineWeb-Edu, 18% DCLM-Edu, 13% English FineWiki, 9% Cosmopedia v2,10% educational Python. These are configured sampling shares, not unique-document counts. Base held-out validation loss: 2.5563359552.
 
 ![Training loss and held-out validation](pretraining-training-loss.png)
 
 *Training loss and held-out validation during pretraining.*
 
-|Adaptation stage|What is inherited|
-|---|---|
-|Grounded evidence QA|15,008 presentations through the selected ancestor|
-|Requested-field extraction|12,115 SNIPS requests|
-|Focused extraction plus replay|5,120 presentations|
-|Native JSON foundation|15,000 source/schema/JSON presentations|
-|Additional extraction SFT|2,048 inherited presentations: 1,536 TRAIN replay + 512 controlled cardinality views|
+The latest full-weight SFT jointly mixes original source/schema/JSON training replay, controlled multiple-value examples and authored natural-language passages across the seven SNIPS domains. The prepared corpus contains **21,144 presentations**: 15,000 original JSON views, 768 cardinality replay views and 5,376 controlled prose presentations (1,753 distinct new texts). The released main checkpoint inherits **19,200 presentations** from one deterministic shuffle: 13,640 original JSON views, 689 cardinality views and 4,871 prose views. It processes **6,153,194 serialized positions**, including **633,561 supervised assistant positions**. Later updates from the completed experiment are not included.
+
+The run uses full-weight BF16 compute with FP32-master AdamW, batch 64, microbatch 2, betas (0.9, 0.95), epsilon 1e-8, zero weight decay and gradient clipping at 1. A fresh optimizer warms up over 20 updates to 5e-6, then follows a cosine schedule toward 1e-6 over the full 331-update experiment. Only assistant JSON and the turn-ending token contribute to mean per-request token cross-entropy. There is no teacher KL, LoRA or RL in this latest stage. Source text, schema and instruction tokens are masked. The inherited earlier cardinality stage used training-only distillation; no teacher is used at inference.
 
-The additional stage completed a 3,072-presentation experiment; the main checkpoint inherits only its selected 2,048-presentation prefix and 600,558 serialized positions. All stages update full model weights. The add-on uses batch 32/micro 2, fresh AdamW, 6-update warmup, peak 2e-6 and a 96-update cosine schedule toward 2e-7. On replay assistant tokens it adds KL from the frozen native JSON foundation; new rows use gold JSON CE alone. No teacher, RL, LoRA or adapter is needed at inference. Exact settings, data separation and stage hashes are in the report/provenance. Newly researched SGD/MIT datasets are not in these weights.
+The full inherited lineage and earlier SFT data are documented in the technical report.
 
 ## Native JSON evaluation
 
-Same original 1,444 SNIPS-derived source/schema requests and raw greedy protocol as the preceding release:
+Same original 1,444 SNIPS-derived source/schema requests and raw greedy protocol as the previous release:
+
+| Metric | Main checkpoint | Previous release |
+| --- | ---: | ---: |
+| Entire request correct | **88.02%** | 83.03% |
+| Field accuracy | **94.17%** | 91.36% |
+| Extracted-value F1 | **92.88%** | 90.40% |
+| Extracted-value precision | **92.42%** | 90.53% |
+| Extracted-value recall | **93.35%** | 90.26% |
+| Missing-field accuracy | **95.22%** | 92.79% |
+| Schema-valid JSON | **100.00%** | 99.79% |
 
-|Metric|Main checkpoint|Previous native JSON release|
-|---|---:|---:|
-|Entire request correct|**83.03%**|82.48%|
-|Field accuracy|**91.36%**|91.31%|
-|Extracted-value F1|**90.40%**|90.48%|
-|Extracted-value precision|**90.53%**|90.55%|
-|Extracted-value recall|**90.26%**|90.41%|
-|Missing-field accuracy|**92.79%**|92.46%|
-|Schema-valid JSON|**99.79%**|99.79%|
+Complete correctness is **1,271/1,444**, compared with 1,199/1,444 for the previous release: **96 improved requests and 24 regressions**, a net gain of 72. All seven domain aggregate correctness totals improve. Paired-source correctness rises from 512/694 to **561/694**. False abstentions fall from 107 to **53**, missed missing fields from 110 to **73**, and unsupported predicted values from 27 to **24**. All 1,444 outputs parse, validate and stop naturally; there are no output-budget failures.
 
-Complete correctness is 1,199/1,444 versus 1,191/1,444: 27 improved requests and 19 regressions. Unsupported values decrease 30→27. Canonical string-array views score 1,163/1,388 (83.79%) with 100% schema validity; renamed-key 21/28, nullable 15/28. The separate 54-example playground scores 44/54 (81.48%), versus 42/54; all 54 are schema-valid. The controlled new two/three-value development panel scores 14/24, versus 0/24. These are observed research panels, not fresh blind benchmarks or a public leaderboard.
+The separate observed 60-case playground scores **49/60 (81.67%)**, compared with 45/60. Its original 54-case subset scores 46/54; the six added three/four/five-item probes score 3/6. Three larger-list probes still fail. The compact 240-request development panel scores 178/240 (74.17%). These panels measure different scopes and must not be pooled. They have prior research exposure and are not fresh blind benchmarks or an external leaderboard.
 
-The main checkpoint was owner-selected despite strict automatic retention checks not passing. Playlist/weather exactness declines on the larger panel; aggregate precision/recall/F1 decline slightly while exact/field/missing correctness improve. The new 48-pattern panel has only 75% schema validity. Full tradeoffs and failures are documented.
+The main checkpoint was selected for aggregate improvement despite failing the strict zero-lost-case development gate. Individual regressions remain; improved aggregate scores do not establish universal retention or production reliability.
 
 ![Complete-request correctness on separate observed panels](complete-request-exact.svg)
 
 ## Limitations
 
-The model remains a focused SNIPS-domain extraction research model. Valid JSON does not prove correct values. It can invent, shorten, omit or misassign names, confuse venue/movie or state/country, and mishandle multiple values or nullable fields. For example, Bergen and Oslo can generate Bergen, Orange and Lake. Broader domains, long documents, corrections, arbitrary schemas, multilingual/chat/coding quality, calibration and prompt-injection robustness are unverified. `success` only means structure-valid natural completion. No blanket reliability/nonregression claim is made.
+A focused English extraction research model for documented flat schemas and mostly SNIPS fields. It can shorten, omit, invent or misassign values, confuse roles, miss list members, and fail on corrections or unfamiliar definitions. Valid JSON does not prove correct extraction. Long documents, nested objects, numeric/boolean outputs, multilingual tasks, general chat and prompt-injection robustness are outside the verified scope. `success` means structural validity and natural completion, not semantic correctness.
 
 ## Main checkpoint identity and package
 
 Selected BF16 weight SHA-256:
 
-`655fb5cfba009163cd8a7f7fe7b170b998bb452c217b1b473cfe10b33de0ccd1`
+`781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560`
 
-The package includes standard model.safetensors(351,242,342 bytes), config, unchanged tokenizer, source archive, examples, evaluation records and provenance. Source/archive/weight checksums are verified by the runner. The standalone verification receipt records raw-output and stop parity across 54 local fixtures; it is not an independent blind accuracy result. Historical revisions are retained for reproducibility. Source uses Apache-2.0; upstream terms apply separately; no trained-weight license is assigned by this update.
+Includes `model.safetensors`, unchanged tokenizer, native MLX runner, source bundle, 60 source/schema/reference examples, raw evaluation records, checksums and provenance. Standalone verification checks generated strings and stop reasons against the saved 60-case protocol; this is packaging parity, not an independent accuracy benchmark. Previous revisions remain available. Source uses Apache-2.0; upstream terms apply separately; no trained-weight license is assigned by this update.

RUN_GUIDE.md

@@ -3,7 +3,7 @@
 The current weights generate the entire JSON object from source text and the
 supplied schema. No fieldwise assembler, constrained decoder or repair is used.
 The runnable package includes model.safetensors, tokenizer/config, native MLX
-model source, a source archive, 54 examples, results and verification records.
+model source, a source archive, 60 examples, results and verification records.
 Verified platform: Apple Silicon macOS, Python 3.11, stock MLX 0.32.3.
 
 ```bash
@@ -11,7 +11,7 @@ python3 -m venv .venv
 source .venv/bin/activate
 python -m pip install huggingface_hub
 # Authenticate if access requires it: hf auth login
-hf download wzebrowski/OpenSML-JSONLite-175M --local-dir OpenSML-JSONLite-175M
+hf download OpenSML/OpenSML-JSONLite-175M --local-dir OpenSML-JSONLite-175M
 cd OpenSML-JSONLite-175M
 python -m pip install -r requirements.txt
 python inference.py --example 29
@@ -24,7 +24,7 @@ with raw `text`, parsed `value`, structural success, stop reason, token count an
 example exact correctness when a reference exists. The reference is compared
 only after generation and never enters the model input. Schema validity does
 not certify extracted values. Some included examples intentionally expose
-remaining errors. The current local diagnostic is 44/54, not 100% correctness.
+remaining errors. The current local diagnostic is 49/60, not 100% correctness.
 
 Example source: `Play the track Blue Lantern by Mira Sol.`
 
@@ -60,11 +60,11 @@ python native_bundle/inference.py --weights "$PWD/model.safetensors" --example 2
 ```
 
 Selected weights:
-`655fb5cfba009163cd8a7f7fe7b170b998bb452c217b1b473cfe10b33de0ccd1`.
+`781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560`.
 The current source module is inference_native/native_json.py. The old labeled
 adapter, if present in the historical source archive, is not used by this
 entry point. Optimizer state and the original training workspace are not needed
 for inference. See the technical report for lineage, metrics and limitations.
 
 
-The main checkpoint includes an additional cardinality/replay extraction SFT stage. The fixed inference instruction/schema/one-call native JSON protocol is unchanged; no frozen teacher is needed after download. Use example 35 to try two artists. Exact identities/history are in checkpoint_provenance.json.
+The main checkpoint adds joint source/schema/JSON SFT with replay, cardinality and authored prose. No teacher is needed at inference. Exact lineage and settings are in checkpoint_provenance.json.

SHA256SUMS

@@ -1,20 +1,21 @@
-dfe55d5efa7268fa3e8f2997dcb148e0145db80960cd460a242bacc81cef61a0  CITATION.cff
-980695fb234fa03113e27bad3ec2a87416168e19f4beaad98bbcad2da920f641  INFERENCE_VERIFICATION.json
-38eaa196a9bcbbb363d207b46a3c42fcc589687b7b9980e7eff218a1a8cf8bdb  JSONLITE_RESULTS.json
+5d659853ddccca76e4090ce3e4ebe15bc6017239752a54d663f5e997a695651a  .gitattributes
+4a49b666643257b6a7f8bc5e7051a4a1389cdd021778d14a736b2f5f1a2c005d  CITATION.cff
+9a4fb5f559d82ed0120a57139eb1e1c616bfdb18c39cca7b1e9bcab24ef968bc  INFERENCE_VERIFICATION.json
+74a074f90bd5e7a26d5f6cd5757ddc1fbfcc72e259e0b23a210fa48d91b35fdb  JSONLITE_RESULTS.json
 c71d239df91726fc519c6eb72d318ec65820627232b2f796219e87dcf35d0ab4  LICENSE
-30291e6ad10c7f9a79108f327fc26e68d52a3d579df36f2ac6d8dacc2316e9bf  PACKAGING_VERIFICATION.json
-c26710cf588d761b33783265fe97224a194927f2b04f8f36aace033a2367285b  README.md
-f48e17b06318efd4fdff1eba8427fcf503fd0b08fdfb4c000b9f8ae0b2e9cf92  RUN_GUIDE.md
-d5de06e61d0d6bd7245e515f41f7d3ca2d0716acc69d460f676b86f449fd078c  SOURCE_NOTICE.md
-974f3a2d6837c08d989a64ba942009c75a5d2015d32c6783c679ba6e8b47f907  TECHNICAL_REPORT.md
-4e9b202c1a70b476d74a4fe9b9a55e9756195a8b5bd3fc63558875a89487a91d  THIRD_PARTY_NOTICES.md
-49d177fb4606673f789fbf06c694643c44504672a9f0c3f75f7eae776cb4590c  checkpoint_metadata.json
-09494951de47b0cb0e2d2b683f27cf07139475b05a3fcad738bfc216ab4bc6a8  checkpoint_provenance.json
-ac810f02fd0428acc76123c281c746c70276fcee6a01b1c7a4e08e06db8097e2  complete-request-exact.svg
-4fd845ef735a67b567f0d46e123994c6dbaa18715c4b2550192720f54a280033  config.json
-2010f51d840d86ddae62fb65ac4eaaa03c90ca12f7272fe70d949f494b74d728  inference.py
-655fb5cfba009163cd8a7f7fe7b170b998bb452c217b1b473cfe10b33de0ccd1  model.safetensors
-a1b63ed6286d6c4f8bc4f7526f4a11fc46f936830daaac22224bfa0c4892755b  native_bundle.zip
+f53194e53a2af5ec857337e8213397ea4f81a7e02d2dbcf0384a00cf1cf7f76f  PACKAGING_VERIFICATION.json
+2728c5efa01d9d59fb1def12a543980c300edcf557fabfa921b08dc0712e68cb  README.md
+5ee8f11f0e87becf1a29c44e2b0f3dacc2c779712b635558be57dd7ef44455d3  RUN_GUIDE.md
+c6191baf937bce61f1b747acfbcb975a17eef6e8a598682a751a1031d1cfee23  SOURCE_NOTICE.md
+6088771ab13fa1080c0ca0f31649deb9f417597d7c2055b0f388e86f41f2c415  TECHNICAL_REPORT.md
+e4e44da050abbd930748ec91731ddf1be16be83a4aa59bd8c4fc63478b384a03  THIRD_PARTY_NOTICES.md
+663087d32872824f7f2fec425ad4054f308913c43da7c3f83521c7011967fdbe  checkpoint_metadata.json
+a365a7697b463bacc6e44e26d11a1e62d43d96b5d52c70c2e9df8d4513f86537  checkpoint_provenance.json
+ecec6f191822d16a3da0b571fa129f5016e1243ca7d567de476a9d0e8ad66646  complete-request-exact.svg
+981e02e7777980705770cacf19f16d76ea74e85e7edb018f875ddb4766afe19d  config.json
+8c0f0f5a91079699d090986d5d241385309b10579f34f271c02f05dca6ddd4bf  inference.py
+781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560  model.safetensors
+c54fd408f6b7ec0c6e80377ee8b6eeffe3bfbf76389e56fdad4c78f54828e2d1  native_bundle.zip
 641fe860d6fd24547fe4d15bd852190f00ddec31bd74eb7ca9eeae74031c2f0c  pretraining-training-loss.png
 c7b1a3c0efb09ef3b5416a24072e48b25e2cc1f122fb5ed638c0f95b72594af9  requirements.txt
 e095f0413a11fa91fbb58f37d00e4c6bcda59beb89b28f6f680db1436a7e8891  tokenizer.json

SOURCE_NOTICE.md

@@ -18,11 +18,9 @@ Pretraining decontamination and perfectly correct semantic golds are not claimed
 Raw pretraining pools, packed arrays, optimizer state, credentials, SSH keys and
 private host/network receipts are excluded. Published metadata replaces private
 workspace paths with [workspace]; original and published hashes are separated.
-54 actual local fixtures, raw predictions and held-out evidence are retained.
+60 actual local fixtures, raw predictions and held-out evidence are retained.
 The canonical held-out 1,388 and schema-stress 56 requests are included as JSONL in
 evaluation/native-json-heldout.jsonl and the prepared native data directory.
 See THIRD_PARTY_NOTICES.md, docs/TECHNICAL_REPORT.md and provenance/LINEAGE.json.
 
-The additional stage uses original SNIPS TRAIN replay plus authored cardinality
-views, with frozen-parent training-only KL. Main inherits 2,048 presentations, not
-the entire 3,072-presentation completed experiment. See the report for exact counts.
+The latest stage jointly replays original SNIPS TRAIN JSON views, controlled cardinality views and authored English prose with original TRAIN values. It adds 19,200 inherited presentations; no teacher KL is used in this stage. Prepared authored prose and targets are included for inspection, with observed evaluation scope and remaining failures in the report.

TECHNICAL_REPORT.md

@@ -1,20 +1,20 @@
 # OpenSML-JSONLite-175M-Instruct: native JSON technical report
 
-**Author:** William Zebrowski · **Parameters:** 175,611,648 · **Date:** 2026-10-14
+**Author:** William Zebrowski · **Parameters:** 175,611,648 · **Date:** 2026-10-16
 
 ## 1. Current artifact and behavior
 
 The main checkpoint directly generates JSON from source text and a supplied JSON Schema. A single native autoregressive request emits the entire object. There is no fieldwise extraction-to-JSON assembler, answer repair, constrained decoder or gold-conditioned filling in the current inference path. Parsing and schema validation happen only after generation; the raw generated string is retained even when incorrect or invalid.
 
 Current artifact: **main checkpoint**, BF16 `model.safetensors`, 351,242,342 bytes.
-SHA-256: `655fb5cfba009163cd8a7f7fe7b170b998bb452c217b1b473cfe10b33de0ccd1`.
+SHA-256: `781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560`.
 This is a custom native MLX architecture, not a Transformers/MLX-LM auto-loader export.
 
 This report supersedes the earlier fieldwise SFT-093 release. Its historical report is retained as `HISTORICAL_SFT093_REPORT.md`; those adapter-assisted scores are not current native JSON scores. No release-readiness or arbitrary-schema reliability claim is made.
 
 ## 2. Exact inherited lineage
 
-`Stage A → grounded QA → requested-field extraction → focused extraction → native JSON foundation → additional extraction SFT`
+`Stage A → grounded QA → requested-field extraction → focused extraction → native JSON foundation → cardinality/replay extraction → joint JSON extraction SFT`
 
 Only those selected checkpoints are inherited. Other trials and failed RL/JSON experiments are research history, not hidden additional ancestors. The architecture and frozen tokenizer remain unchanged. Each SFT stage resets its local counters and AdamW optimizer while continuing the inherited neural weights.
 
@@ -26,9 +26,10 @@ Only those selected checkpoints are inherited. Other trials and failed RL/JSON e
 | SFT-093 focused extraction/replay | 160 | 5,120 | 706,399 |
 | SFT-100 native JSON | 469 | 15,000 | 4,634,126 |
 | Additional extraction SFT | 64 selected of 96 completed | 2,048 | 600,558 |
-| Inherited post-training total | 1,541 updates | 49,291 | 10,881,007 |
+| Joint JSON extraction SFT | 300 of 331 | 19,200 | 6,153,194 |
+| Inherited post-training total | 1,841 updates | 68,491 | 17,034,201 |
 
-Presentation counts include repeated source views. Serialized positions include masked prompt tokens; they are not supervised-output-token counts. The earlier 150M model weights did not initialize this model; the earlier frozen OpenSML tokenizer was reused.
+The latest stage metadata counter records 633,561 supervised assistant positions; its serialized count above was recomputed from the exact selected shuffle prefix. Presentation counts include repeated source views. Serialized positions include masked prompt tokens; they are not supervised-output-token counts. The earlier 150M model weights did not initialize this model; the earlier frozen OpenSML tokenizer was reused.
 
 ## 3. Architecture and parameterization
 
@@ -57,7 +58,7 @@ The same architecture is retained across pretraining and all inherited SFT stage
 | FFN implementation | Packed Metal SwiGLU |
 | Training CE implementation | Metal cross entropy, FP32 reductions |
 
-Implementation evidence: [`model.py`](../provenance/LINEAGE.json), [`precision.py`](../provenance/LINEAGE.json), and the final bundle's `recipe.model` metadata. Weight count and file size are consistent with BF16 parameters plus the safetensors header. GPU allocation readings are not equivalent to file size, total process RSS or system RAM requirements.
+Implementation evidence: [`model.py`](checkpoint_provenance.json), [`precision.py`](checkpoint_provenance.json), and the final bundle's `recipe.model` metadata. Weight count and file size are consistent with BF16 parameters plus the safetensors header. GPU allocation readings are not equivalent to file size, total process RSS or system RAM requirements.
 
 The base initialization is recorded as `normal-0.02-residual-scaled-v1`. Earlier OpenSML-150M **weights** were not used to initialize the base; its frozen tokenizer was reused.
 
@@ -101,7 +102,7 @@ Source weights below are intended processed-token sampling weights, not measured
 
 Public source cards: [SmolLM corpus](https://huggingface.co/datasets/HuggingFaceTB/smollm-corpus/blob/3ba9d605774198c5868892d7a8deda78031a781f/README.md), [DCLM-Edu](https://huggingface.co/datasets/HuggingFaceTB/dclm-edu/blob/dbad8ad71224482740cd9c9d353591adbf62fe04/README.md), [FineWiki](https://huggingface.co/datasets/HuggingFaceFW/finewiki/blob/8bd13e72e6a002407649b3e898535f42ceb1aeb9/README.md). The Python-Edu upstream subset identifies files; this run used the configured text mirror. The mirror's pinned README could not be retrieved during report preparation, so its provenance statement here comes from the local corpus specification, not a newly completed external audit.
 
-The exact corpus specification is [`configs/corpus_20b.json`](../provenance/LINEAGE.json). Training preserves source text and code indentation. NFC/whitespace normalization is used for identity hashing and partitioning, not to rewrite model input.
+The exact corpus specification is [`configs/corpus_20b.json`](checkpoint_provenance.json). Training preserves source text and code indentation. NFC/whitespace normalization is used for identity hashing and partitioning, not to rewrite model input.
 
 The content-split policy hashes identity into 10,000 buckets: 97% training, 1% validation, 1% final and 1% tokenizer development. Identity checks do not establish that a supervised benchmark article never appeared elsewhere in pretraining web data.
 
@@ -161,15 +162,15 @@ The 175.6M-parameter base was trained from random initialization using next-toke
 *Training loss and held-out validation during pretraining. Left: raw training loss and exponential moving average. Right: held-out validation loss.*
 
 
-## 7. Supervised foundation and additional extraction SFT
+## 7. Inherited supervised foundation and prior extraction continuation
 
 The four inherited foundation stages remain unchanged: grounded evidence QA (15,008 presentations through its selected checkpoint), requested-field SNIPS extraction (12,115), focused single-field extraction/replay(5,120), then directly supervised native JSON (15,000). Their original source and checkpoint records remain in provenance. Each stage uses full model weights, fresh AdamW, BF16 computation with FP32 master weights/moments/gradient accumulation, and assistant-output-only supervision. No RL, LoRA or added extraction head is in this lineage.
 
 The native JSON foundation has 12,115 distinct SNIPS TRAIN requests, with 12,115 canonical string-array schemas, 1,443 renamed-key views and 1,442 nullable-string views. Source text, field definitions and the complete schema are input; complete JSON and stop tokens are the assistant target. Original CC0 SNIPS chunk annotations provide source spans. Metadata fidelity and schema validation do not establish perfect semantic annotation. Frozen tokenizer and architecture are unchanged.
 
-### Additional extraction SFT
+### Prior cardinality/replay extraction SFT
 
-This completed continuation starts from the native JSON foundation. It adds matched zero/one/two/three-value supervision for artists and cities, including zero-primary-value cases where an album or country is explicitly present. It retains original JSON TRAIN replay and adds frozen-parent distribution distillation on replay assistant tokens. Every model batch has 24 replay and 8 new views. Public checkpoint name: **main**; the neural tensor is `model.safetensors`.
+This completed continuation starts from the native JSON foundation. It adds matched zero/one/two/three-value supervision for artists and cities, including zero-primary-value cases where an album or country is explicitly present. It retains original JSON TRAIN replay and adds frozen-parent distribution distillation on replay assistant tokens. Every model batch has 24 replay and 8 new views. This is an inherited ancestor of the current main checkpoint.
 
 | Setting | Value |
 |---|---|
@@ -187,82 +188,84 @@ This completed continuation starts from the native JSON foundation. It adds matc
 | Weight decay / clipping |0.01 /1.0|
 | Seed |20261119|
 
-The main checkpoint inherits only the selected prefix; updates after it are not part of these weights. The endpoint's actual learning rate is recorded by the cosine schedule, not the planned final LR. Presentation counts include multiple schemas for the same text and do not mean distinct user queries.
+The current weights inherit only this ancestor’s selected prefix; updates after that ancestor are not part of these weights. The endpoint's actual learning rate is recorded by the cosine schedule, not the planned final LR. Presentation counts include multiple schemas for the same text and do not mean distinct user queries.
 
 For every row, CE is averaged over supervised assistant tokens within each sequence, then averaged across sequences. On original TRAIN replay add 1.0×KL(frozen foundation || student), using the full 32,000-token distribution, temperature 1 and teacher-forced ground-truth histories. Only assistant JSON and turn/document-stop positions participate. The reference network is frozen/stop-gradient; new authored rows use CE alone. This regularizer is training-only: no teacher is needed for inference, and it does not guarantee retention or repair parent errors.
 
 The new views use literal original SNIPS TRAIN values with split-disjoint authored value pools/phrases, excluding observed evaluation source texts and holding new authored DEV/TEST values out of replay. Source text, exact offsets, schema/gold agreement, masks, token budgets and hashes were checked. Twenty-eight tests, GPU accumulation checks, frozen-reference checks and a temporary optimizer/checkpoint-resume smoke passed. Authored examples are programmatic controlled supervision, not independently collected user traffic. SGD, MIT Movie and other newly researched datasets are **not** in this checkpoint.
 
+
+### Latest joint JSON extraction SFT
+
+The latest full-weight SFT jointly mixes original source/schema/JSON training replay, controlled multiple-value examples and authored natural-language passages across the seven SNIPS domains. The prepared corpus contains **21,144 presentations**: 15,000 original JSON views, 768 cardinality replay views and 5,376 controlled prose presentations (1,753 distinct new texts). The released main checkpoint inherits **19,200 presentations** from one deterministic shuffle: 13,640 original JSON views, 689 cardinality views and 4,871 prose views. It processes **6,153,194 serialized positions**, including **633,561 supervised assistant positions**. Later updates from the completed experiment are not included.
+
+The run uses full-weight BF16 compute with FP32-master AdamW, batch 64, microbatch 2, betas (0.9, 0.95), epsilon 1e-8, zero weight decay and gradient clipping at 1. A fresh optimizer warms up over 20 updates to 5e-6, then follows a cosine schedule toward 1e-6 over the full 331-update experiment. Only assistant JSON and the turn-ending token contribute to mean per-request token cross-entropy. There is no teacher KL, LoRA or RL in this latest stage. Source text, schema and instruction tokens are masked. The inherited earlier cardinality stage used training-only distillation; no teacher is used at inference.
+
 ## 8. Prompt, schema and native inference
 
 Supply source text and a flat closed JSON Schema with 1–8 required fields. Supported properties are strings, nullable strings and arrays of strings. Every required key is returned; absent arrays target [] and absent nullable strings target null. Copy case/spelling and complete values; preserve source order. Nested schemas, numbers/booleans, enums and advanced constraints are outside the verified runner contract. The supplied runner rejects overlong inputs without truncation.
 
 The fixed instruction is followed by compact `{"source_text":...,"schema":...}` and native SFT role tokens. It issues exactly one autoregressive whole-object generation with temperature 0, repetition penalty 1 and at most 256 output tokens. No fieldwise assembly, repair, schema-constrained decoder, normalization, reference answer, frozen teacher or secondary model is used at inference. Parsing/schema checks only annotate the raw generated text afterward.
 
-## 9. Evaluation protocol and selection
-
-The continuation checks 404 fixed development requests plus 54 local playground fixtures at its scheduled checkpoints. DEV contains 252 original native JSON controls, 64 previously observed combined-request controls, 8 earlier single-request probes, 32 earlier authored list controls and 48 new cardinality controls. Metrics include exact requests/fields, precision/recall/F1, present/missing fields, false/missed abstentions, unsupported values, strict parsing, schema validity, stopping/budget failures, paired requests and cardinality-group correctness. These panels have prior research exposure; no new blind/general extraction benchmark claim is made.
+## 9. Evaluation and release selection
 
-The main checkpoint is an owner-selected research tradeoff. Strict automatic retention gates did not pass: some domain/variant unsupported values and nullable false abstentions increased. The run's automatic best pointer remains its starting baseline; the published main is explicitly the selected trained checkpoint, not a claim that the automatic gate accepted it. Larger original TEST comparison was performed after the candidate was nominated from development/playground results. Subsequent owner release choice observed that comparison. The original 1,444 panel also has earlier research exposure.
+| Metric | Main checkpoint | Previous release |
+| --- | ---: | ---: |
+| Entire request correct | **88.02%** | 83.03% |
+| Field accuracy | **94.17%** | 91.36% |
+| Extracted-value F1 | **92.88%** | 90.40% |
+| Extracted-value precision | **92.42%** | 90.53% |
+| Extracted-value recall | **93.35%** | 90.26% |
+| Missing-field accuracy | **95.22%** | 92.79% |
+| Schema-valid JSON | **100.00%** | 99.79% |
 
-### Main checkpoint versus previous native JSON release
+Complete correctness is **1,271/1,444**, compared with 1,199/1,444 for the previous release: **96 improved requests and 24 regressions**, a net gain of 72. All seven domain aggregate correctness totals improve. Paired-source correctness rises from 512/694 to **561/694**. False abstentions fall from 107 to **53**, missed missing fields from 110 to **73**, and unsupported predicted values from 27 to **24**. All 1,444 outputs parse, validate and stop naturally; there are no output-budget failures.
 
-Identical original 1,444 source/schema requests, tokenizer, native prompt, raw greedy decoding, 256-token budget and scorer were used. Previous release outputs were identity-checked and rescored from their saved cache. The main checkpoint generated every request, with no dropped cases or token limits. Complete correctness requires natural stopping, strict JSON/schema validity and exact field/value/order agreement. Value scores use(field, value) multisets; literal unsupported-value checks do not detect every wrong association.
+The separate observed 60-case playground scores **49/60 (81.67%)**, compared with 45/60. Its original 54-case subset scores 46/54; the six added three/four/five-item probes score 3/6. Three larger-list probes still fail. The compact 240-request development panel scores 178/240 (74.17%). These panels measure different scopes and must not be pooled. They have prior research exposure and are not fresh blind benchmarks or an external leaderboard.
 
-|Metric|Previous native JSON release|Main checkpoint|
-|---|---:|---:|
-|Entire request correct|82.4792%|83.0332%|
-|Field accuracy|91.3068%|91.3640%|
-|Extracted-value F1|90.4798%|90.3963%|
-|Extracted-value precision|90.5488%|90.5344%|
-|Extracted-value recall|90.4110%|90.2588%|
-|Missing-field accuracy|92.4640%|92.7916%|
-|Schema-valid JSON|99.7922%|99.7922%|
+The main checkpoint was selected for aggregate improvement despite failing the strict zero-lost-case development gate. Individual regressions remain; improved aggregate scores do not establish universal retention or production reliability.
 
-Complete requests increase 1,191→1,199/1,444 (+8).27 previously wrong requests become correct and 19 previously correct become wrong; those 19 schema views cover 18 unique source texts. Paired source correctness increases 501→512/694. Unsupported values 30→27; false abstentions 111→107; missed missing fields 115→110. Strict parsing remains 1,442/1,444 and schema validity 1,441/1,444. These modest aggregate gains coexist with small F1/precision/recall decreases.
+| Metric | Main checkpoint | Previous release |
+| --- | ---: | ---: |
+| Entire request correct | **88.02%** | 83.03% |
+| Field accuracy | **94.17%** | 91.36% |
+| Extracted-value F1 | **92.88%** | 90.40% |
+| Extracted-value precision | **92.42%** | 90.53% |
+| Extracted-value recall | **93.35%** | 90.26% |
+| Missing-field accuracy | **95.22%** | 92.79% |
+| Schema-valid JSON | **100.00%** | 99.79% |
 
-|Domain|Previous correct|Main correct|Requests|
-|---|---:|---:|---:|
-|AddToPlaylist|168|165|206|
-|BookRestaurant|156|162|208|
-|GetWeather|168|164|206|
-|PlayMusic|149|153|208|
-|RateBook|188|193|204|
-|SearchCreativeWork|197|197|208|
-|SearchScreeningEvent|165|165|204|
+### Correctness by domain
 
-Canonical schemas: 1,163/1,388 (83.79%), 100% schema validity. Renamed-key stress 21/28; nullable stress 15/28. Playlist and weather exactness decrease, while restaurants, music and ratings increase. Do not describe this as uniform improvement or guaranteed nonregression. Full per-field counts, NLL and per-domain/variant metrics are in the results JSON.
+| Domain | Previous | Main | Requests |
+| --- | ---: | ---: | ---: |
+| AddToPlaylist | 165 | 172 | 206 |
+| BookRestaurant | 162 | 172 | 208 |
+| GetWeather | 164 | 178 | 206 |
+| PlayMusic | 153 | 164 | 208 |
+| RateBook | 193 | 195 | 204 |
+| SearchCreativeWork | 197 | 199 | 208 |
+| SearchScreeningEvent | 165 | 191 | 204 |
 
-### Development and local examples
+### Protocol and contamination limits
 
-|Panel|Previous|Main|
-|---|---:|---:|
-|Original 252 DEV complete|200/252|206/252|
-|New 48 cardinality DEV complete|15/48|32/48|
-|New two/three-value requests|0/24|14/24|
-|Earlier 8 multi-value probes|0/8|7/8|
-|54 local playground complete|42/54|44/54|
-|54 local playground schema-valid|52/54|54/54|
+Greedy temperature 0, repetition penalty 1, maximum 256 output tokens and the unchanged native serialization. Every request supplies only the source text, schema and fixed extraction instruction. References, gold spans and offsets are used after generation only. No prompt truncation, constrained decoding, secondary model, JSON assembly or repair is used. Strict parsing rejects duplicate keys and nonfinite constants. Complete correctness requires a natural stop, valid JSON and schema, and exact agreement of all fields. Lists retain order; value precision/recall/F1 compare extracted values, while field accuracy includes missing fields. The full machine-readable result includes counts, denominators, paired correctness, unsupported values and stopping metrics.
 
-Local field accuracy 124/135(91.85%); value precision 89.53%, recall 91.67%, F190.59%; missing 49/53(92.45%); unsupported values 3; all 54 stop naturally. Local precision declines from 91.03% despite higher recall/correctness. The new 48-pattern panel has only 36/48(75%) schema-valid outputs. All panels remain separate; do not pool them or claim an external leaderboard result.
+The 1,444 requests comprise the original canonical and schema-stress views; some views reuse the same source. Previously observed tests and playground examples informed research and owner selection. They are not a fresh blind holdout. Synthetic prose shares construction families and training entities, so gains do not establish open-domain generalization. The selected checkpoint is a preserved no-pruning replay artifact identified by its exact hash; metrics are recomputed from that artifact's saved generations, not borrowed from an earlier artifact. No claim of bitwise training replay identity is made.
 
 ## 10. Failure audit and limitations
 
-All 19 newly regressed original TEST requests are schema-valid JSON. Manual primary categories: 4 changed/invented values, 3 shortened spans, 8 wrong field assignments, 2 omissions and 2 especially ambiguous source/schema contracts. Examples include Pete Murray becoming Peanut, a complete movie title losing its final phrase, a venue assigned as a movie, and an explicit city becoming null. A bare music title may not specify track versus album; broad entity_name definitions also conflict with exclusive annotation roles. Those ambiguity flags do not change benchmark golds or excuse clearer errors. Full-case evidence is retained in the report archive.
-
-Multiple-value coverage improves, but errors remain: Bergen and Oslo can generate Bergen, Orange and Lake. Correct JSON structure cannot establish correct extraction. The model is primarily SNIPS-domain research software; unfamiliar fields, long documents, corrections/associations, nested schemas, arbitrary-schema reliability, general chat/coding, multilingual behavior, confidence calibration and prompt-injection robustness remain unestablished. No 95% reliability or universal release-readiness claim is made.
+There are 173 entirely incorrect original-test requests and 24 previously correct requests become incorrect. Errors include incomplete copying, wrong roles, missing or extra values and unsupported values. All seven domain totals improve, but per-request retention is not perfect. On the six longer-list probes only three are correct. Flat schema validity is 100% on this test, not a guarantee for arbitrary prompts, schema types or novel domains. No general chat, reasoning, code, multilingual, long-document or safety readiness claim is made. The authored development panel is small and repeatedly observed.
 
 ## 11. Packaging, verification and reproduction
 
-The main package retains BF16 weights, frozen tokenizer and native MLX architecture. Stock-MLX standalone verification compares all 54 raw outputs and stop reasons against the selected saved playground protocol; the receipt specifies achieved coverage/parity and correctness. This is packaging parity, not a fresh blind accuracy estimate or complete 1,444-case cross-runtime parity claim.
+The main package uses BF16 weights, the unchanged tokenizer and the supplied native MLX runner. The verification receipt records stock-MLX raw-string and stop-reason parity on all 60 observed playground fixtures. This establishes packaging parity within the stated scope, not full 1,444-case cross-runtime parity or an independent benchmark. Standard `model.safetensors` and the public model ID remain unchanged. Previous Hub and Git revisions retain rollback identities.
 
-Hugging Face supplies canonical model.safetensors, config, tokenizer, launcher, native_bundle.zip and model/report/results/provenance records. GitHub ships runnable source, curated JSONL records, preparation/training/evaluation code and audit evidence. Optimizer/packed-array binaries, private workspace paths, credentials and host network logs are excluded. Original workspace manifests record excluded files; provenance/SNAPSHOT_MANIFEST.json lists the actual shipped source inventory. Sanitized metadata hashes are separate from original metadata hashes.
+The source archive includes the frozen latest training recipe, prepared JSONL corpus and input hashes, exact prefix counts, saved generated answers, scorer and matched parent comparisons. Public copies sanitize private workspace paths and distinguish original from distributed hashes. Optimizer binaries and packed pretraining pools are omitted. Reproduction requires the specified parent and layout, compatible dependencies, and fresh local preflight receipts; sanitized audit receipts are not resumable optimizer state. See TRAINING_AND_REPRODUCIBILITY.md, provenance/LINEAGE.json, results/JSONLITE_RESULTS.json and pretraining/native_json_128/ in the archive.
 
-The additional stage has frozen paths to historical baseline receipts and prepared-data inputs. Reproduction requires the specified parent weights, original source dependencies/layout, re-packing omitted arrays and regenerating local preflight receipts; published audit receipts are not ready-to-resume optimizer state. Native inference needs none of those training dependencies. The teacher is only used during training.
+## 12. Attribution and licensing
 
-See provenance/LINEAGE.json, results/JSONLITE_RESULTS.json, pretraining/native_json_105/, examples/cases.json and reports/model-design/SFT-105_cardinality_distillation_20261014/ inside the source archive. Prior native JSON release hashes/revisions remain in provenance/RELEASE_HISTORY.json; old Git/Hub revisions remain accessible.
 
-## 12. Attribution and licensing
 
 Source code retains Apache-2.0; upstream data and dependencies retain their notices. Original SNIPS annotations are CC0; the matched cardinality examples are programmatically authored from TRAIN literals. The earlier SQuAD/Dolly/pretraining source notices remain in THIRD_PARTY_NOTICES.md. Source licensing does not automatically assign a trained-weight license; none is assigned by this update.
 

THIRD_PARTY_NOTICES.md

@@ -18,3 +18,7 @@ MLX, NumPy, Tokenizers and jsonschema are installed dependencies, not vendored d
 No IFEval, Paraloq, SOB, ARC, PIQA, HellaSwag or other experiment's scores are advertised as evaluation of these weights. Older research helpers present in dependency packages do not change the inherited training lineage.
 
 For actual source-file pins and identity-preserving adaptations, see the technical report and `provenance/`. These declarations are attribution records, not a blanket commercial-use or model-weight redistribution determination. The public weight license remains unassigned.
+
+## Latest joint JSON extraction data
+
+The latest stage reuses SNIPS TRAIN source/schema/JSON views and authored controlled prose with original TRAIN values. Original SNIPS attribution and CC0 terms above continue to apply. Authored wording does not make evaluation independent or establish a new weight license. No SGD/MIT dataset was added to this stage.

checkpoint_metadata.json

@@ -1,168 +1,80 @@
 {
-  "extraction_sft": {
+  "joint128": {
     "contract": {
-      "code": {
-        "pretraining/extract_068/data.py": "497b0b12c8896c2045f470c1c2de0330472663f1fd5b67746734a3f7907ee971",
-        "pretraining/extract_068/evaluate.py": "a7ff05b0d2383421e83d0247e613017715485b3a5fffdcc7240b90387a756c06",
-        "pretraining/extract_068/runtime.py": "7b97df4835d03210c77da36f53340aa83d5915b9cfa7aae968aef6961422c2f1",
-        "pretraining/extract_068/train.py": "f44af1b00a7e36c3dd4c42c30dc63b73bc0d5917495f1985cc6bc3c5b902b6c5",
-        "pretraining/extraction_001/__init__.py": "f1e824c48131bc441020c864929875bfa44daebba75b15181cc401f7b73ae8a2",
-        "pretraining/extraction_001/data.py": "59843f3beca3aecbe8e905af8c969fa9fd6a42554da8d2b1054d027fb771a659",
-        "pretraining/extraction_001/evaluate.py": "b22da13e22b5ebffeb217200e5ff431773d85e6ae394045e96c97ed1bc7e7c54",
-        "pretraining/extraction_001/train.py": "071ef9293bb825753b47b3704e1df1e69f33b792c45c5748f226e40c6c054e26",
-        "pretraining/json_evidence_binding/__init__.py": "357162457d07712f819559c1f1a6769323941355531cb6a04f657ace44d5c52c",
-        "pretraining/json_evidence_binding/data.py": "fd828fcaa202085ea38a4a927e31f7bfc09e19a8965cd170f095408744e3a8a0",
-        "pretraining/json_evidence_binding/evaluate.py": "25fc997a4035ab44f6e5ac47988abc38886efb2c23d8cc270f014e8c0a04cc71",
-        "pretraining/json_evidence_binding/objective.py": "5bcfe1d57e445ca79a82daf81a8d664bb1c8178aaf3ff4189665e205a95537aa",
-        "pretraining/json_evidence_binding/order.py": "37759caee925d4eb2a6e8273075524090516017c199fbc9ebe75edee9b1a7cd6",
-        "pretraining/json_evidence_binding/runtime.py": "4d6695edd7eccfd26ca31a282e76ae663fe149b86f6fc42e8edc4e7625ca15ec",
-        "pretraining/json_evidence_binding/storage.py": "a93d3af323904983f016870865e90ee89a02df4c66d15e502ef9a2ebfff99424",
-        "pretraining/json_evidence_binding/train.py": "2e20b5a09501145a68500994eed4a8e5134148d805dafd93726ea09afca23ebf",
-        "pretraining/json_sft/__init__.py": "aa9dcf15611647559bbc4ab1d8dfa3746cf731f30e4b5fd35bf53a1ed00bea95",
-        "pretraining/json_sft/adapters.py": "09dd78767c9f7e0dbaa0184e64d0011afadd7a326986766026dfcc64106ab8d9",
-        "pretraining/json_sft/benchmark.py": "75ae0df6b0d6676805d6c1ef3b7b73893f473df001dcc2a542f31fb074dd3601",
-        "pretraining/json_sft/business.py": "5c31088adce5f85ce8980d159fa0f9c39a0e1d9a7725560838a2852ac5cfbff4",
-        "pretraining/json_sft/constrained.py": "dc21aa142fe525f3559cd5638e46ef287555efba8239a06072a092fc10700cf5",
-        "pretraining/json_sft/contrasts.py": "e504c5f879011963b82728d73fe334585a640d419c4e3d47d919a4cc89f396fd",
-        "pretraining/json_sft/corpus.py": "37d7b74e0478853903953934b5d2006fd059e1ca2a613cc69f6788e9f34d0e6d",
-        "pretraining/json_sft/decomposed.py": "46f863e628bcc5fedfd6f34f7c759d4b0998091d8252b48729027ce9ef2a01b9",
-        "pretraining/json_sft/diagnostics.py": "92f750807aebb9cb38154c4599d468850ce25ef86496de8e660a77e538e08f03",
-        "pretraining/json_sft/evaluate.py": "2c3076f69ef2d7721f25a79618563d102484bedbf2912f684b644b9135d25d17",
-        "pretraining/json_sft/expansion.py": "231274b7e6d486d39bf962ff310ad4557ccfd84f8cac329cf57ed5890aed7cc7",
-        "pretraining/json_sft/grounded.py": "f668379c5f904ee36c267f6897f810cdb04f80e19a8fd5fdfb791b31f902e8c9",
-        "pretraining/json_sft/narrative.py": "9d507dbc7c2eff7a4b77064393dc65e65c6e3a1866b4375df3d8b66c755d5be1",
-        "pretraining/json_sft/ordered.py": "0ac1b6dd3139425c06094cf5e185ffab5389a377ecaf58b6320978908df4b7ef",
-        "pretraining/json_sft/ordered_serialization.py": "14c23f875b45d17d8d07cb106afc29a6b72cb9b7bb5a393ee0ee55bf84b5aa1a",
-        "pretraining/json_sft/plain.py": "64dd1060d5a674ef04bba3ea3965b7e89ab4faa58bbc3c437816ed2023023843",
-        "pretraining/json_sft/precise.py": "7696328b55290548f05562a9562ea898301a992b4161000f5f2dd7d1bbaf621a",
-        "pretraining/json_sft/reader.py": "ff288fddabd0525f8fc6a8f2a6aeaedf309cfb954448d8c6aa57e122e8628d6e",
-        "pretraining/json_sft/relational.py": "d2e5d3f850b5b543876fafac8b3d57c3ea7c6983437ba6744026ed86701e929e",
-        "pretraining/json_sft/schedule.py": "50ec8a06b0f881714451501b6325a0a64c7cbff159b57f1214f0a3a6ac83014c",
-        "pretraining/json_sft/serialization.py": "d04338a44b569c3f38ed42b0e1135738c1c119387134945335d51bb354c2213f",
-        "pretraining/json_sft/serving.py": "d533a59791bd2516dd1ec4cd2493c554161c0fe1e3603d55fa12631491cddbf8",
-        "pretraining/json_sft/slots.py": "8db7eb31a53899b9e4306af97fad6219ff6c8054a0df67ceffe0dbe927e5df24",
-        "pretraining/json_sft/train.py": "abe0613af70c559dca777cc1fc3c905be0f9583fb6aa30442881da8d0a95f316",
-        "pretraining/json_sft/transfer.py": "ff5bc65fae8da93164f830cff5a20fcff42132bcbf02a94de672af9653cf340d",
-        "pretraining/json_sft/unquoted.py": "969263db40f1a5561acbafdfb21dbb74ce85dd394c2bc9309cd208f61c977289",
-        "pretraining/json_sft/value_weights.py": "043cde3b06edcbc0bd332eb8ace4257e1864af77c239204c120261b95696598b",
-        "pretraining/native_json_100/__init__.py": "8d4b1be2f7d75122c982cd881c19a41b1d39594c5e30fbbe2bc87fcccf9c0242",
-        "pretraining/native_json_100/data.py": "594a83b846c97836be009e61d9fcf9a1b4ceb0d0f9bb491bbb5b56f02a433edf",
-        "pretraining/native_json_100/evaluate.py": "3c03d63aeb85ee85d857f219ee3aa4c190f34532ab6643cb487ca8e41d9779c3",
-        "pretraining/native_json_100/runtime.py": "9b04f8b33fceb717fa6e9290a75bcc7c8365ad6162d3f59de19e08b913a034c4",
-        "pretraining/native_json_100/train.py": "63823c5bc7a4b44cb57d7cd4bead7e97ecf309b09816e4d81327a9b7d795ea72",
-        "pretraining/native_json_100/verify.py": "aa1452e3994eb534f30185c69ba1fcac5dc47081a51e547c31db132e8c369118",
-        "pretraining/native_json_105/__init__.py": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855",
-        "pretraining/native_json_105/data.py": "b4934d571851f934164a2f0dce310486dc70e45c8222f8ae5c91a9fd987be303",
-        "pretraining/native_json_105/evaluate.py": "bf439186ebf0819f54e2e1dafb120e3abc0e05b53e2367f3adb6abc09f66817b",
-        "pretraining/native_json_105/runtime.py": "cd00debd869794c50af2856e51437199bbe2221bfb1b7192e36332167ef1ee24",
-        "pretraining/native_json_105/train.py": "fec330fbcf8d6f3b76cc4c86a30bcf00f7215cbc57b1af31ebcc05c69f5c3d4d",
-        "pretraining/native_json_105/verify.py": "e21a616dba7b402aeaea40c204678fdbfc16dda97606d74179c7a213cfb57d78",
-        "pretraining/qa_064/data.py": "a5d86a1c73285f420c728b5e8331da468eb8846086a8bbe69ce61df6e64fef7f",
-        "pretraining/qa_064/evaluate.py": "b0b8bd42b4842065e3ef8ba3c1fee42c95464f0f8ec49d030e4ab4f1998f94ae",
-        "pretraining/qa_064/runtime.py": "7b97df4835d03210c77da36f53340aa83d5915b9cfa7aae968aef6961422c2f1",
-        "pretraining/qa_064/train.py": "9400aa40b0f5b332334e11fc8b62f5d06464fe1188603bb65585302461d3bf19",
-        "pretraining/qa_065/controlled.py": "55799050001b55a642705064bf79ce45625eca81ebfe77d7f70cba8435726f51",
-        "pretraining/qa_065/data.py": "1e4878ce7bd9e48d8d5c79eafac31093eaf7cdcdd4a6aeb5f9b6519967e47e57",
-        "pretraining/qa_065/evaluate.py": "f8e4a209a7cdb42cded3cb5ffdb2aabfa0d12cc13c10c2f40f33907c532f74e5",
-        "pretraining/qa_065/runtime.py": "7b97df4835d03210c77da36f53340aa83d5915b9cfa7aae968aef6961422c2f1",
-        "pretraining/qa_065/train.py": "52b11949018bc87c6c80dc3bee7571b2ca95967ebb47fe450a5f5552c46d4a16",
-        "pretraining/qa_066/controlled.py": "fc101c488049fd836e53422a72d13d8d02f253153726e9e4d5eda0eb19ab9baa",
-        "pretraining/qa_066/data.py": "4b89e4107bd7076270151989871805af4162d76a299e72d40881fdc26750015a",
-        "pretraining/qa_066/evaluate.py": "d998fb255daabc15b6070bffaf9adeeebdbe631e54be05b83b473f79b3c8642c",
-        "pretraining/qa_066/runtime.py": "7b97df4835d03210c77da36f53340aa83d5915b9cfa7aae968aef6961422c2f1",
-        "pretraining/qa_066/train.py": "3da0be266aa1f047addba8d0c696245ed71e4fde236ecfc8b383c69e8b32e783",
-        "pretraining/runs/sft_100_native_json_20261014/evaluations/development_0000469.identity.json": "77385825353ef560758a91527a4d47113b319694097fb62147d3e88cb123279e",
-        "pretraining/runs/sft_100_native_json_20261014/evaluations/development_0000469.json": "bdb12afb6ab0a6d0066894f3d18c8fe6e81aeaa4c4e7cfe579452d4273aa0585",
-        "pretraining/runs/sft_100_native_json_20261014/final-test-identity.json": "80ac5bd9eb25dcdd99cb1bf2f4cbaa20e0c79ef334b4c37e310af44a2cb2cb48",
-        "pretraining/runs/sft_100_native_json_20261014/final-test.json": "8f0efd027a57d34f8e22f2ac5d2d54b72a1103a2d853b3704514947559c2e3b0",
-        "pretraining/scripts/run_sft_105_cardinality_distillation.sh": "4d4ca56869958045f7504b08684d1610422cee323e146648af01c75a908b60eb",
+      "files": {
+        "pretraining/data/sft_128_joint_extraction_20261016/manifest.json": "5172c7702b2c0b6fa9583052007acb69630bed6b23b5fb4d3ce62d3666ca1a00",
+        "pretraining/json_sampling_119/common.py": "b3ef439b09b8ca21d9bb454a9afe377f202ea75bda522e51e7b284399b39d0c0",
+        "pretraining/json_sampling_119/data.py": "5e560b14b795a58f2ca2cbe6d7fb19924b8ec633788cdef8875d80449e3734a4",
+        "pretraining/json_sampling_119/run.py": "84fe9caab3c196fec5e71c56a1c4ab8e7c63309fe4529811680d72e4a5cec747",
+        "pretraining/native_json_128/__init__.py": "fcdbfacd29cfb9f3235160fe00d4c9f4f4c35cb27d469f1e6f3796f137b373b2",
+        "pretraining/native_json_128/common.py": "354cda3dd100321fa858e64ae22e4ecc27e09d97f2c824811e3b9f53f205be52",
+        "pretraining/native_json_128/data.py": "874ab8eae8281b1967ea7ed22a0bad122e0c830a7ee8391ae9d201e0d6867087",
+        "pretraining/native_json_128/evaluate.py": "2dc61a8269c7c279ca13814f7e0b07ddbee477308ac5abc87d1461ca75cfedc8",
+        "pretraining/native_json_128/runtime.py": "5a83f2ba3f900f69390d10c2a67d10cd2d7654a1b22345f059a02488b0239ef8",
+        "pretraining/native_json_128/scoring.py": "1f4d2c69ebc640f993b77cb09ecf280d2df1dbf3f20d1060f7c8a5ce6780dd97",
+        "pretraining/native_json_128/train.py": "dded6b80814d12f621eafe0ddcc41481fa80cc98a2182be442fb09c31a5735e7",
+        "pretraining/native_json_128/verify.py": "738c709a8358a0818f6dbd42d81a9db25ea76569cededae345e1a0d474344779",
+        "pretraining/runs/sft_105_cardinality_distillation_20261014/step_0000064_cdd8be1e8d5a/model.safetensors.json": "8b8c0049d731f8124eaec78a213d05c24a520bc4f190dc82975d763f43351bbc",
+        "pretraining/scripts/run_sft_128_joint_extraction.sh": "dcf97394bf39afd46c5591bcc19abd8802422d4ab3e5d9fd8845c5e46f493ead",
         "pretraining/sft/data.py": "70348a9df1473da88330695d5409f17b929b9bc1dc79f77525272f6d87317aa1",
         "pretraining/sft/evaluate.py": "80624551512112850145ccb57d6435d7a1bae86d0b95f8f52d56b8cb6f102a76",
         "pretraining/sft/quality.py": "2abf5926a22a33b4f29c699855b24f7e459b6e631aa6ef780876ddc205f17758",
-        "pretraining/sml_v1/__init__.py": "7e9e02294e03e6b8a12fd202ab288f2f9b0073bb11b8cb2a5fa82211b0d1dfa4",
-        "pretraining/sml_v1/batch_transition.py": "e6778ba8a8215f0b3c20386e768ea3e15b5939e7919935046689733eb96eba28",
         "pretraining/sml_v1/checkpoint_bundle.py": "73cbf0c17cf74a44bb7c74a8f9e5bb63671057a1f664a6a4af0e0777b2c8f1a4",
-        "pretraining/sml_v1/checkpoint_metadata.py": "0aaa7019ab53fd8b5e6b0e397f970026b4e8d38e12bf5d5f0fcb7f686de9b82c",
         "pretraining/sml_v1/common.py": "ec4dda551cbe45cbd30d64c7c82379a8849ce02a3ad1a597a293844b5dc86351",
-        "pretraining/sml_v1/continuation.py": "3c5248646c27bcdf6bd4010742fe6e6d0aa9f059b543e183b60e00e3dbe034a3",
-        "pretraining/sml_v1/corpus.py": "be77769180342f6a8b1d5619e5b3b64f20ca40d3424219168cccf4789bb02038",
-        "pretraining/sml_v1/data.py": "411284cda81b07f10bc624ebd8f7fe9debd2911e950971f96b204dfd1cee6a6d",
-        "pretraining/sml_v1/data_review.py": "a71eabcc526344b6502cae999020da63c950b2d80c13077629bb86c8757a18ed",
-        "pretraining/sml_v1/diagnostic_eval.py": "317c05d4d17239e62d1316797892288f5e4eae467b3e66a4ca4f15d189960a9b",
         "pretraining/sml_v1/fast_loss.py": "f99edeedb6cb328ef728d05643f65ea20371437f7e5806009ae3ad0885bafbf1",
         "pretraining/sml_v1/fast_swiglu.py": "59d1adb31400e3c288914b4c0cc8a21681a73ca3dfa7c722ea4e13d21dc38528",
-        "pretraining/sml_v1/inference.py": "4255b93bbc18d120318d83fb26b2fda1fe16d19ca1423250c77188940cc4f008",
-        "pretraining/sml_v1/jaccl_control.py": "a3dfbd61de43fda14daf4bbd921ccb382ea8af2c8342457fe16d1ad0908e88fe",
-        "pretraining/sml_v1/launch.py": "41e6a98d9b41dc844063fc28ce81a08d83b919306b9daf0716b76a0bc21365fc",
-        "pretraining/sml_v1/lr_scheduler.py": "e141132efbd5f1beab4b187d4eab15efd46fcf9710d6a2ba7c882f62362a77cf",
-        "pretraining/sml_v1/lr_trial.py": "2f643abaa6ac3f86e47b37632df4d25fa34d0b25e77b59ffecc0290b2385e7dc",
         "pretraining/sml_v1/model.py": "1eee6f8afe78767ff6b9c604ccf2328c98b24bc94dbad6291a0ad699dd7345fe",
-        "pretraining/sml_v1/playground_publish.py": "bffd8f8c5a74a0bf9c3ca070e6511cc1ab8ceb8343c0092a3d711b4fb9ece21c",
         "pretraining/sml_v1/precision.py": "b9f535fca724153703ed57a300eceb5b34a8f8381fc65bf5c4c5a0f7e066303c",
-        "pretraining/sml_v1/prepare_data.py": "3d78ed45dbab52a671d889fe937a210a832e55426bbdc4f905ddbd202ccf2336",
-        "pretraining/sml_v1/pretrain.py": "456a498d50e991cd83363588392b91e58a8bdeec7ceaf9950671925b6a86a46b",
-        "pretraining/sml_v1/recipe.py": "d354731910ac16868b21717a089fce5199a898dec39c6561b3d10c89bfa8a2d1",
-        "pretraining/sml_v1/research_stage_b.py": "5234bc6b824e4b07f63dda48e5353bbdb5f5abac4c98c299b642d1815ce85d94",
-        "pretraining/sml_v1/sample_prompts.py": "1173c8c94dc11c6b8e446c26804a543b68e366986f01d249384082629dca369b",
-        "pretraining/sml_v1/stable_schedule.py": "631451a45b020120610b7116e8e63276ba5043011359fbd13e49e63ec8e2b02b",
-        "pretraining/sml_v1/stage_b.py": "f80f065f6892eea951e94a7ff6925d5a1bff9419a7384e0756cf2faeb92e4775",
-        "pretraining/sml_v1/stage_b_data.py": "c065be118b0c6b8948a15d4b81e7fd7371f4e699ea3c49d34fa0abde258e0505",
-        "pretraining/sml_v1/stage_b_eval.py": "943fd969994e896faff14481b94eb121d23b6fef00244874d5d5361df6bd8cce",
-        "pretraining/sml_v1/stream.py": "70e36e12d33547bfce505bd0302605e3d66c9ee65fc0facb26ccf5341bde8b4b",
         "pretraining/sml_v1/tokenization.py": "254e1f5d60cc4498a9f1209955752db28b91839f48d3121d674f3f5ac42e52e5",
-        "pretraining/sml_v1/topology_transition.py": "adc3a0a6f19e7cfb3073f4c1e0432f664e5fb709141ceb4148aca7cda08ddaab",
-        "pretraining/sml_v1/train_tokenizer.py": "90c58a30a50cb625e36222825768a134aba5c81df87de8655ab95de2bd47402c",
-        "pretraining/tests/test_native_json_105.py": "698d9410a900dee6b81dede19f20ea06b721692b2d5a09ebb4ee646d9786701a",
+        "pretraining/tests/test_native_json_128.py": "7547e08bf1f06d87bdcb34197b412ae9555c9e3a503b3d608d21c27e9f423c3c",
         "reference/opensml-150m/tokenizer/bytebpe32k_v1/corpus.json": "e4c47b7b6aba08f948de7c2c7cb94c58dbd09dde2ded6ac185c60856ee686d27",
         "reference/opensml-150m/tokenizer/bytebpe32k_v1/manifest.json": "1194292c6d906ac19e6f01c6ad2b42825143fad426a22aa8856d903776701ecb",
         "reference/opensml-150m/tokenizer/bytebpe32k_v1/report.json": "e871f690eabebf82da173f0e406f38b79a10cd7d2647bb3592c3fc35ae9a2f16",
         "reference/opensml-150m/tokenizer/bytebpe32k_v1/tokenizer.json": "e095f0413a11fa91fbb58f37d00e4c6bcda59beb89b28f6f680db1436a7e8891",
         "reference/opensml-150m/tokenizer/bytebpe32k_v1/tokenizer_config.json": "619f500553367eb76f5be00a98625130ec857b9c367561334984096c029ba2a9",
-        "reports/model-design/SFT-105_cardinality_distillation_20261014/parent-development.json": "581a70411556d0af2e9aa6868af29c614d89f94012179c6b14b897b0afa0dfe0"
+        "reports/model-design/SFT-105_cardinality_distillation_20261014/original-test-1444-step64.identity.json": "e72aeb0100d8fd145096cc8df88b20c23a4630a3c2e4b3114fa9935f10dbc96b",
+        "reports/model-design/SFT-105_cardinality_distillation_20261014/original-test-1444-step64.json": "2c730e69196aa4177ccc7bf1b09bde7a044597b7f2afde108344c9c4579a0c44",
+        "reports/model-design/SFT-128_joint_extraction_20261016/source-validation.json": "b37530d0ef3716a15e977a949a31dcca953bdb5abf32513885e8285b4f810314"
       },
-      "manifest": "3680d620c3cefd00473ee5fcabe60b5e3a1351ef5b2d8c4ef763efec0081a218",
-      "parent": "[workspace]/pretraining/runs/sft_100_native_json_20261014/step_0000469_25528943984d",
-      "parent_metadata": "3864a0eaa43015d286e4da9327b0d3cea4fa4f5b80f336baabe2a549379e2e36",
-      "parent_sha256": "656f45015aab548465d93bed0b57e687497163741885da7364872851fa17e9a4",
+      "parent_sha256": "655fb5cfba009163cd8a7f7fe7b170b998bb452c217b1b473cfe10b33de0ccd1",
       "plan": {
-        "batch_size": 32,
+        "adaptation": "all_parameters",
+        "batch_size": 64,
         "betas": [
           0.9,
           0.95
         ],
-        "distillation_temperature": 1.0,
+        "clip": 1.0,
+        "context": 2048,
         "epochs": 1,
         "eps": 1e-08,
-        "eval_steps": [
-          0,
-          32,
-          64,
-          96
-        ],
-        "final_lr": 2e-07,
-        "grad_clip": 1.0,
+        "eval_every": 100,
+        "final_lr": 1e-06,
         "max_tokens": 256,
         "microbatch": 2,
-        "order_policy": "prepared_prefix",
-        "padding_policy": "micro_multiple128",
-        "peak_lr": 2e-06,
-        "replay_kl_weight": 1.0,
-        "seed": 20261119,
-        "total_updates": 96,
-        "warmup_updates": 6,
-        "weight_decay": 0.01
+        "objective": "Mean per-request assistant-JSON token cross entropy, including turn EOS. No teacher KL, no inference repair or assembly.",
+        "order": "One deterministic joint shuffle, all prepared records exactly once; no late domain-only phase.",
+        "peak_lr": 5e-06,
+        "precision": "BF16 compute / FP32 master AdamW",
+        "seed": 202610160128,
+        "warmup": 20,
+        "weight_decay": 0.0
       },
       "versions": {
         "jsonschema": "4.26.0",
-        "mlx": "0.32.3.dev20260920+7241f12",
+        "mlx": "0.32.3",
+        "mlx-metal": "0.32.3",
         "numpy": "2.4.2",
         "tokenizers": "0.22.2"
       }
     },
-    "order_hash": "8e0ef3d752534f0aa8e45465436974b77384cc65b3facba7f32c8bc29bcf0750",
-    "records": 2048,
-    "serialization": "native-sft-chat-v1",
-    "total_steps": 96
+    "order_hash": "8d3c7c9d4a1e34303d2f6c6b411f50f8cc72a72f45a3dd6a75c7a60f88715be3",
+    "records": 19200,
+    "total": 331
+  },
+  "preservation_policy": {
+    "pruning": false,
+    "user_requested": true
   },
   "recipe": {
     "model": {
@@ -183,6 +95,6 @@
       "vocab_size": 32000
     }
   },
-  "step": 64,
-  "tokens": 600558
+  "step": 300,
+  "tokens": 633561
 }

checkpoint_provenance.json

@@ -57,14 +57,27 @@
       "weight_bytes": 351242342,
       "original_metadata_sha256": "8b8c0049d731f8124eaec78a213d05c24a520bc4f190dc82975d763f43351bbc",
       "published_metadata_sha256": "49d177fb4606673f789fbf06c694643c44504672a9f0c3f75f7eae776cb4590c"
+    },
+    {
+      "stage": "sft-128",
+      "public_stage": "joint JSON extraction SFT",
+      "step": 300,
+      "records": 19200,
+      "serialized_positions": 6153194,
+      "supervised_assistant_positions": 633561,
+      "token_counter": "supervised assistant positions; differs from earlier serialized counters",
+      "weight_sha256": "781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560",
+      "weight_bytes": 351242342,
+      "original_metadata_sha256": "663087d32872824f7f2fec425ad4054f308913c43da7c3f83521c7011967fdbe",
+      "published_metadata_sha256": "663087d32872824f7f2fec425ad4054f308913c43da7c3f83521c7011967fdbe"
     }
   ],
   "weights_in_git": false,
   "public_weight_location": "https://huggingface.co/OpenSML/OpenSML-JSONLite-175M",
   "model_weight_license": "not yet assigned",
-  "scope": "Stage A -> grounded QA -> requested-field extraction -> focused extraction -> native JSON foundation -> cardinality/replay distillation continuation. Only the selected prefix of the final continuation is inherited; complete historical update counts are in stages.",
-  "selected_stage": "additional extraction SFT",
-  "selected_update": 64,
+  "scope": "Selected lineage: Stage A -> grounded QA -> requested-field extraction -> focused extraction -> native JSON foundation -> cardinality/replay distillation -> joint JSON extraction SFT. Later updates are not inherited.",
+  "selected_stage": "joint JSON extraction SFT",
+  "selected_update": 300,
   "weight_location": "https://huggingface.co/OpenSML/OpenSML-JSONLite-175M",
   "public_checkpoint_name": "main"
 }

config.json

@@ -18,7 +18,7 @@
     "rope_base": 10000.0,
     "vocab_size": 32000
   },
-  "weight_sha256": "655fb5cfba009163cd8a7f7fe7b170b998bb452c217b1b473cfe10b33de0ccd1",
+  "weight_sha256": "781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560",
   "serialization": "native-sft-chat-v1",
   "output_mode": "native-json",
   "context_limit": 2048,
@@ -29,8 +29,8 @@
   "json_assembly": false,
   "output_repair": false,
   "constrained_decoding": false,
-  "selected_stage": "additional extraction SFT",
-  "selected_update": 64,
+  "selected_stage": "joint JSON extraction SFT",
+  "selected_update": 300,
   "checkpoint_name": "main",
   "inference_teacher_required": false
 }

inference.py

@@ -12,7 +12,7 @@ import tempfile
 import zipfile
 
 ROOT = Path(__file__).resolve().parent
-BUNDLE_SHA256 = "a1b63ed6286d6c4f8bc4f7526f4a11fc46f936830daaac22224bfa0c4892755b"
+BUNDLE_SHA256 = "c54fd408f6b7ec0c6e80377ee8b6eeffe3bfbf76389e56fdad4c78f54828e2d1"
 
 def main():
     archive = ROOT / "native_bundle.zip"

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:655fb5cfba009163cd8a7f7fe7b170b998bb452c217b1b473cfe10b33de0ccd1
+oid sha256:781597baeb9b51d535fc27cd6f941dbc202e01d62c8d963587a3f3d565128560
 size 351242342

native_bundle.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a1b63ed6286d6c4f8bc4f7526f4a11fc46f936830daaac22224bfa0c4892755b
-size 7638568
+oid sha256:c54fd408f6b7ec0c6e80377ee8b6eeffe3bfbf76389e56fdad4c78f54828e2d1
+size 12831619