ГЛОБАЛЬНЫЙ РЕФАКТОРИНГ АРХИТЕКТУРЫ - Stage 4.7: внедрить stage4_contract_compliance_rate в P0 quality-gap и baseline gate

llm_normalizer/backend/src/eval/p0_acceptance_gate.ts

@@ -161,6 +161,13 @@ function buildQualityGapChecks(input: {
       threshold: thresholds.followup_context_retention_score_min,
       comparator: ">=",
       passed: metrics.followup_context_retention_score >= thresholds.followup_context_retention_score_min
+    },
+    {
+      metric: "stage4_contract_compliance_rate",
+      value: metrics.stage4_contract_compliance_rate,
+      threshold: thresholds.stage4_contract_compliance_rate_min,
+      comparator: ">=",
+      passed: metrics.stage4_contract_compliance_rate >= thresholds.stage4_contract_compliance_rate_min
     }
   ];
 }

llm_normalizer/backend/src/eval/p0_eval_runner.ts

@@ -15,6 +15,7 @@ import {
 } from "../config";
 import { AssistantService } from "../services/assistantService";
 import { AssistantSessionStore } from "../services/assistantSessionStore";
+import { buildStage4AnswerContractAuditV1 } from "../services/assistantStage4AnswerContractAudit";
 import { NormalizerService } from "../services/normalizerService";
 import type { AssistantMessageResponsePayload } from "../types/assistant";
 import type { EvalRunMode, NormalizeRequestPayload } from "../types/normalizer";
@@ -1291,6 +1292,10 @@ function computeQualityGapMetrics(caseRecords: P0CaseRecord[]): {
   const mechanismSpecificityScores = caseRecords.map((item) => item.metric_subscores.mechanism_specificity_score);
   const followupCases = caseRecords.filter((item) => item.query_class === "followup_investigation");
   const followupRetainedCases = followupCases.filter((item) => item.checks.followup_context_retained).length;
+  const stage4ContractAuditedCases = caseRecords
+    .map((item) => buildStage4AnswerContractAuditV1(item.assistant_reply ?? ""))
+    .filter((item) => item.schema_version === "stage4_answer_contract_audit_v1");
+  const stage4ContractCompliantCases = stage4ContractAuditedCases.filter((item) => item.is_stage4_shape === true).length;
 
   return {
     raw: {
@@ -1299,11 +1304,15 @@ function computeQualityGapMetrics(caseRecords: P0CaseRecord[]): {
       mechanism_specificity_score: average(mechanismSpecificityScores),
       followup_context_retention_score: round2(
         followupCases.length === 0 ? 1 : followupRetainedCases / Math.max(1, followupCases.length)
+      ),
+      stage4_contract_compliance_rate: round2(
+        stage4ContractAuditedCases.length === 0 ? 0 : stage4ContractCompliantCases / Math.max(1, stage4ContractAuditedCases.length)
       )
     },
     denominators: {
       cases_total: caseRecords.length,
-      followup_cases_total: followupCases.length
+      followup_cases_total: followupCases.length,
+      stage4_contract_audited_cases_total: stage4ContractAuditedCases.length
     }
   };
 }

llm_normalizer/backend/src/eval/p0_metric_definitions.ts

@@ -68,6 +68,7 @@ export interface P0QualityGapMetricVector {
   false_confidence_rate: number;
   mechanism_specificity_score: number;
   followup_context_retention_score: number;
+  stage4_contract_compliance_rate: number;
 }
 
 export type P0QualityGapMetricName = keyof P0QualityGapMetricVector;
@@ -147,6 +148,11 @@ export const P0_QUALITY_GAP_METRIC_DEFINITIONS: Record<P0QualityGapMetricName, P
     description: "Rate of correct context retention in follow-up investigation cases.",
     unit: "rate",
     direction: "higher_is_better"
+  },
+  stage4_contract_compliance_rate: {
+    description: "Share of answers that satisfy the Stage 4 user-facing answer contract shape.",
+    unit: "rate",
+    direction: "higher_is_better"
   }
 };
 
@@ -183,6 +189,7 @@ export interface P0QualityGapThresholds {
   false_confidence_rate_max: number;
   mechanism_specificity_score_min: number;
   followup_context_retention_score_min: number;
+  stage4_contract_compliance_rate_min: number;
 }
 
 export const P0_DEFAULT_ACCEPTANCE_THRESHOLDS: P0AcceptanceThresholds = {
@@ -200,7 +207,8 @@ export const P0_DEFAULT_QUALITY_GAP_THRESHOLDS: P0QualityGapThresholds = {
   generic_explanation_rate_max: 0.2,
   false_confidence_rate_max: 0.12,
   mechanism_specificity_score_min: 3.0,
-  followup_context_retention_score_min: 0.75
+  followup_context_retention_score_min: 0.75,
+  stage4_contract_compliance_rate_min: 0.9
 };
 
 function isProblemUnitType(value: unknown): value is ProblemUnitType {