ГЛОБАЛЬНЫЙ РЕФАКТОРИНГ АРХИТЕКТУРЫ - Рефакторинг этапов Stage 3.6 Усилина оркестрация и детект data-scope для живого сленга. Убрана шаблонность soft-refusal в limited-ответах. Подрезана словарнаю жирность в фильтр-экстракторе

llm_normalizer/backend/src/routes/eval.ts

@@ -202,6 +202,11 @@ export const __evalRouteTestUtils = {
   normalizeRuntimeQuestions
 };
 
+export const __evalRouteAsyncTestUtils = {
+  readReportedCaseIds,
+  syncJobWithSessions
+};
+
 function normalizeCaseIds(value: unknown): string[] | undefined {
   if (!Array.isArray(value)) {
     return undefined;
@@ -365,10 +370,30 @@ function readSessionConversation(runId: string, caseId: string): EvalAsyncCaseIn
   }
 }
 
+function readReportedCaseIds(job: EvalAsyncJob): Set<string> {
+  const output = new Set<string>();
+  const reportRecord = toRecord(job.report);
+  const reportResults = toArray(reportRecord?.results)
+    .map((item) => toRecord(item))
+    .filter((item): item is Record<string, unknown> => item !== null);
+
+  for (const result of reportResults) {
+    const caseId = toStringSafe(result.case_id);
+    if (!caseId) continue;
+    output.add(caseId);
+  }
+  return output;
+}
+
+function isTerminalCaseStatus(status: EvalAsyncStatus): boolean {
+  return status === "completed" || status === "failed";
+}
+
 function syncJobWithSessions(job: EvalAsyncJob): void {
   if (!job.run_id || !job.eval_target.startsWith("assistant_")) {
     return;
   }
+  const reportCaseIds = readReportedCaseIds(job);
   let completed = 0;
   let hasRunning = false;
   for (const item of job.cases) {
@@ -376,11 +401,16 @@ function syncJobWithSessions(job: EvalAsyncJob): void {
     item.messages = messages;
     const assistantMessages = messages.filter((entry) => entry.role === "assistant").length;
     const userMessages = messages.filter((entry) => entry.role === "user").length;
-    if (assistantMessages >= item.turns_total && item.turns_total > 0) {
+    const reportMarkedDone = reportCaseIds.has(item.case_id);
+    if ((assistantMessages >= item.turns_total && item.turns_total > 0) || reportMarkedDone) {
       item.status = "completed";
       completed += 1;
       continue;
     }
+    if (isTerminalCaseStatus(item.status) && isTerminalCaseStatus(job.status)) {
+      completed += 1;
+      continue;
+    }
     if (userMessages > 0 || messages.length > 0) {
       item.status = "running";
       hasRunning = true;