ГЛОБАЛЬНЫЙ РЕФАКТОРИНГ АРХИТЕКТУРЫ - Рефакторинг этапов Stage 3.2 В оркестрации добавлено семантический арбитраж fallbackов (учет guard_hints и extraction) и followup-override для aggregate/unsupported кейсов. \ Убрано залипание в address_lane при carryover-контексте для кейса типа какие обороты ... за 2020, но сохранил защиту поддержанных интентов (list_open_contracts и т.п.), кроме strict deep-диагностики.

2026-04-11 17:04:50 +03:00 · 2026-04-11 17:04:50 +03:00 · c2fa360eb9
parent a7ccc92a9c
commit c2fa360eb9
5 changed files with 332 additions and 16 deletions
--- a/docs/TECH/1CLLMARCH-FACT.md
+++ b/docs/TECH/1CLLMARCH-FACT.md
@ -2366,6 +2366,22 @@ Implemented in current pass (Stage 3.1 kick-off):
   - `assistantSemanticExtractionContract.test.ts` (new)
   - Existing router/chat regressions revalidated (`assistantLivingRouter.test.ts`, `assistantLivingChatMode.test.ts`).

+Implemented in current pass (Stage 3.2 semantic route arbitration):
+1. Extended orchestration fallback arbitration with semantic hints:
+   - `resolveAssistantOrchestrationDecision(...)` now evaluates semantic extraction hints (`guard_hints`, `extraction`) in addition to lexical flags.
+   - Added followup-aware override path so unsupported aggregate carryover queries can still be redirected to deep-analysis lane.
+2. Reduced false route stickiness in followup turns:
+   - Aggregate/unsupported questions with stale followup context no longer remain in address lane by default.
+   - Supported intents (for example `list_open_contracts`) remain protected from accidental deep fallback unless strict deep-investigation cue is detected.
+3. Extended orchestration debug contract:
+   - Added `semantic_route_arbitration` block (`supported_address_intent_detected`, `semantic_deep_investigation_hint_detected`, `semantic_aggregate_shape_detected`, `followup_semantic_override_to_deep_allowed`).
+4. Added/updated regression coverage for real failures:
+   - `assistantLivingRouter.test.ts` now includes followup-carryover aggregate scenario (`какие обороты ... за 2020 год`) and validates deep fallback.
+   - Existing list-open-contracts guard scenario remains green (`Покажи незакрытые договоры ...` stays in address lane).
+5. Validation snapshot:
+   - Stage 3 focused suite: `10` files / `76` tests passed.
+   - Type build: `npm --prefix llm_normalizer/backend run build` passed.
+
 Acceptance (Stage 3):
 1. LLM outputs strictly validated schema for extraction/decomposition (no free-form).
 2. Deterministic guards can block or downgrade answers when evidence insufficient.
--- a/llm_normalizer/backend/dist/services/assistantService.js
+++ b/llm_normalizer/backend/dist/services/assistantService.js
@ -3429,13 +3429,35 @@ function resolveAssistantOrchestrationDecision(input) {
        hasAddressFollowupContextSignal(effectiveAddressUserMessage) ||
        hasAddressFollowupContextSignal(repairedRawUserMessage) ||
        hasAddressFollowupContextSignal(repairedEffectiveAddressUserMessage));
-    const unsupportedIntentOrMode = modeDetection.mode !== "address_query" &&
-        (intentResolution.intent === "unknown" || llmContractMode === "unsupported");
-    const unsupportedAddressIntentFallbackToDeep = Boolean(!followupContext &&
-        baseToolGate?.runAddressLane &&
+    const supportedAddressIntentDetected = !strictDeepInvestigationCueDetected &&
+        Boolean((intentResolution.intent && ADDRESS_INTENTS_KEEP_ADDRESS_LANE.has(intentResolution.intent)) ||
+            (llmContractIntent && ADDRESS_INTENTS_KEEP_ADDRESS_LANE.has(llmContractIntent)));
+    const semanticGuardHints = semanticExtractionContract?.guard_hints &&
+        typeof semanticExtractionContract.guard_hints === "object"
+        ? semanticExtractionContract.guard_hints
+        : null;
+    const semanticExtraction = semanticExtractionContract?.extraction &&
+        typeof semanticExtractionContract.extraction === "object"
+        ? semanticExtractionContract.extraction
+        : null;
+    const semanticDeepInvestigationHintDetected = semanticGuardHints?.deep_investigation_signal_detected === true;
+    const semanticAggregateShapeDetected = semanticExtraction?.query_shape === "AGGREGATE_LOOKUP" ||
+        semanticExtraction?.aggregation_profile === "management_profile";
+    const followupSemanticOverrideToDeepAllowed = Boolean(followupContext &&
+        !supportedAddressIntentDetected &&
+        (llmContractMode === "unsupported" ||
+            semanticAggregateShapeDetected ||
+            semanticDeepInvestigationHintDetected ||
+            !semanticApplyCanonicalRecommended));
+    const unsupportedIntentOrMode = (modeDetection.mode !== "address_query" && intentResolution.intent === "unknown") ||
+        llmContractMode === "unsupported";
+    const unsupportedAddressIntentFallbackToDeep = Boolean(baseToolGate?.runAddressLane &&
        unsupportedIntentOrMode &&
        strongDataSignal &&
-        !preserveAddressLaneSignal);
+        !preserveAddressLaneSignal &&
+        !keepAddressLaneByIntent &&
+        !supportedAddressIntentDetected &&
+        (!followupContext || followupSemanticOverrideToDeepAllowed));
    const deepAnalysisPreferenceDetected = Boolean(hasDeepAnalysisPreferenceSignal(rawUserMessage) ||
        hasDeepAnalysisPreferenceSignal(repairedRawUserMessage) ||
        hasDeepAnalysisPreferenceSignal(effectiveAddressUserMessage) ||
@ -3448,14 +3470,19 @@ function resolveAssistantOrchestrationDecision(input) {
        toNonEmptyString(followupContext.previous_intent) === "vat_payable_forecast" &&
        /(?:\u043f\u043e\u0447\u0435\u043c\u0443|why).*(?:\u043f\u0440\u043e\u0433\u043d\u043e\u0437|forecast).*(?:\u0443\u043f\u043b\u0430\u0442|payable|\b0\b)/iu.test(compactWhitespace(`${repairedRawUserMessage} ${repairedEffectiveAddressUserMessage}`)));
    const deepAnalysisSignalFallbackToDeep = Boolean(baseToolGate?.runAddressLane &&
-        deepAnalysisPreferenceDetected &&
+        (deepAnalysisPreferenceDetected || semanticDeepInvestigationHintDetected) &&
        !keepAddressLaneByIntent &&
+        !supportedAddressIntentDetected &&
        !vatExplainFollowupSignal &&
-        (!followupContext || !dataRetrievalSignal));
+        (!followupContext || !dataRetrievalSignal || followupSemanticOverrideToDeepAllowed));
    const aggregateAnalyticsFallbackToDeep = Boolean(baseToolGate?.runAddressLane &&
        aggregateBusinessAnalyticsSignal &&
        !keepAddressLaneByIntent &&
-        !followupContext);
+        !supportedAddressIntentDetected &&
+        (!followupContext ||
+            llmContractMode === "unsupported" ||
+            semanticAggregateShapeDetected ||
+            !semanticApplyCanonicalRecommended));
    const deepSessionContinuationFallbackToDeep = Boolean(!followupContext &&
        baseToolGate?.runAddressLane &&
        hasDeepSessionContinuationSignal({
@ -3541,6 +3568,12 @@ function resolveAssistantOrchestrationDecision(input) {
            semantic_contract_valid: semanticContractValid,
            semantic_apply_canonical_recommended: semanticApplyCanonicalRecommended,
            semantic_reason_codes: semanticReasonCodes,
+            semantic_route_arbitration: {
+                supported_address_intent_detected: supportedAddressIntentDetected,
+                semantic_deep_investigation_hint_detected: semanticDeepInvestigationHintDetected,
+                semantic_aggregate_shape_detected: semanticAggregateShapeDetected,
+                followup_semantic_override_to_deep_allowed: followupSemanticOverrideToDeepAllowed
+            },
            followup_context_detected: Boolean(followupContext),
            unsupported_address_intent_fallback_to_deep: unsupportedAddressIntentFallbackToDeep,
            deep_analysis_signal_fallback_to_deep: deepAnalysisSignalFallbackToDeep,
--- a/llm_normalizer/backend/src/services/assistantService.ts
+++ b/llm_normalizer/backend/src/services/assistantService.ts
@ -3385,13 +3385,35 @@ export function resolveAssistantOrchestrationDecision(input) {
        hasAddressFollowupContextSignal(effectiveAddressUserMessage) ||
        hasAddressFollowupContextSignal(repairedRawUserMessage) ||
        hasAddressFollowupContextSignal(repairedEffectiveAddressUserMessage));
-    const unsupportedIntentOrMode = modeDetection.mode !== "address_query" &&
-        (intentResolution.intent === "unknown" || llmContractMode === "unsupported");
-    const unsupportedAddressIntentFallbackToDeep = Boolean(!followupContext &&
-        baseToolGate?.runAddressLane &&
+    const supportedAddressIntentDetected = !strictDeepInvestigationCueDetected &&
+        Boolean((intentResolution.intent && ADDRESS_INTENTS_KEEP_ADDRESS_LANE.has(intentResolution.intent)) ||
+            (llmContractIntent && ADDRESS_INTENTS_KEEP_ADDRESS_LANE.has(llmContractIntent)));
+    const semanticGuardHints = semanticExtractionContract?.guard_hints &&
+        typeof semanticExtractionContract.guard_hints === "object"
+        ? semanticExtractionContract.guard_hints
+        : null;
+    const semanticExtraction = semanticExtractionContract?.extraction &&
+        typeof semanticExtractionContract.extraction === "object"
+        ? semanticExtractionContract.extraction
+        : null;
+    const semanticDeepInvestigationHintDetected = semanticGuardHints?.deep_investigation_signal_detected === true;
+    const semanticAggregateShapeDetected = semanticExtraction?.query_shape === "AGGREGATE_LOOKUP" ||
+        semanticExtraction?.aggregation_profile === "management_profile";
+    const followupSemanticOverrideToDeepAllowed = Boolean(followupContext &&
+        !supportedAddressIntentDetected &&
+        (llmContractMode === "unsupported" ||
+            semanticAggregateShapeDetected ||
+            semanticDeepInvestigationHintDetected ||
+            !semanticApplyCanonicalRecommended));
+    const unsupportedIntentOrMode = (modeDetection.mode !== "address_query" && intentResolution.intent === "unknown") ||
+        llmContractMode === "unsupported";
+    const unsupportedAddressIntentFallbackToDeep = Boolean(baseToolGate?.runAddressLane &&
        unsupportedIntentOrMode &&
        strongDataSignal &&
-        !preserveAddressLaneSignal);
+        !preserveAddressLaneSignal &&
+        !keepAddressLaneByIntent &&
+        !supportedAddressIntentDetected &&
+        (!followupContext || followupSemanticOverrideToDeepAllowed));
    const deepAnalysisPreferenceDetected = Boolean(hasDeepAnalysisPreferenceSignal(rawUserMessage) ||
        hasDeepAnalysisPreferenceSignal(repairedRawUserMessage) ||
        hasDeepAnalysisPreferenceSignal(effectiveAddressUserMessage) ||
@ -3404,14 +3426,19 @@ export function resolveAssistantOrchestrationDecision(input) {
        toNonEmptyString(followupContext.previous_intent) === "vat_payable_forecast" &&
        /(?:\u043f\u043e\u0447\u0435\u043c\u0443|why).*(?:\u043f\u0440\u043e\u0433\u043d\u043e\u0437|forecast).*(?:\u0443\u043f\u043b\u0430\u0442|payable|\b0\b)/iu.test(compactWhitespace(`${repairedRawUserMessage} ${repairedEffectiveAddressUserMessage}`)));
    const deepAnalysisSignalFallbackToDeep = Boolean(baseToolGate?.runAddressLane &&
-        deepAnalysisPreferenceDetected &&
+        (deepAnalysisPreferenceDetected || semanticDeepInvestigationHintDetected) &&
        !keepAddressLaneByIntent &&
+        !supportedAddressIntentDetected &&
        !vatExplainFollowupSignal &&
-        (!followupContext || !dataRetrievalSignal));
+        (!followupContext || !dataRetrievalSignal || followupSemanticOverrideToDeepAllowed));
    const aggregateAnalyticsFallbackToDeep = Boolean(baseToolGate?.runAddressLane &&
        aggregateBusinessAnalyticsSignal &&
        !keepAddressLaneByIntent &&
-        !followupContext);
+        !supportedAddressIntentDetected &&
+        (!followupContext ||
+            llmContractMode === "unsupported" ||
+            semanticAggregateShapeDetected ||
+            !semanticApplyCanonicalRecommended));
    const deepSessionContinuationFallbackToDeep = Boolean(!followupContext &&
        baseToolGate?.runAddressLane &&
        hasDeepSessionContinuationSignal({
@ -3497,6 +3524,12 @@ export function resolveAssistantOrchestrationDecision(input) {
            semantic_contract_valid: semanticContractValid,
            semantic_apply_canonical_recommended: semanticApplyCanonicalRecommended,
            semantic_reason_codes: semanticReasonCodes,
+            semantic_route_arbitration: {
+                supported_address_intent_detected: supportedAddressIntentDetected,
+                semantic_deep_investigation_hint_detected: semanticDeepInvestigationHintDetected,
+                semantic_aggregate_shape_detected: semanticAggregateShapeDetected,
+                followup_semantic_override_to_deep_allowed: followupSemanticOverrideToDeepAllowed
+            },
            followup_context_detected: Boolean(followupContext),
            unsupported_address_intent_fallback_to_deep: unsupportedAddressIntentFallbackToDeep,
            deep_analysis_signal_fallback_to_deep: deepAnalysisSignalFallbackToDeep,
--- a/llm_normalizer/backend/tests/assistantLivingRouter.test.ts
+++ b/llm_normalizer/backend/tests/assistantLivingRouter.test.ts
@ -228,6 +228,50 @@ describe("assistant orchestration contract", () => {
    ]).toContain(String(decision.livingReason));
  });

+  it("routes unsupported turnover query to deep even with followup context carryover", () => {
+    const decision = resolveAssistantOrchestrationDecision({
+      rawUserMessage: "\u043a\u0430\u043a\u0438\u0435 \u043e\u0431\u043e\u0440\u043e\u0442\u044b \u043f\u043e \u0430\u043b\u044c\u0442\u0435\u0440\u043d\u0430\u0442\u0438\u0432\u0435 \u0437\u0430 2020 \u0433\u043e\u0434",
+      effectiveAddressUserMessage: "\u041e\u0431\u043e\u0440\u043e\u0442\u044b \u043f\u043e \u0441\u0447\u0435\u0442\u0443 '\u0410\u043b\u044c\u0442\u0435\u0440\u043d\u0430\u0442\u0438\u0432\u0430' \u0437\u0430 2020 \u0433\u043e\u0434.",
+      followupContext: {
+        previous_intent: "list_documents_by_contract",
+        previous_filters: {
+          organization: "\u041e\u041e\u041e \u0410\u043b\u044c\u0442\u0435\u0440\u043d\u0430\u0442\u0438\u0432\u0430 \u041f\u043b\u044e\u0441"
+        }
+      },
+      llmPreDecomposeMeta: {
+        applied: true,
+        llmCanonicalCandidateDetected: true,
+        predecomposeContract: {
+          mode: "unsupported",
+          mode_confidence: "low",
+          intent: "account_balance_snapshot",
+          intent_confidence: "high"
+        },
+        semanticExtractionContract: {
+          extraction: {
+            query_shape: "AGGREGATE_LOOKUP",
+            aggregation_profile: "management_profile"
+          },
+          guard_hints: {
+            deep_investigation_signal_detected: false
+          }
+        }
+      } as any,
+      useMock: false
+    });
+
+    expect(decision.runAddressLane).toBe(false);
+    expect(decision.toolGateDecision).toBe("skip_address_lane");
+    expect(decision.livingMode).toBe("deep_analysis");
+    expect([
+      "address_signal_unsupported_intent_fallback_to_deep",
+      "aggregate_analytics_signal_fallback_to_deep"
+    ]).toContain(String(decision.toolGateReason));
+    expect(
+      decision.orchestrationContract?.semantic_route_arbitration?.followup_semantic_override_to_deep_allowed
+    ).toBe(true);
+  });
+
  it("routes profitability ranking query to deep analysis instead of address lane", () => {
    const decision = resolveAssistantOrchestrationDecision({
      rawUserMessage: "\u043a\u0430\u043a\u043e\u0439 \u0441\u0430\u043c\u044b\u0439 \u0434\u043e\u0445\u043e\u0434\u043d\u044b\u0439 \u0433\u043e\u0434?",
--- a/llm_normalizer/data/eval_cases/assistant_autogen_runtime_job-O0P9xlQ61X.json
+++ b/llm_normalizer/data/eval_cases/assistant_autogen_runtime_job-O0P9xlQ61X.json
@ -0,0 +1,190 @@
+{
+  "suite_id": "assistant_autogen_runtime_job-O0P9xlQ61X",
+  "suite_version": "0.1.0",
+  "schema_version": "assistant_autogen_runtime_v0_1",
+  "scenario_count": 15,
+  "case_ids": [
+    "AUTO-001",
+    "AUTO-002",
+    "AUTO-003",
+    "AUTO-004",
+    "AUTO-005",
+    "AUTO-006",
+    "AUTO-007",
+    "AUTO-008",
+    "AUTO-009",
+    "AUTO-010",
+    "AUTO-011",
+    "AUTO-012",
+    "AUTO-013",
+    "AUTO-014",
+    "AUTO-015"
+  ],
+  "cases": [
+    {
+      "case_id": "AUTO-001",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Кому из контрагентов мы уже месяц отдаем товары, но на счетах все еще красуется минусовое сальдо - это реально зеленый свет для ручного вмешательства?"
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-002",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Где у нас накопились авансы к отгрузкам, которые уже давно пора закрыть или хотя бы перепроверить, чтобы не подозревать худшее?"
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-003",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Покажи контрагентов, по которым сальдо у нас выглядит так, будто оно врет - ну точно не совпадает с тем, что они нам прислали. Это уже критично для сверки."
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-004",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Сколько заказчиков у нас на этот момент могут считаться долгожителями по своим задолженностям?"
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-005",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "В каких случаях мы видим ситуацию, когда документы есть, а денег - нет и пока не предвидится?"
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-006",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Какие контрагенты висят с закрытыми отгрузками, но с открытыми документами оплаты, что явно выглядит как кейс для ручной проверки?"
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-007",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Покажи контрагентов, у которых есть неоплаченные задолженности по договорам на конец месяца - это уже красный свет для бухгалтера."
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-008",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "По каким заказчикам мы можем выделить непростую картину: сальдо нулевое, а история платежей явно говорит о том, что все не так просто?"
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-009",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Какие контрагенты у нас на этом моменте могут быть причислены к тем, кто вообще не платит уже несколько месяцев?"
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-010",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "В каких случаях мы видим зависшие отгрузки, которые уже давно пора закрыть - это грозит проблемами в отчетности."
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-011",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Покажи контрагентов, по которым на конец месяца сальдо выглядит так, будто документы собраны криво и их нужно перепроверить."
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-012",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Какие у нас зависшие авансы или предоплаты уже давно пора либо закрыть, либо хотя бы проверить - это уже не просто вопрос времени?"
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-013",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "По каким контрагентам мы можем заметить такую картину: оплачено меньше, чем отгружено, и это явно требует вмешательства бухгалтера."
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-014",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Какие незакрытые документы по договорам у нас уже давно пора проверить - это грозит серьезными проблемами?"
+        }
+      ]
+    },
+    {
+      "case_id": "AUTO-015",
+      "scenario_tag": "autogen_runtime",
+      "question_type": "direct",
+      "broadness_level": "medium",
+      "turns": [
+        {
+          "user_message": "Покажи контрагентов, чьи заказы на отгрузку еще не оплачены, но сальдо уже отрицательное - это явный признак того, что нужно вмешаться."
+        }
+      ]
+    }
+  ]
+}