@tuned-tensor/local 0.2.9 → 0.4.0

package/dist/contracts.d.ts

@@ -1,36 +1,15 @@
 import { z } from "zod";
-export declare const trainingMethodSchema: z.ZodEnum<{
-    sft: "sft";
-    dpo: "dpo";
-}>;
-export declare const datasetFormatSchema: z.ZodEnum<{
-    chat_jsonl: "chat_jsonl";
-    document_ocr_chat_jsonl: "document_ocr_chat_jsonl";
-    preference_jsonl: "preference_jsonl";
-}>;
-export declare const documentInputAssetSchema: z.ZodObject<{
-    type: z.ZodDefault<z.ZodLiteral<"image">>;
-    mime_type: z.ZodOptional<z.ZodString>;
-    image: z.ZodOptional<z.ZodString>;
-    data_uri: z.ZodOptional<z.ZodString>;
-    uri: z.ZodOptional<z.ZodString>;
-    path: z.ZodOptional<z.ZodString>;
-    page: z.ZodOptional<z.ZodNumber>;
-}, z.core.$strip>;
+/**
+ * TT Local deliberately has one training contract today: text supervised
+ * fine-tuning with a LoRA adapter. Additional methods should only be added
+ * after they have their own end-to-end CUDA acceptance test.
+ */
+export declare const datasetFormatSchema: z.ZodLiteral<"chat_jsonl">;
+export declare const baseModelRevisionSchema: z.ZodPipe<z.ZodString, z.ZodTransform<string, string>>;
 export declare const behaviorSpecExampleSchema: z.ZodObject<{
     input: z.ZodString;
     output: z.ZodString;
-    input_assets: z.ZodOptional<z.ZodArray<z.ZodObject<{
-        type: z.ZodDefault<z.ZodLiteral<"image">>;
-        mime_type: z.ZodOptional<z.ZodString>;
-        image: z.ZodOptional<z.ZodString>;
-        data_uri: z.ZodOptional<z.ZodString>;
-        uri: z.ZodOptional<z.ZodString>;
-        path: z.ZodOptional<z.ZodString>;
-        page: z.ZodOptional<z.ZodNumber>;
-    }, z.core.$strip>>>;
-    modality: z.ZodOptional<z.ZodLiteral<"document_ocr">>;
-}, z.core.$strip>;
+}, z.core.$strict>;
 export declare const specSnapshotSchema: z.ZodObject<{
     name: z.ZodString;
     description: z.ZodDefault<z.ZodString>;
@@ -39,30 +18,16 @@ export declare const specSnapshotSchema: z.ZodObject<{
     examples: z.ZodDefault<z.ZodArray<z.ZodObject<{
         input: z.ZodString;
         output: z.ZodString;
-        input_assets: z.ZodOptional<z.ZodArray<z.ZodObject<{
-            type: z.ZodDefault<z.ZodLiteral<"image">>;
-            mime_type: z.ZodOptional<z.ZodString>;
-            image: z.ZodOptional<z.ZodString>;
-            data_uri: z.ZodOptional<z.ZodString>;
-            uri: z.ZodOptional<z.ZodString>;
-            path: z.ZodOptional<z.ZodString>;
-            page: z.ZodOptional<z.ZodNumber>;
-        }, z.core.$strip>>>;
-        modality: z.ZodOptional<z.ZodLiteral<"document_ocr">>;
-    }, z.core.$strip>>>;
+    }, z.core.$strict>>>;
     constraints: z.ZodDefault<z.ZodArray<z.ZodString>>;
     base_model: z.ZodPipe<z.ZodString, z.ZodTransform<string, string>>;
-}, z.core.$strip>;
+}, z.core.$strict>;
 export declare const datasetPrebuiltSchema: z.ZodObject<{
     training: z.ZodString;
     validation: z.ZodOptional<z.ZodString>;
     test: z.ZodOptional<z.ZodString>;
-    format: z.ZodOptional<z.ZodEnum<{
-        chat_jsonl: "chat_jsonl";
-        document_ocr_chat_jsonl: "document_ocr_chat_jsonl";
-        preference_jsonl: "preference_jsonl";
-    }>>;
-}, z.core.$strip>;
+    format: z.ZodDefault<z.ZodLiteral<"chat_jsonl">>;
+}, z.core.$strict>;
 export declare const fineTuneHyperparametersSchema: z.ZodObject<{
     n_epochs: z.ZodDefault<z.ZodNumber>;
     learning_rate: z.ZodOptional<z.ZodNumber>;
@@ -72,39 +37,14 @@ export declare const fineTuneHyperparametersSchema: z.ZodObject<{
     lora_dropout: z.ZodOptional<z.ZodNumber>;
     max_seq_length: z.ZodOptional<z.ZodNumber>;
     gradient_accumulation_steps: z.ZodOptional<z.ZodNumber>;
-    save_adapter_only: z.ZodDefault<z.ZodBoolean>;
-    augment: z.ZodDefault<z.ZodBoolean>;
-    use_llm_judge: z.ZodDefault<z.ZodBoolean>;
     max_eval_examples: z.ZodOptional<z.ZodNumber>;
-    chat_template_kwargs: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
-    dpo_beta: z.ZodOptional<z.ZodNumber>;
-    dpo_loss_type: z.ZodOptional<z.ZodString>;
-    dpo_label_smoothing: z.ZodOptional<z.ZodNumber>;
-    dpo_reference_free: z.ZodOptional<z.ZodBoolean>;
-    max_prompt_length: z.ZodOptional<z.ZodNumber>;
-    max_completion_length: z.ZodOptional<z.ZodNumber>;
-    base_model_revision: z.ZodOptional<z.ZodString>;
-    parent_model_artifact: z.ZodOptional<z.ZodString>;
-}, z.core.$loose>;
-export declare const modelArtifactMetadataSchema: z.ZodObject<{
-    framework: z.ZodOptional<z.ZodString>;
-    format: z.ZodOptional<z.ZodString>;
-    entrypoint: z.ZodOptional<z.ZodString>;
-    servable: z.ZodOptional<z.ZodBoolean>;
-    notes: z.ZodOptional<z.ZodString>;
-}, z.core.$loose>;
-export declare const localArtifactsSchema: z.ZodObject<{
-    prefix: z.ZodOptional<z.ZodString>;
-}, z.core.$strip>;
+    base_model_revision: z.ZodOptional<z.ZodPipe<z.ZodString, z.ZodTransform<string, string>>>;
+}, z.core.$strict>;
 export declare const fineTuneRunRequestSchema: z.ZodObject<{
     run_id: z.ZodString;
     user_id: z.ZodString;
     behavior_spec_id: z.ZodString;
     run_number: z.ZodNumber;
-    training_method: z.ZodDefault<z.ZodEnum<{
-        sft: "sft";
-        dpo: "dpo";
-    }>>;
     spec_snapshot: z.ZodObject<{
         name: z.ZodString;
         description: z.ZodDefault<z.ZodString>;
@@ -113,20 +53,10 @@ export declare const fineTuneRunRequestSchema: z.ZodObject<{
         examples: z.ZodDefault<z.ZodArray<z.ZodObject<{
             input: z.ZodString;
             output: z.ZodString;
-            input_assets: z.ZodOptional<z.ZodArray<z.ZodObject<{
-                type: z.ZodDefault<z.ZodLiteral<"image">>;
-                mime_type: z.ZodOptional<z.ZodString>;
-                image: z.ZodOptional<z.ZodString>;
-                data_uri: z.ZodOptional<z.ZodString>;
-                uri: z.ZodOptional<z.ZodString>;
-                path: z.ZodOptional<z.ZodString>;
-                page: z.ZodOptional<z.ZodNumber>;
-            }, z.core.$strip>>>;
-            modality: z.ZodOptional<z.ZodLiteral<"document_ocr">>;
-        }, z.core.$strip>>>;
+        }, z.core.$strict>>>;
         constraints: z.ZodDefault<z.ZodArray<z.ZodString>>;
         base_model: z.ZodPipe<z.ZodString, z.ZodTransform<string, string>>;
-    }, z.core.$strip>;
+    }, z.core.$strict>;
     hyperparameters: z.ZodDefault<z.ZodObject<{
         n_epochs: z.ZodDefault<z.ZodNumber>;
         learning_rate: z.ZodOptional<z.ZodNumber>;
@@ -136,34 +66,16 @@ export declare const fineTuneRunRequestSchema: z.ZodObject<{
         lora_dropout: z.ZodOptional<z.ZodNumber>;
         max_seq_length: z.ZodOptional<z.ZodNumber>;
         gradient_accumulation_steps: z.ZodOptional<z.ZodNumber>;
-        save_adapter_only: z.ZodDefault<z.ZodBoolean>;
-        augment: z.ZodDefault<z.ZodBoolean>;
-        use_llm_judge: z.ZodDefault<z.ZodBoolean>;
         max_eval_examples: z.ZodOptional<z.ZodNumber>;
-        chat_template_kwargs: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
-        dpo_beta: z.ZodOptional<z.ZodNumber>;
-        dpo_loss_type: z.ZodOptional<z.ZodString>;
-        dpo_label_smoothing: z.ZodOptional<z.ZodNumber>;
-        dpo_reference_free: z.ZodOptional<z.ZodBoolean>;
-        max_prompt_length: z.ZodOptional<z.ZodNumber>;
-        max_completion_length: z.ZodOptional<z.ZodNumber>;
-        base_model_revision: z.ZodOptional<z.ZodString>;
-        parent_model_artifact: z.ZodOptional<z.ZodString>;
-    }, z.core.$loose>>;
-    artifacts: z.ZodOptional<z.ZodObject<{
-        prefix: z.ZodOptional<z.ZodString>;
-    }, z.core.$strip>>;
+        base_model_revision: z.ZodOptional<z.ZodPipe<z.ZodString, z.ZodTransform<string, string>>>;
+    }, z.core.$strict>>;
     dataset_prebuilt: z.ZodOptional<z.ZodObject<{
         training: z.ZodString;
         validation: z.ZodOptional<z.ZodString>;
         test: z.ZodOptional<z.ZodString>;
-        format: z.ZodOptional<z.ZodEnum<{
-            chat_jsonl: "chat_jsonl";
-            document_ocr_chat_jsonl: "document_ocr_chat_jsonl";
-            preference_jsonl: "preference_jsonl";
-        }>>;
-    }, z.core.$strip>>;
-}, z.core.$strip>;
+        format: z.ZodDefault<z.ZodLiteral<"chat_jsonl">>;
+    }, z.core.$strict>>;
+}, z.core.$strict>;
 export declare const localBehaviorSpecFileSchema: z.ZodObject<{
     name: z.ZodString;
     description: z.ZodDefault<z.ZodString>;
@@ -172,26 +84,10 @@ export declare const localBehaviorSpecFileSchema: z.ZodObject<{
     examples: z.ZodDefault<z.ZodArray<z.ZodObject<{
         input: z.ZodString;
         output: z.ZodString;
-        input_assets: z.ZodOptional<z.ZodArray<z.ZodObject<{
-            type: z.ZodDefault<z.ZodLiteral<"image">>;
-            mime_type: z.ZodOptional<z.ZodString>;
-            image: z.ZodOptional<z.ZodString>;
-            data_uri: z.ZodOptional<z.ZodString>;
-            uri: z.ZodOptional<z.ZodString>;
-            path: z.ZodOptional<z.ZodString>;
-            page: z.ZodOptional<z.ZodNumber>;
-        }, z.core.$strip>>>;
-        modality: z.ZodOptional<z.ZodLiteral<"document_ocr">>;
-    }, z.core.$strip>>>;
+    }, z.core.$strict>>>;
     constraints: z.ZodDefault<z.ZodArray<z.ZodString>>;
     base_model: z.ZodPipe<z.ZodString, z.ZodTransform<string, string>>;
     id: z.ZodOptional<z.ZodString>;
-    user_id: z.ZodDefault<z.ZodString>;
-    run_number: z.ZodDefault<z.ZodNumber>;
-    training_method: z.ZodDefault<z.ZodEnum<{
-        sft: "sft";
-        dpo: "dpo";
-    }>>;
     hyperparameters: z.ZodOptional<z.ZodObject<{
         n_epochs: z.ZodDefault<z.ZodNumber>;
         learning_rate: z.ZodOptional<z.ZodNumber>;
@@ -201,34 +97,16 @@ export declare const localBehaviorSpecFileSchema: z.ZodObject<{
         lora_dropout: z.ZodOptional<z.ZodNumber>;
         max_seq_length: z.ZodOptional<z.ZodNumber>;
         gradient_accumulation_steps: z.ZodOptional<z.ZodNumber>;
-        save_adapter_only: z.ZodDefault<z.ZodBoolean>;
-        augment: z.ZodDefault<z.ZodBoolean>;
-        use_llm_judge: z.ZodDefault<z.ZodBoolean>;
         max_eval_examples: z.ZodOptional<z.ZodNumber>;
-        chat_template_kwargs: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
-        dpo_beta: z.ZodOptional<z.ZodNumber>;
-        dpo_loss_type: z.ZodOptional<z.ZodString>;
-        dpo_label_smoothing: z.ZodOptional<z.ZodNumber>;
-        dpo_reference_free: z.ZodOptional<z.ZodBoolean>;
-        max_prompt_length: z.ZodOptional<z.ZodNumber>;
-        max_completion_length: z.ZodOptional<z.ZodNumber>;
-        base_model_revision: z.ZodOptional<z.ZodString>;
-        parent_model_artifact: z.ZodOptional<z.ZodString>;
-    }, z.core.$loose>>;
-    artifacts: z.ZodOptional<z.ZodObject<{
-        prefix: z.ZodOptional<z.ZodString>;
-    }, z.core.$strip>>;
+        base_model_revision: z.ZodOptional<z.ZodPipe<z.ZodString, z.ZodTransform<string, string>>>;
+    }, z.core.$strict>>;
     dataset_prebuilt: z.ZodOptional<z.ZodObject<{
         training: z.ZodString;
         validation: z.ZodOptional<z.ZodString>;
         test: z.ZodOptional<z.ZodString>;
-        format: z.ZodOptional<z.ZodEnum<{
-            chat_jsonl: "chat_jsonl";
-            document_ocr_chat_jsonl: "document_ocr_chat_jsonl";
-            preference_jsonl: "preference_jsonl";
-        }>>;
-    }, z.core.$strip>>;
-}, z.core.$strip>;
+        format: z.ZodDefault<z.ZodLiteral<"chat_jsonl">>;
+    }, z.core.$strict>>;
+}, z.core.$strict>;
 export declare const evalExampleResultSchema: z.ZodObject<{
     prompt: z.ZodString;
     expected: z.ZodString;
@@ -238,13 +116,11 @@ export declare const evalExampleResultSchema: z.ZodObject<{
     reasoning: z.ZodNullable<z.ZodString>;
     latency_ms: z.ZodNumber;
     scored_by: z.ZodOptional<z.ZodEnum<{
-        llm_judge: "llm_judge";
         exact_match: "exact_match";
         json_fields: "json_fields";
         heuristic: "heuristic";
-        exact_match_fallback: "exact_match_fallback";
     }>>;
-}, z.core.$strip>;
+}, z.core.$strict>;
 export declare const jsonFieldMetricsSchema: z.ZodObject<{
     fields: z.ZodArray<z.ZodString>;
     valid_json_count: z.ZodNumber;
@@ -257,14 +133,14 @@ export declare const jsonFieldMetricsSchema: z.ZodObject<{
         correct: z.ZodNumber;
         total: z.ZodNumber;
         accuracy: z.ZodNumber;
-    }, z.core.$strip>>;
-}, z.core.$strip>;
+    }, z.core.$strict>>;
+}, z.core.$strict>;
 export declare const evalSplitSchema: z.ZodEnum<{
     spec_examples: "spec_examples";
     spec_holdout: "spec_holdout";
     prebuilt_test: "prebuilt_test";
     prebuilt_validation: "prebuilt_validation";
-    prebuilt_training: "prebuilt_training";
+    general_regression: "general_regression";
 }>;
 export declare const evalReportSchema: z.ZodObject<{
     kind: z.ZodEnum<{
@@ -281,7 +157,7 @@ export declare const evalReportSchema: z.ZodObject<{
         spec_holdout: "spec_holdout";
         prebuilt_test: "prebuilt_test";
         prebuilt_validation: "prebuilt_validation";
-        prebuilt_training: "prebuilt_training";
+        general_regression: "general_regression";
     }>>;
     eval_sample_seed: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
     avg_score: z.ZodNumber;
@@ -289,9 +165,6 @@ export declare const evalReportSchema: z.ZodObject<{
     exact_match_rate: z.ZodNumber;
     avg_token_f1: z.ZodOptional<z.ZodNumber>;
     avg_latency_ms: z.ZodNumber;
-    judge_scored_count: z.ZodOptional<z.ZodNumber>;
-    fallback_scored_count: z.ZodOptional<z.ZodNumber>;
-    judge_only_avg_score: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
     cached: z.ZodOptional<z.ZodBoolean>;
     cache_key: z.ZodOptional<z.ZodString>;
     results: z.ZodArray<z.ZodObject<{
@@ -303,30 +176,22 @@ export declare const evalReportSchema: z.ZodObject<{
         reasoning: z.ZodNullable<z.ZodString>;
         latency_ms: z.ZodNumber;
         scored_by: z.ZodOptional<z.ZodEnum<{
-            llm_judge: "llm_judge";
             exact_match: "exact_match";
             json_fields: "json_fields";
             heuristic: "heuristic";
-            exact_match_fallback: "exact_match_fallback";
         }>>;
-    }, z.core.$strip>>;
+    }, z.core.$strict>>;
     artifact_uri: z.ZodString;
     scoring_method: z.ZodEnum<{
-        command: "command";
-        llm_judge: "llm_judge";
         exact_match: "exact_match";
         json_fields: "json_fields";
         heuristic: "heuristic";
     }>;
-    judge_model_id: z.ZodOptional<z.ZodNullable<z.ZodString>>;
     inference_provider: z.ZodOptional<z.ZodEnum<{
-        command: "command";
         none: "none";
-        batch_command: "batch_command";
         transformers: "transformers";
     }>>;
     scoring_mode: z.ZodOptional<z.ZodEnum<{
-        llm_judge: "llm_judge";
         exact_match: "exact_match";
         json_fields: "json_fields";
     }>>;
@@ -342,67 +207,34 @@ export declare const evalReportSchema: z.ZodObject<{
             correct: z.ZodNumber;
             total: z.ZodNumber;
             accuracy: z.ZodNumber;
-        }, z.core.$strip>>;
-    }, z.core.$strip>>;
+        }, z.core.$strict>>;
+    }, z.core.$strict>>;
     generation_config: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
     log_uri: z.ZodOptional<z.ZodString>;
-}, z.core.$strip>;
-export declare const regressionCategorySchema: z.ZodEnum<{
-    fallback: "fallback";
-    other: "other";
-    style: "style";
-    factual: "factual";
-    omission: "omission";
-}>;
+}, z.core.$strict>;
 export declare const comparisonReportSchema: z.ZodObject<{
     avg_score_delta: z.ZodNumber;
     pass_rate_delta: z.ZodNumber;
     exact_match_rate_delta: z.ZodNumber;
     token_f1_delta: z.ZodOptional<z.ZodNumber>;
-    judge_only_avg_score_delta: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
     regressions: z.ZodNumber;
     improvements: z.ZodNumber;
-    regression_taxonomy: z.ZodOptional<z.ZodRecord<z.ZodEnum<{
-        fallback: "fallback";
-        other: "other";
-        style: "style";
-        factual: "factual";
-        omission: "omission";
-    }>, z.ZodNumber>>;
     regressed_examples: z.ZodArray<z.ZodObject<{
         prompt: z.ZodString;
         old_score: z.ZodNumber;
         new_score: z.ZodNumber;
-        category: z.ZodOptional<z.ZodEnum<{
-            fallback: "fallback";
-            other: "other";
-            style: "style";
-            factual: "factual";
-            omission: "omission";
-        }>>;
-    }, z.core.$strip>>;
-}, z.core.$strip>;
+    }, z.core.$strict>>;
+}, z.core.$strict>;
 export declare const trainingReportSchema: z.ZodObject<{
-    provider: z.ZodEnum<{
-        "local-uv": "local-uv";
-        "local-command": "local-command";
-    }>;
+    provider: z.ZodLiteral<"local-uv">;
     training_job_name: z.ZodString;
     model_artifact_uri: z.ZodOptional<z.ZodString>;
     base_model_artifact_uri: z.ZodOptional<z.ZodString>;
-    parent_model_artifact_uri: z.ZodOptional<z.ZodString>;
-    artifact_metadata: z.ZodOptional<z.ZodObject<{
-        framework: z.ZodOptional<z.ZodString>;
-        format: z.ZodOptional<z.ZodString>;
-        entrypoint: z.ZodOptional<z.ZodString>;
-        servable: z.ZodOptional<z.ZodBoolean>;
-        notes: z.ZodOptional<z.ZodString>;
-    }, z.core.$loose>>;
     metrics: z.ZodNullable<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
     exit_code: z.ZodNullable<z.ZodNumber>;
     log_uri: z.ZodString;
     command: z.ZodOptional<z.ZodArray<z.ZodString>>;
-}, z.core.$strip>;
+}, z.core.$strict>;
 export declare const runReportSchema: z.ZodObject<{
     run_id: z.ZodString;
     behavior_spec_id: z.ZodString;
@@ -429,7 +261,7 @@ export declare const runReportSchema: z.ZodObject<{
             spec_holdout: "spec_holdout";
             prebuilt_test: "prebuilt_test";
             prebuilt_validation: "prebuilt_validation";
-            prebuilt_training: "prebuilt_training";
+            general_regression: "general_regression";
         }>>;
         eval_sample_seed: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
         avg_score: z.ZodNumber;
@@ -437,9 +269,6 @@ export declare const runReportSchema: z.ZodObject<{
         exact_match_rate: z.ZodNumber;
         avg_token_f1: z.ZodOptional<z.ZodNumber>;
         avg_latency_ms: z.ZodNumber;
-        judge_scored_count: z.ZodOptional<z.ZodNumber>;
-        fallback_scored_count: z.ZodOptional<z.ZodNumber>;
-        judge_only_avg_score: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
         cached: z.ZodOptional<z.ZodBoolean>;
         cache_key: z.ZodOptional<z.ZodString>;
         results: z.ZodArray<z.ZodObject<{
@@ -451,30 +280,22 @@ export declare const runReportSchema: z.ZodObject<{
             reasoning: z.ZodNullable<z.ZodString>;
             latency_ms: z.ZodNumber;
             scored_by: z.ZodOptional<z.ZodEnum<{
-                llm_judge: "llm_judge";
                 exact_match: "exact_match";
                 json_fields: "json_fields";
                 heuristic: "heuristic";
-                exact_match_fallback: "exact_match_fallback";
             }>>;
-        }, z.core.$strip>>;
+        }, z.core.$strict>>;
         artifact_uri: z.ZodString;
         scoring_method: z.ZodEnum<{
-            command: "command";
-            llm_judge: "llm_judge";
             exact_match: "exact_match";
             json_fields: "json_fields";
             heuristic: "heuristic";
         }>;
-        judge_model_id: z.ZodOptional<z.ZodNullable<z.ZodString>>;
         inference_provider: z.ZodOptional<z.ZodEnum<{
-            command: "command";
             none: "none";
-            batch_command: "batch_command";
             transformers: "transformers";
         }>>;
         scoring_mode: z.ZodOptional<z.ZodEnum<{
-            llm_judge: "llm_judge";
             exact_match: "exact_match";
             json_fields: "json_fields";
         }>>;
@@ -490,11 +311,11 @@ export declare const runReportSchema: z.ZodObject<{
                 correct: z.ZodNumber;
                 total: z.ZodNumber;
                 accuracy: z.ZodNumber;
-            }, z.core.$strip>>;
-        }, z.core.$strip>>;
+            }, z.core.$strict>>;
+        }, z.core.$strict>>;
         generation_config: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
         log_uri: z.ZodOptional<z.ZodString>;
-    }, z.core.$strip>;
+    }, z.core.$strict>;
     candidate: z.ZodObject<{
         kind: z.ZodEnum<{
             candidate: "candidate";
@@ -510,7 +331,7 @@ export declare const runReportSchema: z.ZodObject<{
             spec_holdout: "spec_holdout";
             prebuilt_test: "prebuilt_test";
             prebuilt_validation: "prebuilt_validation";
-            prebuilt_training: "prebuilt_training";
+            general_regression: "general_regression";
         }>>;
         eval_sample_seed: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
         avg_score: z.ZodNumber;
@@ -518,9 +339,6 @@ export declare const runReportSchema: z.ZodObject<{
         exact_match_rate: z.ZodNumber;
         avg_token_f1: z.ZodOptional<z.ZodNumber>;
         avg_latency_ms: z.ZodNumber;
-        judge_scored_count: z.ZodOptional<z.ZodNumber>;
-        fallback_scored_count: z.ZodOptional<z.ZodNumber>;
-        judge_only_avg_score: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
         cached: z.ZodOptional<z.ZodBoolean>;
         cache_key: z.ZodOptional<z.ZodString>;
         results: z.ZodArray<z.ZodObject<{
@@ -532,30 +350,22 @@ export declare const runReportSchema: z.ZodObject<{
             reasoning: z.ZodNullable<z.ZodString>;
             latency_ms: z.ZodNumber;
             scored_by: z.ZodOptional<z.ZodEnum<{
-                llm_judge: "llm_judge";
                 exact_match: "exact_match";
                 json_fields: "json_fields";
                 heuristic: "heuristic";
-                exact_match_fallback: "exact_match_fallback";
             }>>;
-        }, z.core.$strip>>;
+        }, z.core.$strict>>;
         artifact_uri: z.ZodString;
         scoring_method: z.ZodEnum<{
-            command: "command";
-            llm_judge: "llm_judge";
             exact_match: "exact_match";
             json_fields: "json_fields";
             heuristic: "heuristic";
         }>;
-        judge_model_id: z.ZodOptional<z.ZodNullable<z.ZodString>>;
         inference_provider: z.ZodOptional<z.ZodEnum<{
-            command: "command";
             none: "none";
-            batch_command: "batch_command";
             transformers: "transformers";
         }>>;
         scoring_mode: z.ZodOptional<z.ZodEnum<{
-            llm_judge: "llm_judge";
             exact_match: "exact_match";
             json_fields: "json_fields";
         }>>;
@@ -571,80 +381,210 @@ export declare const runReportSchema: z.ZodObject<{
                 correct: z.ZodNumber;
                 total: z.ZodNumber;
                 accuracy: z.ZodNumber;
-            }, z.core.$strip>>;
-        }, z.core.$strip>>;
+            }, z.core.$strict>>;
+        }, z.core.$strict>>;
         generation_config: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
         log_uri: z.ZodOptional<z.ZodString>;
-    }, z.core.$strip>;
+    }, z.core.$strict>;
     comparison: z.ZodObject<{
         avg_score_delta: z.ZodNumber;
         pass_rate_delta: z.ZodNumber;
         exact_match_rate_delta: z.ZodNumber;
         token_f1_delta: z.ZodOptional<z.ZodNumber>;
-        judge_only_avg_score_delta: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
         regressions: z.ZodNumber;
         improvements: z.ZodNumber;
-        regression_taxonomy: z.ZodOptional<z.ZodRecord<z.ZodEnum<{
-            fallback: "fallback";
-            other: "other";
-            style: "style";
-            factual: "factual";
-            omission: "omission";
-        }>, z.ZodNumber>>;
         regressed_examples: z.ZodArray<z.ZodObject<{
             prompt: z.ZodString;
             old_score: z.ZodNumber;
             new_score: z.ZodNumber;
-            category: z.ZodOptional<z.ZodEnum<{
-                fallback: "fallback";
-                other: "other";
-                style: "style";
-                factual: "factual";
-                omission: "omission";
+        }, z.core.$strict>>;
+    }, z.core.$strict>;
+    general_regression: z.ZodOptional<z.ZodObject<{
+        dataset_uri: z.ZodString;
+        dataset_sha256: z.ZodString;
+        baseline: z.ZodObject<{
+            kind: z.ZodEnum<{
+                candidate: "candidate";
+                baseline: "baseline";
+            }>;
+            model_id: z.ZodString;
+            total: z.ZodNumber;
+            eval_examples_total: z.ZodNumber;
+            eval_examples_used: z.ZodNumber;
+            eval_truncated: z.ZodBoolean;
+            eval_split: z.ZodOptional<z.ZodEnum<{
+                spec_examples: "spec_examples";
+                spec_holdout: "spec_holdout";
+                prebuilt_test: "prebuilt_test";
+                prebuilt_validation: "prebuilt_validation";
+                general_regression: "general_regression";
             }>>;
-        }, z.core.$strip>>;
-    }, z.core.$strip>;
+            eval_sample_seed: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
+            avg_score: z.ZodNumber;
+            pass_rate: z.ZodNumber;
+            exact_match_rate: z.ZodNumber;
+            avg_token_f1: z.ZodOptional<z.ZodNumber>;
+            avg_latency_ms: z.ZodNumber;
+            cached: z.ZodOptional<z.ZodBoolean>;
+            cache_key: z.ZodOptional<z.ZodString>;
+            results: z.ZodArray<z.ZodObject<{
+                prompt: z.ZodString;
+                expected: z.ZodString;
+                actual: z.ZodString;
+                passed: z.ZodBoolean;
+                score: z.ZodNumber;
+                reasoning: z.ZodNullable<z.ZodString>;
+                latency_ms: z.ZodNumber;
+                scored_by: z.ZodOptional<z.ZodEnum<{
+                    exact_match: "exact_match";
+                    json_fields: "json_fields";
+                    heuristic: "heuristic";
+                }>>;
+            }, z.core.$strict>>;
+            artifact_uri: z.ZodString;
+            scoring_method: z.ZodEnum<{
+                exact_match: "exact_match";
+                json_fields: "json_fields";
+                heuristic: "heuristic";
+            }>;
+            inference_provider: z.ZodOptional<z.ZodEnum<{
+                none: "none";
+                transformers: "transformers";
+            }>>;
+            scoring_mode: z.ZodOptional<z.ZodEnum<{
+                exact_match: "exact_match";
+                json_fields: "json_fields";
+            }>>;
+            json_field_metrics: z.ZodOptional<z.ZodObject<{
+                fields: z.ZodArray<z.ZodString>;
+                valid_json_count: z.ZodNumber;
+                valid_json_rate: z.ZodNumber;
+                schema_match_count: z.ZodNumber;
+                schema_match_rate: z.ZodNumber;
+                all_fields_match_count: z.ZodNumber;
+                all_fields_match_rate: z.ZodNumber;
+                field_accuracy: z.ZodRecord<z.ZodString, z.ZodObject<{
+                    correct: z.ZodNumber;
+                    total: z.ZodNumber;
+                    accuracy: z.ZodNumber;
+                }, z.core.$strict>>;
+            }, z.core.$strict>>;
+            generation_config: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
+            log_uri: z.ZodOptional<z.ZodString>;
+        }, z.core.$strict>;
+        candidate: z.ZodObject<{
+            kind: z.ZodEnum<{
+                candidate: "candidate";
+                baseline: "baseline";
+            }>;
+            model_id: z.ZodString;
+            total: z.ZodNumber;
+            eval_examples_total: z.ZodNumber;
+            eval_examples_used: z.ZodNumber;
+            eval_truncated: z.ZodBoolean;
+            eval_split: z.ZodOptional<z.ZodEnum<{
+                spec_examples: "spec_examples";
+                spec_holdout: "spec_holdout";
+                prebuilt_test: "prebuilt_test";
+                prebuilt_validation: "prebuilt_validation";
+                general_regression: "general_regression";
+            }>>;
+            eval_sample_seed: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
+            avg_score: z.ZodNumber;
+            pass_rate: z.ZodNumber;
+            exact_match_rate: z.ZodNumber;
+            avg_token_f1: z.ZodOptional<z.ZodNumber>;
+            avg_latency_ms: z.ZodNumber;
+            cached: z.ZodOptional<z.ZodBoolean>;
+            cache_key: z.ZodOptional<z.ZodString>;
+            results: z.ZodArray<z.ZodObject<{
+                prompt: z.ZodString;
+                expected: z.ZodString;
+                actual: z.ZodString;
+                passed: z.ZodBoolean;
+                score: z.ZodNumber;
+                reasoning: z.ZodNullable<z.ZodString>;
+                latency_ms: z.ZodNumber;
+                scored_by: z.ZodOptional<z.ZodEnum<{
+                    exact_match: "exact_match";
+                    json_fields: "json_fields";
+                    heuristic: "heuristic";
+                }>>;
+            }, z.core.$strict>>;
+            artifact_uri: z.ZodString;
+            scoring_method: z.ZodEnum<{
+                exact_match: "exact_match";
+                json_fields: "json_fields";
+                heuristic: "heuristic";
+            }>;
+            inference_provider: z.ZodOptional<z.ZodEnum<{
+                none: "none";
+                transformers: "transformers";
+            }>>;
+            scoring_mode: z.ZodOptional<z.ZodEnum<{
+                exact_match: "exact_match";
+                json_fields: "json_fields";
+            }>>;
+            json_field_metrics: z.ZodOptional<z.ZodObject<{
+                fields: z.ZodArray<z.ZodString>;
+                valid_json_count: z.ZodNumber;
+                valid_json_rate: z.ZodNumber;
+                schema_match_count: z.ZodNumber;
+                schema_match_rate: z.ZodNumber;
+                all_fields_match_count: z.ZodNumber;
+                all_fields_match_rate: z.ZodNumber;
+                field_accuracy: z.ZodRecord<z.ZodString, z.ZodObject<{
+                    correct: z.ZodNumber;
+                    total: z.ZodNumber;
+                    accuracy: z.ZodNumber;
+                }, z.core.$strict>>;
+            }, z.core.$strict>>;
+            generation_config: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
+            log_uri: z.ZodOptional<z.ZodString>;
+        }, z.core.$strict>;
+        comparison: z.ZodObject<{
+            avg_score_delta: z.ZodNumber;
+            pass_rate_delta: z.ZodNumber;
+            exact_match_rate_delta: z.ZodNumber;
+            token_f1_delta: z.ZodOptional<z.ZodNumber>;
+            regressions: z.ZodNumber;
+            improvements: z.ZodNumber;
+            regressed_examples: z.ZodArray<z.ZodObject<{
+                prompt: z.ZodString;
+                old_score: z.ZodNumber;
+                new_score: z.ZodNumber;
+            }, z.core.$strict>>;
+        }, z.core.$strict>;
+        policy: z.ZodObject<{
+            max_score_drop: z.ZodNumber;
+            max_pass_rate_drop: z.ZodNumber;
+        }, z.core.$strict>;
+        passed: z.ZodBoolean;
+        failures: z.ZodArray<z.ZodString>;
+    }, z.core.$strict>>;
     training: z.ZodObject<{
-        provider: z.ZodEnum<{
-            "local-uv": "local-uv";
-            "local-command": "local-command";
-        }>;
+        provider: z.ZodLiteral<"local-uv">;
         training_job_name: z.ZodString;
         model_artifact_uri: z.ZodOptional<z.ZodString>;
         base_model_artifact_uri: z.ZodOptional<z.ZodString>;
-        parent_model_artifact_uri: z.ZodOptional<z.ZodString>;
-        artifact_metadata: z.ZodOptional<z.ZodObject<{
-            framework: z.ZodOptional<z.ZodString>;
-            format: z.ZodOptional<z.ZodString>;
-            entrypoint: z.ZodOptional<z.ZodString>;
-            servable: z.ZodOptional<z.ZodBoolean>;
-            notes: z.ZodOptional<z.ZodString>;
-        }, z.core.$loose>>;
         metrics: z.ZodNullable<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
         exit_code: z.ZodNullable<z.ZodNumber>;
         log_uri: z.ZodString;
         command: z.ZodOptional<z.ZodArray<z.ZodString>>;
-    }, z.core.$strip>;
+    }, z.core.$strict>;
     artifact_uris: z.ZodObject<{
         dataset: z.ZodString;
         baseline_eval: z.ZodString;
         candidate_eval: z.ZodString;
+        general_baseline_eval: z.ZodOptional<z.ZodString>;
+        general_candidate_eval: z.ZodOptional<z.ZodString>;
         report: z.ZodString;
-    }, z.core.$strip>;
+    }, z.core.$strict>;
     run_metadata: z.ZodObject<{
         base_model: z.ZodString;
         fine_tuned_model_id: z.ZodString;
-        parent_model_artifact: z.ZodOptional<z.ZodNullable<z.ZodString>>;
-        training_method: z.ZodDefault<z.ZodEnum<{
-            sft: "sft";
-            dpo: "dpo";
-        }>>;
         dataset_prebuilt: z.ZodBoolean;
-        dataset_format: z.ZodOptional<z.ZodNullable<z.ZodEnum<{
-            chat_jsonl: "chat_jsonl";
-            document_ocr_chat_jsonl: "document_ocr_chat_jsonl";
-            preference_jsonl: "preference_jsonl";
-        }>>>;
+        dataset_format: z.ZodOptional<z.ZodNullable<z.ZodLiteral<"chat_jsonl">>>;
         dataset_uri: z.ZodString;
         spec_example_count: z.ZodNumber;
         training_example_count: z.ZodNullable<z.ZodNumber>;
@@ -655,123 +595,53 @@ export declare const runReportSchema: z.ZodObject<{
             spec_holdout: "spec_holdout";
             prebuilt_test: "prebuilt_test";
             prebuilt_validation: "prebuilt_validation";
-            prebuilt_training: "prebuilt_training";
+            general_regression: "general_regression";
         }>>;
         eval_sample_seed: z.ZodOptional<z.ZodNullable<z.ZodNumber>>;
         started_at: z.ZodString;
         completed_at: z.ZodString;
         elapsed_ms: z.ZodNumber;
         elapsed_seconds: z.ZodNumber;
-    }, z.core.$strip>;
+    }, z.core.$strict>;
     created_at: z.ZodString;
-}, z.core.$strip>;
-export declare const commandSchema: z.ZodArray<z.ZodString>;
-export declare const labelingConfigSchema: z.ZodObject<{
-    model: z.ZodOptional<z.ZodString>;
-    maxTokens: z.ZodDefault<z.ZodNumber>;
-    temperature: z.ZodDefault<z.ZodNumber>;
-    concurrency: z.ZodDefault<z.ZodNumber>;
-    minIntervalMs: z.ZodDefault<z.ZodNumber>;
-    maxAttempts: z.ZodDefault<z.ZodNumber>;
-    maxRows: z.ZodDefault<z.ZodNumber>;
-    timeoutMs: z.ZodDefault<z.ZodNumber>;
-}, z.core.$strip>;
+}, z.core.$strict>;
 export declare const localRunnerConfigSchema: z.ZodObject<{
     storeRoot: z.ZodOptional<z.ZodString>;
     artifactRoot: z.ZodDefault<z.ZodString>;
     dryRun: z.ZodDefault<z.ZodBoolean>;
-    training: z.ZodDefault<z.ZodObject<{
-        backend: z.ZodDefault<z.ZodEnum<{
-            command: "command";
-            uv: "uv";
-        }>>;
-        command: z.ZodOptional<z.ZodArray<z.ZodString>>;
-        artifact: z.ZodOptional<z.ZodObject<{
-            framework: z.ZodOptional<z.ZodString>;
-            format: z.ZodOptional<z.ZodString>;
-            entrypoint: z.ZodOptional<z.ZodString>;
-            servable: z.ZodOptional<z.ZodBoolean>;
-            notes: z.ZodOptional<z.ZodString>;
-        }, z.core.$loose>>;
-        project: z.ZodDefault<z.ZodString>;
-        cwd: z.ZodOptional<z.ZodString>;
-        module: z.ZodOptional<z.ZodString>;
-        script: z.ZodOptional<z.ZodString>;
-        args: z.ZodDefault<z.ZodArray<z.ZodString>>;
-        with: z.ZodDefault<z.ZodArray<z.ZodString>>;
-        env: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodString>>;
-    }, z.core.$strip>>;
     paths: z.ZodDefault<z.ZodObject<{
         baseModel: z.ZodOptional<z.ZodString>;
         modelCache: z.ZodOptional<z.ZodString>;
-    }, z.core.$strip>>;
+    }, z.core.$strict>>;
     evaluation: z.ZodDefault<z.ZodObject<{
-        baselineCommand: z.ZodOptional<z.ZodArray<z.ZodString>>;
-        candidateCommand: z.ZodOptional<z.ZodArray<z.ZodString>>;
         inference: z.ZodDefault<z.ZodObject<{
-            provider: z.ZodDefault<z.ZodEnum<{
-                command: "command";
-                none: "none";
-                batch_command: "batch_command";
-                transformers: "transformers";
-            }>>;
-            command: z.ZodOptional<z.ZodArray<z.ZodString>>;
-            project: z.ZodDefault<z.ZodString>;
-            cwd: z.ZodOptional<z.ZodString>;
-            module: z.ZodOptional<z.ZodString>;
-            script: z.ZodDefault<z.ZodString>;
-            args: z.ZodDefault<z.ZodArray<z.ZodString>>;
-            with: z.ZodDefault<z.ZodArray<z.ZodString>>;
-            env: z.ZodDefault<z.ZodRecord<z.ZodString, z.ZodString>>;
             maxNewTokens: z.ZodDefault<z.ZodNumber>;
             temperature: z.ZodDefault<z.ZodNumber>;
             topP: z.ZodDefault<z.ZodNumber>;
-            trustRemoteCode: z.ZodDefault<z.ZodBoolean>;
             device: z.ZodDefault<z.ZodEnum<{
-                auto: "auto";
                 cpu: "cpu";
                 cuda: "cuda";
-                mps: "mps";
             }>>;
-            chatTemplateKwargs: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
-        }, z.core.$strip>>;
+        }, z.core.$strict>>;
         scoring: z.ZodDefault<z.ZodObject<{
             mode: z.ZodDefault<z.ZodEnum<{
-                llm_judge: "llm_judge";
                 exact_match: "exact_match";
                 json_fields: "json_fields";
             }>>;
-            fallback: z.ZodDefault<z.ZodEnum<{
-                fail: "fail";
-                exact_match: "exact_match";
-            }>>;
             fields: z.ZodOptional<z.ZodArray<z.ZodString>>;
-        }, z.core.$strip>>;
+        }, z.core.$strict>>;
         timeoutMs: z.ZodDefault<z.ZodNumber>;
         maxExamples: z.ZodOptional<z.ZodNumber>;
         sampleSeed: z.ZodOptional<z.ZodNumber>;
-        allowPrebuiltTrainingEval: z.ZodDefault<z.ZodBoolean>;
         baselineCache: z.ZodDefault<z.ZodBoolean>;
-    }, z.core.$strip>>;
-    llm: z.ZodOptional<z.ZodObject<{
-        provider: z.ZodDefault<z.ZodLiteral<"openrouter">>;
-        model: z.ZodDefault<z.ZodString>;
-        apiKeyEnv: z.ZodDefault<z.ZodString>;
-        appName: z.ZodOptional<z.ZodString>;
-        siteUrl: z.ZodOptional<z.ZodString>;
-    }, z.core.$strip>>;
-    labeling: z.ZodDefault<z.ZodObject<{
-        model: z.ZodOptional<z.ZodString>;
-        maxTokens: z.ZodDefault<z.ZodNumber>;
-        temperature: z.ZodDefault<z.ZodNumber>;
-        concurrency: z.ZodDefault<z.ZodNumber>;
-        minIntervalMs: z.ZodDefault<z.ZodNumber>;
-        maxAttempts: z.ZodDefault<z.ZodNumber>;
-        maxRows: z.ZodDefault<z.ZodNumber>;
-        timeoutMs: z.ZodDefault<z.ZodNumber>;
-    }, z.core.$strip>>;
-}, z.core.$strip>;
-export type DocumentInputAsset = z.infer<typeof documentInputAssetSchema>;
+        generalRegression: z.ZodOptional<z.ZodObject<{
+            dataset: z.ZodString;
+            systemPrompt: z.ZodOptional<z.ZodString>;
+            maxScoreDrop: z.ZodDefault<z.ZodNumber>;
+            maxPassRateDrop: z.ZodDefault<z.ZodNumber>;
+        }, z.core.$strict>>;
+    }, z.core.$strict>>;
+}, z.core.$strict>;
 export type BehaviorSpecExample = z.infer<typeof behaviorSpecExampleSchema>;
 export type SpecSnapshot = z.infer<typeof specSnapshotSchema>;
 export type FineTuneHyperparameters = z.infer<typeof fineTuneHyperparametersSchema>;
@@ -783,5 +653,4 @@ export type EvalReport = z.infer<typeof evalReportSchema>;
 export type ComparisonReport = z.infer<typeof comparisonReportSchema>;
 export type TrainingReport = z.infer<typeof trainingReportSchema>;
 export type RunReport = z.infer<typeof runReportSchema>;
-export type LocalLabelingConfig = z.infer<typeof labelingConfigSchema>;
 export type LocalRunnerConfig = z.infer<typeof localRunnerConfigSchema>;