ГЛОБАЛЬНЫЙ РЕФАКТОРИНГ АРХИТЕКТУРЫ - Рефакторинг этапов 2.2 - Усилена оркестрацию в deep/address гейте и follow-up binding в assistantService.ts. Починен кейс UTF-8 follow-up refinement (теперь followup_state_usage.applied=true в нужном сценарии). Убраны регрессии по assistantLivingRouter и stage3 lifecycle probe. Корректное поведение для llm canonical candidate (чтобы не уезжало в clarification_required там, где должен быть address factual).

2026-04-11 14:56:14 +03:00 · 2026-04-11 14:56:14 +03:00 · b5bd4fd737
parent 19f5f19d8e
commit b5bd4fd737
46 changed files with 2471 additions and 370 deletions
--- a/docs/TECH/1CLLMARCH-FACT.md
+++ b/docs/TECH/1CLLMARCH-FACT.md
@ -1,6 +1,6 @@
 # 1CLLMARCH Fact Check And Stabilization Plan

-Updated at: 2026-04-10  
+Updated at: 2026-04-11  
 Source baseline: `docs/TECH/1CLLMARCH.md`

 ## 1. Purpose
@ -2096,7 +2096,236 @@ Validation:
   - `assistantDeepTurnAnalysisRuntimeAdapter.test.ts`
   - `assistantDeepTurnAnalysisAttemptRuntimeAdapter.test.ts`

-Status: **In progress (Phase 2.1 + 2.2 + 2.3 + 2.4 + 2.5 + 2.6 + 2.7 + 2.8 + 2.9 + 2.10 + 2.11 + 2.12 + 2.13 + 2.14 + 2.15 + 2.16 + 2.17 + 2.18 + 2.19 + 2.20 + 2.21 + 2.22 + 2.23 + 2.24 + 2.25 + 2.26 + 2.27 + 2.28 + 2.29 + 2.30 + 2.31 + 2.32 + 2.33 + 2.34 + 2.35 + 2.36 + 2.37 + 2.38 + 2.39 + 2.40 + 2.41 + 2.42 + 2.43 + 2.44 + 2.45 + 2.46 + 2.47 + 2.48 + 2.49 + 2.50 + 2.51 + 2.52 + 2.53 + 2.54 + 2.55 + 2.56 + 2.57 + 2.58 + 2.59 + 2.60 + 2.61 + 2.62 + 2.63 + 2.64 + 2.65 + 2.66 + 2.67 + 2.68 + 2.69 + 2.70 + 2.71 + 2.72 + 2.73 + 2.74 + 2.75 + 2.76 + 2.77 + 2.78 + 2.79 + 2.80 + 2.81 + 2.82 + 2.83 + 2.84 + 2.85 + 2.86 + 2.87 + 2.88 + 2.89 + 2.90 + 2.91 + 2.92 + 2.93 + 2.94 + 2.95 + 2.96 + 2.97 + 2.98 + 2.99 + 2.100 + 2.101 + 2.102 + 2.103 + 2.104 + 2.105 + 2.106 + 2.107 + 2.108 + 2.109 + 2.110 + 2.111 + 2.112 completed)**
+Implemented in current pass (Phase 2.113 + 2.114 + 2.115 + 2.116):
+1. Tightened temporal/polarity guard typing in deep plan runtime:
+   - `assistantDeepTurnPlanRuntimeAdapter.ts`
+   - `temporalGuard` now typed to `TemporalGuardAudit`;
+   - `domainPolarityGuardInitial` now typed to `DomainPolarityGuardAudit`.
+2. Preserved behavior:
+   - no runtime logic changes; type alignment only.
+
+Validation:
+1. `npm run build` passed.
+2. Targeted deep plan pack passed:
+   - `assistantDeepTurnPlanRuntimeAdapter.test.ts`
+   - `assistantDeepTurnAnalysisAttemptRuntimeAdapter.test.ts`
+
+Implemented in current pass (Phase 2.117 + 2.118 + 2.119 + 2.120):
+1. Tightened audit typing along deep response/packaging chain:
+   - `assistantDebugPayloadAssembler.ts`
+   - `assistantMessageLogAssembler.ts`
+   - `assistantDeepTurnPackaging.ts`
+   - `assistantDeepTurnPackagingRuntimeAdapter.ts`
+   - `assistantDeepTurnInputBuilder.ts`
+   - `assistantDeepTurnResponseRuntimeAdapter.ts`
+   - audit contracts now use concrete types:
+     - `TemporalGuardAudit`, `DomainPolarityGuardAudit`, `ClaimBoundAnchorAudit`,
+       `TargetedEvidenceAcquisitionAudit`, `EvidenceAdmissibilityAudit`,
+       `GroundedAnswerEligibilityAudit`,
+       `RbpLiveRouteAuditDebug | null`, `FaLiveRouteAuditDebug | null`.
+2. Preserved behavior:
+   - no runtime logic changes; type alignment only.
+
+Validation:
+1. `npm run build` passed.
+2. Targeted deep response/packaging pack passed:
+   - `assistantDeepTurnResponseRuntimeAdapter.test.ts`
+   - `assistantDeepTurnResponseAttemptRuntimeAdapter.test.ts`
+   - `assistantDeepTurnPackagingRuntimeAdapter.test.ts`
+   - `assistantDeepTurnPackaging.test.ts`
+
+Implemented in current pass (Phase 2.121 + 2.122 + 2.123 + 2.124):
+1. Tightened retrieval call/raw types across deep response chain:
+   - `assistantDeepTurnRetrievalRuntimeAdapter.ts`
+   - `assistantDeepTurnResponseRuntimeAdapter.ts`
+   - `assistantDeepTurnPackaging.ts`
+   - `assistantDeepTurnPackagingRuntimeAdapter.ts`
+   - `assistantDeepTurnInputBuilder.ts`
+   - `assistantMessageLogAssembler.ts`
+   - `assistantEvidenceBundleAssembler.ts`
+   - `assistantOrchestrationContracts.ts`
+   - introduced `AssistantRetrievalRawResult` and typed `AssistantRetrievalRawResultRecord`;
+   - retrieval calls now use `AssistantRetrievalCallRecord[]` end-to-end.
+2. Preserved behavior:
+   - no runtime logic changes; only stronger typing and safe raw normalization.
+
+Validation:
+1. `npm run build` passed.
+2. Targeted deep retrieval/response pack passed:
+   - `assistantDeepTurnResponseRuntimeAdapter.test.ts`
+   - `assistantDeepTurnRetrievalRuntimeAdapter.test.ts`
+   - `assistantDeepTurnPackaging.test.ts`
+
+Implemented in current pass (Phase 2.125 + 2.126 + 2.127 + 2.128):
+1. Tightened deep turn execution state + fallback metadata typing:
+   - `assistantDeepTurnPrePackagingContext.ts`
+   - `assistantDeepTurnPackaging.ts`
+   - `assistantDeepTurnPackagingRuntimeAdapter.ts`
+   - `assistantDeepTurnInputBuilder.ts`
+   - `assistantDeepTurnResponseRuntimeAdapter.ts`
+   - `assistantDeepTurnNormalizationRuntimeAdapter.ts`
+   - added `AssistantExecutionStateRecord` + `AssistantAddressRuntimeMetaForDeep`;
+   - `fallback_type` now `AssistantFallbackType`;
+   - `problem_answer_mode` now `AssistantProblemAnswerMode`;
+   - `problem_unit_ids_used` now `string[]`;
+   - `investigationStateSnapshot` now `InvestigationStateWithProblemUnits | null`.
+2. Aligned debug/log payload typing with contracts:
+   - `assistantDebugPayloadAssembler.ts`
+   - `assistantMessageLogAssembler.ts`
+   - `assistantOrchestrationContractsV1` and `outcomeClassV1` now typed;
+   - `answerStructureV11` now typed.
+3. Preserved behavior:
+   - no runtime logic changes; type alignment only.
+
+Validation:
+1. Not run in this pass (type-only changes).
+
+Implemented in current pass (Phase 2.129 + 2.130 + 2.131 + 2.132):
+1. Typified debug route structures end-to-end:
+   - `assistantQueryPlanning.ts`
+   - `assistantDeepTurnPrePackagingContext.ts`
+   - `assistantDeepTurnPackaging.ts`
+   - `assistantDeepTurnPackagingRuntimeAdapter.ts`
+   - `assistantDeepTurnInputBuilder.ts`
+   - `assistantDeepTurnResponseRuntimeAdapter.ts`
+   - `assistantDebugPayloadAssembler.ts`
+   - `assistantMessageLogAssembler.ts`
+   - introduced `AssistantDebugRouteRecord` union (legacy vs deterministic debug routes);
+   - `routes` now typed in `AssistantDebugPayload`.
+2. Preserved behavior:
+   - no runtime logic changes; type alignment only.
+
+Validation:
+1. Not run in this pass (type-only changes).
+
+Implemented in current pass (Phase 2.133):
+1. Tightened deep retrieval runtime input:
+   - `assistantDeepTurnRetrievalRuntimeAdapter.ts`
+   - `executeRouteRuntime` now returns `AssistantRetrievalRawResult` (explicit union).
+2. Preserved behavior:
+   - no runtime logic changes; type alignment only.
+
+Validation:
+1. Not run in this pass (type-only changes).
+
+Implemented in current pass (Phase 2.134 + 2.135):
+1. Typed live-route plan audit contract:
+   - `assistantDeepTurnPlanRuntimeAdapter.ts`
+   - introduced `AssistantLiveRoutePlanAudit` for plan enforcement audits.
+2. Typed deep analysis log payload plumbing:
+   - `assistantMessageLogAssembler.ts`
+   - `assistantDeepTurnPackaging.ts`
+   - `assistantDeepTurnPackagingRuntimeAdapter.ts`
+   - `assistantDeepTurnFinalizeRuntimeAdapter.ts`
+   - introduced `DeepAnalysisLogDetails` alias and used it end-to-end.
+3. Preserved behavior:
+   - no runtime logic changes; type alignment only.
+
+Validation:
+1. Not run in this pass (type-only changes).
+
+Implemented in current pass (Phase 2.136):
+1. Tightened business-scope resolution contract shape:
+   - `assistantDeepTurnContextRuntimeAdapter.ts`
+   - removed index-signature `unknown` on business scope resolution;
+   - `resolveBusinessScopeFromLiveContext` now uses `AssistantBusinessScopeResolution` explicitly.
+2. Preserved behavior:
+   - no runtime logic changes; type alignment only.
+
+Validation:
+1. Not run in this pass (type-only changes).
+
+Implemented in current pass (Phase 2.137 + 2.138):
+1. Tightened company anchor normalization input types:
+   - `assistantDeepTurnCompositionRuntimeAdapter.ts`
+   - `toStringArray` now accepts `string[] | null | undefined`;
+   - company anchor normalization now consumes `Partial<CompanyAnchorSet>`.
+2. Narrowed response runtime normalizers:
+   - `assistantDeepTurnResponseRuntimeAdapter.ts`
+   - runtime analysis context and business scope normalizers now take typed inputs.
+3. Preserved behavior:
+   - no runtime logic changes; type alignment only.
+
+Validation:
+1. Not run in this pass (type-only changes).
+
+Implemented in current pass (Phase 2.139 + 2.140):
+1. Tightened analysis attempt builder normalizers:
+   - `assistantDeepTurnAnalysisAttemptInputBuilder.ts`
+   - removed `unknown` from anchor/period helpers; now `Partial<CompanyAnchorSet>` + typed primary period.
+2. Narrowed response runtime normalization helpers:
+   - `assistantDeepTurnResponseRuntimeAdapter.ts`
+   - execution plan normalization now consumes typed `AssistantExecutionPlanItem[]`;
+   - helper signatures no longer accept `unknown` where input is already typed.
+3. Preserved behavior:
+   - no runtime logic changes; type alignment only.
+
+Validation:
+1. Not run in this pass (type-only changes).
+
+Implemented in current pass (Phase 2.141 + 2.142 + 2.143 + 2.144):
+1. Tightened attempt input defaults:
+   - `assistantDeepTurnAttemptInputBuilder.ts`
+   - response attempt default type is now `AssistantMessageResponsePayload`.
+2. Tightened retrieval raw result typing:
+   - `assistantDeepTurnRetrievalRuntimeAdapter.ts`
+   - introduced `AssistantRetrievalRawResultLike` + list item union.
+3. Tightened deep packaging normalization:
+   - `assistantDeepTurnPackaging.ts`
+   - normalized fragments extracted without `Record<string, unknown>` cast.
+4. Simplified response runtime normalizers:
+   - `assistantDeepTurnResponseRuntimeAdapter.ts`
+   - removed `toRecordObject` casts and normalized from typed inputs.
+5. Preserved behavior:
+   - no runtime logic changes; type alignment only.
+
+Validation:
+1. Not run in this pass (type-only changes).
+
+Implemented in current pass (Phase 2.145 + 2.146):
+1. Tightened retrieval raw result field shapes:
+   - `assistantDeepTurnRetrievalRuntimeAdapter.ts`
+   - introduced `AssistantRetrievalFieldValue` / `AssistantRetrievalRecord`;
+   - removed `unknown` from raw result record types.
+2. Tightened normalized fragment extraction:
+   - `assistantDeepTurnPackaging.ts`
+   - normalized fragments now typed to `NormalizedQueryV2*` fragments.
+3. Preserved behavior:
+   - no runtime logic changes; type alignment only.
+
+Validation:
+1. Not run in this pass (type-only changes).
+
+Status: **Completed (Phase 2.1–2.146)**
+
+### Stage 2 Completion Report (Summary)
+1. Orchestration monolith decomposed into explicit modules:
+   - QueryFrame, ExecutionPlan, EvidenceBundle, Coverage/Grounding, Answer package, Debug payload, Log details.
+2. Deep lane now uses stable contracts end-to-end:
+   - `assistant_orchestration_contracts_v1`, `assistant_evidence_bundle_v1`, `assistant_coverage_contract_v1`.
+3. Audit & trace coverage standardized:
+   - temporal/polarity/claim/evidence guards, live-route audits, followup usage, outcome class.
+4. Type hardening complete across deep chain:
+   - normalized payloads, execution plan, retrieval calls/raw, debug routes, runtime meta, investigation state.
+5. Behavior preserved throughout refactor (no route/answer regressions by design).
+
+### Stage 2 Closure Audit (2026-04-11)
+1. Fixed runtime-critical context loss in `assistantTurnRuntimeDepsAdapter.ts`:
+   - unbound session store/logger/normalizer methods caused `TypeError` at `assistantSessionStore.ensureSession(...)` and mass `500` responses in API tests.
+2. Added safe method wrappers in deps adapter:
+   - `ensureSession`, `appendItem`, `getSession`, `persistSession`, `setInvestigationState`, `normalize`.
+3. Added regression guard:
+   - `assistantTurnRuntimeDepsAdapter.test.ts` now includes a stateful instance-context test to prevent `this` loss regressions.
+4. Validation gates (fact):
+   - `npm run build` passed.
+   - Combined Stage 2 regression validation passed: `37` files / `95` tests (deep-turn adapters/builders/packaging, orchestration contracts/runtime, MCP bridge, followup continuity, wave10 corrective regression).
+5. Scope note:
+   - Full backend suite still has red tests in Stage 3/4 probes and long-running acceptance packs; this is tracked under Stage 3 backlog and is not a Stage 2 blocker.
+
+### Stage 2 Remaining Risks (Known)
+1. Final answer quality still template-heavy and brittle.
+2. Lexical routing pressure remains high (dictionary overfitting risk).
+3. Deterministic guards still compensate for weak semantic parsing.

 ## Stage 3 (P2): Hybrid Semantic Layer (LLM + Deterministic Guards)

@ -2105,6 +2334,27 @@ Goal:
 2. Keep deterministic guardrails as verifier, not primary “brain”.
 3. Reduce dictionary overfitting and false route drifts.

+Plan (Stage 3):
+1. **Schema-first semantic extraction**
+   - Strict JSON schema for: entities, time scope, intent, ambiguity, success criteria.
+   - Hard validation + retry/repair loop.
+2. **LLM decomposition with guardrails**
+   - Decomposition produces executable plan candidates.
+   - Deterministic guards validate: domain polarity, temporal window, claim-bound anchors.
+3. **Evidence-first reasoning**
+   - LLM only summarizes from evidence bundle, never invents facts.
+4. **Context binding**
+   - Carryover only via typed followup state, not free-text memory.
+5. **Quality gates**
+   - Coverage critic threshold before final answer.
+   - Reason-code taxonomy normalized.
+
+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.
+3. False route drifts and generic responses reduced in regression packs.
+4. Manual markup shows increase in “correct/grounded” labels.
+
 Status: Planned

 ## Stage 4 (P2): Human-Centric Answer Layer
--- a/llm_normalizer/backend/dist/services/assistantDeepTurnAnalysisAttemptInputBuilder.js
+++ b/llm_normalizer/backend/dist/services/assistantDeepTurnAnalysisAttemptInputBuilder.js
@ -16,12 +16,6 @@ const KNOWN_GUARD_DOMAINS = [
    "month_close_costs_20_44",
    "fixed_asset_amortization"
 ];
-function toRecordObject(value) {
-    if (!value || typeof value !== "object" || Array.isArray(value)) {
-        return null;
-    }
-    return value;
-}
 function toStringArray(value) {
    if (!Array.isArray(value)) {
        return [];
@ -31,10 +25,10 @@ function toStringArray(value) {
        .filter((item) => item.length > 0);
 }
 function toCompanyAnchorSet(value) {
-    const source = toRecordObject(value);
-    if (!source) {
+    if (!value || typeof value !== "object" || Array.isArray(value)) {
        return null;
    }
+    const source = value;
    return {
        contract_numbers: toStringArray(source.contract_numbers),
        document_numbers: toStringArray(source.document_numbers),
@ -47,13 +41,12 @@ function toCompanyAnchorSet(value) {
    };
 }
 function toClaimBoundPrimaryPeriod(value) {
-    const source = toRecordObject(value);
-    if (!source) {
+    if (!value || typeof value !== "object") {
        return null;
    }
-    const from = typeof source.from === "string" ? source.from.trim() : "";
-    const to = typeof source.to === "string" ? source.to.trim() : "";
-    const granularity = source.granularity === "day" || source.granularity === "month" ? source.granularity : null;
+    const from = typeof value.from === "string" ? value.from.trim() : "";
+    const to = typeof value.to === "string" ? value.to.trim() : "";
+    const granularity = value.granularity === "day" || value.granularity === "month" ? value.granularity : null;
    if (!from || !to || !granularity) {
        return null;
    }
--- a/llm_normalizer/backend/dist/services/assistantDeepTurnNormalizationRuntimeAdapter.js
+++ b/llm_normalizer/backend/dist/services/assistantDeepTurnNormalizationRuntimeAdapter.js
@ -2,13 +2,16 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.buildAssistantDeepTurnNormalizationRuntime = buildAssistantDeepTurnNormalizationRuntime;
 async function buildAssistantDeepTurnNormalizationRuntime(input) {
-    const followupBinding = input.featureInvestigationStateV1 &&
+    const investigationState = input.sessionInvestigationState;
+    const canUseFollowupBinding = input.featureInvestigationStateV1 &&
        input.featureStateFollowupBindingV1 &&
-        Boolean(input.sessionInvestigationState)
+        investigationState !== null &&
+        investigationState !== undefined;
+    const followupBinding = canUseFollowupBinding
        ? input.buildFollowupStateBinding({
            userMessage: input.userMessage,
            payloadContext: input.payload.context,
-            investigationState: input.sessionInvestigationState
+            investigationState
        })
        : {
            normalizedQuestion: input.userMessage,
--- a/llm_normalizer/backend/dist/services/assistantDeepTurnPackaging.js
+++ b/llm_normalizer/backend/dist/services/assistantDeepTurnPackaging.js
@ -6,9 +6,16 @@ const assistantContractsBundleAssembler_1 = require("./assistantContractsBundleA
 const assistantDeepResponseAssembler_1 = require("./assistantDeepResponseAssembler");
 const assistantDebugPayloadAssembler_1 = require("./assistantDebugPayloadAssembler");
 const assistantMessageLogAssembler_1 = require("./assistantMessageLogAssembler");
+function extractNormalizedFragments(normalized) {
+    if (!normalized || typeof normalized !== "object") {
+        return [];
+    }
+    const source = normalized;
+    return Array.isArray(source.fragments) ? source.fragments : [];
+}
 function assembleAssistantDeepTurnPackaging(input) {
-    const normalizedPayload = (input.normalized.normalized ?? null);
-    const normalizedFragments = Array.isArray(normalizedPayload?.["fragments"]) ? normalizedPayload?.["fragments"] : [];
+    const normalizedPayload = input.normalized.normalized ?? null;
+    const normalizedFragments = extractNormalizedFragments(normalizedPayload);
    const evidenceBundleAssembly = (0, assistantEvidenceBundleAssembler_1.assembleAssistantEvidenceBundle)({
        retrievalCalls: input.retrievalCalls,
        retrievalResults: input.retrievalResults
--- a/llm_normalizer/backend/dist/services/assistantDeepTurnResponseRuntimeAdapter.js
+++ b/llm_normalizer/backend/dist/services/assistantDeepTurnResponseRuntimeAdapter.js
@ -3,12 +3,6 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.runAssistantDeepTurnResponseRuntime = runAssistantDeepTurnResponseRuntime;
 const assistantDeepTurnPackagingRuntimeAdapter_1 = require("./assistantDeepTurnPackagingRuntimeAdapter");
 const assistantDeepTurnFinalizeRuntimeAdapter_1 = require("./assistantDeepTurnFinalizeRuntimeAdapter");
-function toRecordObject(value) {
-    if (!value || typeof value !== "object") {
-        return null;
-    }
-    return value;
-}
 function toNullableString(value) {
    if (typeof value !== "string") {
        return null;
@ -31,31 +25,17 @@ function normalizeExecutionPlan(value) {
    if (!Array.isArray(value)) {
        return [];
    }
-    return value.map((item, index) => {
-        const source = toRecordObject(item);
-        return {
-            fragment_id: toNullableString(source?.fragment_id) ?? `fragment_${index + 1}`,
-            requirement_ids: toStringArray(source?.requirement_ids),
-            route: toNullableString(source?.route) ?? "unknown_route",
-            should_execute: Boolean(source?.should_execute),
-            no_route_reason: toNullableString(source?.no_route_reason),
-            clarification_reason: toNullableString(source?.clarification_reason)
-        };
-    });
-}
-function normalizeRecordArray(value) {
-    if (!Array.isArray(value)) {
-        return [];
-    }
-    return value
-        .map((item) => toRecordObject(item))
-        .filter((item) => Boolean(item));
-}
-function normalizeRecord(value) {
-    return toRecordObject(value) ?? {};
+    return value.map((item, index) => ({
+        fragment_id: toNullableString(item.fragment_id) ?? `fragment_${index + 1}`,
+        requirement_ids: toStringArray(item.requirement_ids),
+        route: toNullableString(item.route) ?? "unknown_route",
+        should_execute: Boolean(item.should_execute),
+        no_route_reason: toNullableString(item.no_route_reason ?? null),
+        clarification_reason: toNullableString(item.clarification_reason ?? null)
+    }));
 }
 function normalizeRuntimeAnalysisContext(value) {
-    const source = toRecordObject(value);
+    const source = value;
    return {
        active: Boolean(source?.active),
        as_of_date: toNullableString(source?.as_of_date),
@ -66,7 +46,7 @@ function normalizeRuntimeAnalysisContext(value) {
    };
 }
 function normalizeBusinessScopeResolution(value) {
-    const source = toRecordObject(value);
+    const source = value;
    return {
        business_scope_raw: toStringArray(source?.business_scope_raw),
        business_scope_resolved: toStringArray(source?.business_scope_resolved),
@ -75,7 +55,7 @@ function normalizeBusinessScopeResolution(value) {
    };
 }
 function normalizeAddressRuntimeMetaForDeep(value) {
-    const source = toRecordObject(value);
+    const source = value;
    if (!source) {
        return null;
    }
@ -87,8 +67,8 @@ function normalizeAddressRuntimeMetaForDeep(value) {
        fallbackRuleHit: toNullableString(source.fallbackRuleHit),
        toolGateDecision: toNullableString(source.toolGateDecision),
        toolGateReason: toNullableString(source.toolGateReason),
-        predecomposeContract: toRecordObject(source.predecomposeContract),
-        orchestrationContract: toRecordObject(source.orchestrationContract)
+        predecomposeContract: source.predecomposeContract ?? null,
+        orchestrationContract: source.orchestrationContract ?? null
    };
 }
 function runAssistantDeepTurnResponseRuntime(input) {
@ -107,21 +87,21 @@ function runAssistantDeepTurnResponseRuntime(input) {
        coverageEvaluationRequirements: input.coverageEvaluationRequirements,
        coverageReport: input.coverageReport,
        groundingCheck: input.groundingCheck,
-        retrievalCalls: normalizeRecordArray(input.retrievalCalls),
-        retrievalResultsRaw: Array.isArray(input.retrievalResultsRaw) ? input.retrievalResultsRaw : [],
+        retrievalCalls: input.retrievalCalls,
+        retrievalResultsRaw: input.retrievalResultsRaw,
        retrievalResults: input.retrievalResults,
        questionTypeClass: input.questionTypeClass,
        companyAnchors: input.companyAnchors,
        runtimeAnalysisContext: normalizeRuntimeAnalysisContext(input.runtimeAnalysisContext),
        businessScopeResolution: normalizeBusinessScopeResolution(input.businessScopeResolution),
-        temporalGuard: normalizeRecord(input.temporalGuard),
-        polarityAudit: normalizeRecord(input.polarityAudit),
-        claimAnchorAudit: normalizeRecord(input.claimAnchorAudit),
+        temporalGuard: input.temporalGuard,
+        polarityAudit: input.polarityAudit,
+        claimAnchorAudit: input.claimAnchorAudit,
        targetedEvidenceAudit: input.targetedEvidenceAudit,
        evidenceAdmissibilityGateAudit: input.evidenceAdmissibilityGateAudit,
        rbpLiveRouteAudit: input.rbpLiveRouteAudit ?? null,
        faLiveRouteAudit: input.faLiveRouteAudit ?? null,
-        groundedAnswerEligibilityGuard: normalizeRecord(input.groundedAnswerEligibilityGuard),
+        groundedAnswerEligibilityGuard: input.groundedAnswerEligibilityGuard,
        followupStateUsage: input.followupStateUsage,
        followupApplied: input.followupApplied,
        composition: input.composition,
--- a/llm_normalizer/backend/dist/services/assistantDeepTurnRetrievalRuntimeAdapter.js
+++ b/llm_normalizer/backend/dist/services/assistantDeepTurnRetrievalRuntimeAdapter.js
@ -20,6 +20,21 @@ function buildRouteExecutorErrorRawResult(route, message) {
        errors: [message]
    };
 }
+function normalizeRawResult(value) {
+    if (value === null || value === undefined) {
+        return null;
+    }
+    if (typeof value === "string" || typeof value === "number" || typeof value === "boolean") {
+        return value;
+    }
+    if (Array.isArray(value)) {
+        return value;
+    }
+    if (typeof value === "object") {
+        return value;
+    }
+    return null;
+}
 async function executeAssistantDeepTurnRetrievalPlan(input) {
    const normalizeRetrievalResultSafe = input.normalizeRetrievalResultFn ?? retrievalResultNormalizer_1.normalizeRetrievalResult;
    const retrievalCalls = [];
@ -50,12 +65,13 @@ async function executeAssistantDeepTurnRetrievalPlan(input) {
            const raw = await input.executeRouteRuntime(planItem.route, planItem.fragment_text, {
                temporalHint: input.liveTemporalHint
            });
+            const normalizedRaw = normalizeRawResult(raw);
            retrievalResultsRaw.push({
                fragment_id: planItem.fragment_id,
                route: planItem.route,
-                raw_result: raw
+                raw_result: normalizedRaw
            });
-            retrievalResults.push(normalizeRetrievalResultSafe(planItem.fragment_id, planItem.requirement_ids, planItem.route, raw));
+            retrievalResults.push(normalizeRetrievalResultSafe(planItem.fragment_id, planItem.requirement_ids, planItem.route, normalizedRaw));
        }
        catch (error) {
            const message = error instanceof Error ? error.message : String(error);
--- a/llm_normalizer/backend/dist/services/assistantTurnRuntimeDepsAdapter.js
+++ b/llm_normalizer/backend/dist/services/assistantTurnRuntimeDepsAdapter.js
@ -4,12 +4,12 @@ exports.buildAssistantTurnRuntimeDeps = buildAssistantTurnRuntimeDeps;
 function buildAssistantTurnRuntimeDeps(input) {
    return {
        ...input.helpers,
-        ensureSession: input.sessions.ensureSession,
-        appendItem: input.sessions.appendItem,
-        getSession: input.sessions.getSession,
-        persistSession: input.sessionLogger.persistSession,
-        setInvestigationState: input.sessions.setInvestigationState,
-        normalize: input.normalizerService.normalize,
+        ensureSession: (sessionId) => input.sessions.ensureSession(sessionId),
+        appendItem: (sessionId, item) => input.sessions.appendItem(sessionId, item),
+        getSession: (sessionId) => input.sessions.getSession(sessionId),
+        persistSession: (session) => input.sessionLogger.persistSession(session),
+        setInvestigationState: (sessionId, state) => input.sessions.setInvestigationState(sessionId, state),
+        normalize: (payload) => input.normalizerService.normalize(payload),
        executeRouteRuntime: (route, fragmentText, options) => input.dataLayer.executeRouteRuntime(route, fragmentText, options),
        tryAddressQueryHandle: (messageUsed, options) => input.addressQueryService.tryHandle(messageUsed, options),
        chatClient: input.chatClient,
--- a/llm_normalizer/backend/src/services/assistantAddressOrchestrationRuntimeAdapter.ts
+++ b/llm_normalizer/backend/src/services/assistantAddressOrchestrationRuntimeAdapter.ts
@ -22,6 +22,7 @@ export interface BuildAssistantAddressOrchestrationRuntimeInput {
    effectiveAddressUserMessage: string;
    followupContext: unknown;
    llmPreDecomposeMeta: Record<string, unknown>;
+    sessionItems?: unknown[];
    useMock: boolean;
  }) => Record<string, unknown>;
  buildAddressDialogContinuationContractV2: (
@ -102,6 +103,7 @@ export async function buildAssistantAddressOrchestrationRuntime(
    effectiveAddressUserMessage: addressInputMessage,
    followupContext,
    llmPreDecomposeMeta: addressPreDecompose,
+    sessionItems: input.sessionItems,
    useMock: input.useMock
  });
  const dialogContinuationContract = input.buildAddressDialogContinuationContractV2(
--- a/llm_normalizer/backend/src/services/assistantDebugPayloadAssembler.ts
+++ b/llm_normalizer/backend/src/services/assistantDebugPayloadAssembler.ts
@ -1,4 +1,28 @@
-import type { AssistantDebugPayload } from "../types/assistant";
+import type {
+  AssistantAddressRuntimeMetaForDeep,
+  AssistantDebugPayload,
+  AssistantDebugRouteRecord,
+  AssistantFallbackType,
+  AssistantProblemAnswerMode,
+  AssistantRequirement,
+  UnifiedRetrievalResult,
+  FaLiveRouteAuditDebug,
+  RbpLiveRouteAuditDebug
+} from "../types/assistant";
+import type { NormalizeResponsePayload, RouteHintSummary } from "../types/normalizer";
+import type { AnswerStructureV11 } from "../types/stage1Contracts";
+import type { InvestigationStateWithProblemUnits } from "../types/stage2ProblemUnits";
+import type { ClaimBoundAnchorAudit, TargetedEvidenceAcquisitionAudit } from "./assistantClaimBoundEvidence";
+import type { AssistantContractsBundleV1 } from "./assistantContractsBundleAssembler";
+import type { AssistantOutcomeClassV1 } from "./assistantOrchestrationContracts";
+import type { AssistantFollowupUsage } from "./assistantFollowupUsage";
+import type { CompanyAnchorSet } from "./companyAnchorResolver";
+import type {
+  DomainPolarityGuardAudit,
+  EvidenceAdmissibilityAudit,
+  GroundedAnswerEligibilityAudit,
+  TemporalGuardAudit
+} from "./assistantRuntimeGuards";

 type RetrievalStatusItem = AssistantDebugPayload["retrieval_status"][number];

@ -6,18 +30,18 @@ export interface DeepAnalysisDebugPayloadInput {
  traceId: string;
  promptVersion: string;
  schemaVersion: string;
-  fallbackType: unknown;
-  routeSummary: unknown;
+  fallbackType: AssistantFallbackType;
+  routeSummary: RouteHintSummary | null;
  fragments: unknown[];
-  requirementsExtracted: unknown[];
-  coverageReport: unknown;
-  routes: Array<Record<string, unknown>>;
+  requirementsExtracted: AssistantRequirement[];
+  coverageReport: AssistantDebugPayload["coverage_report"];
+  routes: AssistantDebugRouteRecord[];
  retrievalStatus: RetrievalStatusItem[];
-  retrievalResults: unknown[];
-  groundingCheck: unknown;
+  retrievalResults: UnifiedRetrievalResult[];
+  groundingCheck: AssistantDebugPayload["answer_grounding_check"];
  droppedIntentSegments: string[];
  questionTypeClass: string;
-  companyAnchors: unknown;
+  companyAnchors: CompanyAnchorSet | null;
  runtimeAnalysisContext: {
    active: boolean;
    as_of_date: string | null;
@ -32,40 +56,27 @@ export interface DeepAnalysisDebugPayloadInput {
    company_grounding_applied?: boolean;
    scope_resolution_reason?: string[];
  };
-  temporalGuard: Record<string, unknown>;
-  polarityAudit: Record<string, unknown>;
-  claimAnchorAudit: Record<string, unknown>;
-  targetedEvidenceAudit: unknown;
-  evidenceAdmissibilityGateAudit: unknown;
-  rbpLiveRouteAudit: unknown | null;
-  faLiveRouteAudit: unknown | null;
-  groundedAnswerEligibilityGuard: Record<string, unknown>;
-  followupStateUsage: unknown | null;
+  temporalGuard: TemporalGuardAudit;
+  polarityAudit: DomainPolarityGuardAudit;
+  claimAnchorAudit: ClaimBoundAnchorAudit;
+  targetedEvidenceAudit: TargetedEvidenceAcquisitionAudit;
+  evidenceAdmissibilityGateAudit: EvidenceAdmissibilityAudit;
+  rbpLiveRouteAudit: RbpLiveRouteAuditDebug | null;
+  faLiveRouteAudit: FaLiveRouteAuditDebug | null;
+  groundedAnswerEligibilityGuard: GroundedAnswerEligibilityAudit;
+  followupStateUsage: AssistantFollowupUsage | null;
  compositionDebug: {
    problem_centric_answer_applied?: boolean;
    problem_units_used_count?: number;
-    problem_answer_mode?: string;
+    problem_answer_mode?: AssistantProblemAnswerMode;
    problem_unit_ids_used?: string[];
  };
-  addressRuntimeMetaForDeep:
-    | {
-        attempted?: boolean;
-        applied?: boolean;
-        reason?: string | null;
-        provider?: string | null;
-        fallbackRuleHit?: string | null;
-        toolGateDecision?: string | null;
-        toolGateReason?: string | null;
-        predecomposeContract?: unknown;
-        orchestrationContract?: unknown;
-      }
-    | null
-    | undefined;
-  outcomeClassV1: unknown;
-  assistantOrchestrationContractsV1: unknown;
-  answerStructureV11: unknown;
-  investigationStateSnapshot: unknown;
-  normalizedPayload: unknown;
+  addressRuntimeMetaForDeep: AssistantAddressRuntimeMetaForDeep | null | undefined;
+  outcomeClassV1: AssistantOutcomeClassV1;
+  assistantOrchestrationContractsV1: AssistantContractsBundleV1["assistantOrchestrationContractsV1"];
+  answerStructureV11: AnswerStructureV11 | null;
+  investigationStateSnapshot: InvestigationStateWithProblemUnits | null;
+  normalizedPayload: NormalizeResponsePayload["normalized"];
 }

 function toAnalysisContext(input: DeepAnalysisDebugPayloadInput["runtimeAnalysisContext"]): Record<string, unknown> | null {
--- a/llm_normalizer/backend/src/services/assistantDeepTurnAnalysisAttemptInputBuilder.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnAnalysisAttemptInputBuilder.ts
@ -49,14 +49,7 @@ const KNOWN_GUARD_DOMAINS = [
  "fixed_asset_amortization"
 ] as const;

-function toRecordObject(value: unknown): Record<string, unknown> | null {
-  if (!value || typeof value !== "object" || Array.isArray(value)) {
-    return null;
-  }
-  return value as Record<string, unknown>;
-}
-
-function toStringArray(value: unknown): string[] {
+function toStringArray(value: string[] | null | undefined): string[] {
  if (!Array.isArray(value)) {
    return [];
  }
@ -65,11 +58,11 @@ function toStringArray(value: unknown): string[] {
    .filter((item) => item.length > 0);
 }

-function toCompanyAnchorSet(value: unknown): CompanyAnchorSet | null {
-  const source = toRecordObject(value);
-  if (!source) {
+function toCompanyAnchorSet(value: Partial<CompanyAnchorSet> | null | undefined): CompanyAnchorSet | null {
+  if (!value || typeof value !== "object" || Array.isArray(value)) {
    return null;
  }
+  const source: Partial<CompanyAnchorSet> = value;
  return {
    contract_numbers: toStringArray(source.contract_numbers),
    document_numbers: toStringArray(source.document_numbers),
@ -82,14 +75,13 @@ function toCompanyAnchorSet(value: unknown): CompanyAnchorSet | null {
  };
 }

-function toClaimBoundPrimaryPeriod(value: unknown): ClaimBoundPrimaryPeriod {
-  const source = toRecordObject(value);
-  if (!source) {
+function toClaimBoundPrimaryPeriod(value: ClaimBoundPrimaryPeriod | null | undefined): ClaimBoundPrimaryPeriod {
+  if (!value || typeof value !== "object") {
    return null;
  }
-  const from = typeof source.from === "string" ? source.from.trim() : "";
-  const to = typeof source.to === "string" ? source.to.trim() : "";
-  const granularity = source.granularity === "day" || source.granularity === "month" ? source.granularity : null;
+  const from = typeof value.from === "string" ? value.from.trim() : "";
+  const to = typeof value.to === "string" ? value.to.trim() : "";
+  const granularity = value.granularity === "day" || value.granularity === "month" ? value.granularity : null;
  if (!from || !to || !granularity) {
    return null;
  }
@ -101,7 +93,7 @@ function toClaimBoundPrimaryPeriod(value: unknown): ClaimBoundPrimaryPeriod {
 }

 function normalizeFocusDomainForGuards(
-  value: unknown
+  value: string | null | undefined
 ): AssistantDeepTurnRetrievalGuardPipelineInput["focusDomainForGuards"] {
  const normalized = typeof value === "string" ? value.trim() : "";
  if (!normalized) {
--- a/llm_normalizer/backend/src/services/assistantDeepTurnAttemptInputBuilder.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnAttemptInputBuilder.ts
@ -5,6 +5,7 @@ import type {
 import type { RunAssistantDeepTurnAnalysisAttemptRuntimeInput } from "./assistantDeepTurnAnalysisAttemptRuntimeAdapter";
 import type { RunAssistantDeepTurnAnalysisRuntimeOutput } from "./assistantDeepTurnAnalysisRuntimeAdapter";
 import type { RunAssistantDeepTurnResponseAttemptRuntimeInput } from "./assistantDeepTurnResponseAttemptRuntimeAdapter";
+import type { AssistantMessageResponsePayload } from "../types/assistant";
 import { isAssistantFollowupApplied } from "./assistantFollowupUsage";

 export interface BuildAssistantDeepTurnNormalizationRuntimeInputInput {
@ -88,7 +89,7 @@ export function buildAssistantDeepTurnAnalysisAttemptRuntimeInput(
  };
 }

-export interface BuildAssistantDeepTurnResponseAttemptRuntimeInputInput<ResponseType = unknown> {
+export interface BuildAssistantDeepTurnResponseAttemptRuntimeInputInput<ResponseType = AssistantMessageResponsePayload> {
  featureInvestigationStateV1: boolean;
  featureContractsV11: boolean;
  featureAnswerPolicyV11: boolean;
@ -115,7 +116,7 @@ export interface BuildAssistantDeepTurnResponseAttemptRuntimeInputInput<Response
  deepTurnAnalysisRuntime: RunAssistantDeepTurnAnalysisRuntimeOutput;
 }

-export function buildAssistantDeepTurnResponseAttemptRuntimeInput<ResponseType = unknown>(
+export function buildAssistantDeepTurnResponseAttemptRuntimeInput<ResponseType = AssistantMessageResponsePayload>(
  input: BuildAssistantDeepTurnResponseAttemptRuntimeInputInput<ResponseType>
 ): RunAssistantDeepTurnResponseAttemptRuntimeInput<ResponseType> {
  return {
--- a/llm_normalizer/backend/src/services/assistantDeepTurnCompositionRuntimeAdapter.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnCompositionRuntimeAdapter.ts
@ -16,7 +16,7 @@ export interface BuildAssistantDeepTurnCompositionInput {
  groundingCheck: AnswerGroundingCheck;
  followupUsage: AssistantFollowupUsage | null | undefined;
  investigationState: InvestigationStateWithProblemUnits | null | undefined;
-  companyAnchors: unknown;
+  companyAnchors: CompanyAnchorSet | null;
  normalizedPayload: NormalizeResponsePayload["normalized"];
  featureAnswerPolicyV11: boolean;
  featureProblemCentricAnswerV1: boolean;
@ -34,7 +34,7 @@ export interface AssistantDeepTurnCompositionOutput {
  composition: ReturnType<typeof composeAssistantAnswer>;
 }

-function toStringArray(value: unknown): string[] {
+function toStringArray(value: string[] | null | undefined): string[] {
  if (!Array.isArray(value)) {
    return [];
  }
@ -43,11 +43,11 @@ function toStringArray(value: unknown): string[] {
    .filter((item) => item.length > 0);
 }

-function normalizeCompanyAnchorSet(value: unknown): CompanyAnchorSet | null {
+function normalizeCompanyAnchorSet(value: Partial<CompanyAnchorSet> | null | undefined): CompanyAnchorSet | null {
  if (!value || typeof value !== "object" || Array.isArray(value)) {
    return null;
  }
-  const source = value as Record<string, unknown>;
+  const source: Partial<CompanyAnchorSet> = value;
  return {
    contract_numbers: toStringArray(source.contract_numbers),
    document_numbers: toStringArray(source.document_numbers),
--- a/llm_normalizer/backend/src/services/assistantDeepTurnContextRuntimeAdapter.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnContextRuntimeAdapter.ts
@ -61,14 +61,11 @@ export interface BuildAssistantDeepTurnRuntimeContextInput {
    userMessage: string;
    companyAnchors: CompanyAnchorSet;
    focusDomainHint: string | null;
-    primaryPeriod: unknown;
+    primaryPeriod: ClaimBoundAnchorAudit["primary_period"] | null;
  }) => ClaimBoundAnchorAudit;
  resolveBusinessScopeFromLiveContext: (input: {
-    current: {
-      route_summary_resolved: RouteHintSummary | null;
-      [key: string]: unknown;
-    };
-    temporalGuard: unknown;
+    current: AssistantBusinessScopeResolution;
+    temporalGuard: TemporalGuardAudit;
    claimType: string;
    focusDomainHint: string | null;
    userMessage: string;
@ -83,7 +80,6 @@ export interface AssistantBusinessScopeResolution {
  business_scope_resolved?: string[];
  company_grounding_applied?: boolean;
  scope_resolution_reason?: string[];
-  [key: string]: unknown;
 }

 export interface BuildAssistantDeepTurnRuntimeContextOutput {
--- a/llm_normalizer/backend/src/services/assistantDeepTurnFinalizeRuntimeAdapter.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnFinalizeRuntimeAdapter.ts
@ -1,4 +1,5 @@
 import type { AssistantConversationItem, AssistantDebugPayload, AssistantMessageResponsePayload, AssistantReplyType } from "../types/assistant";
+import type { DeepAnalysisLogDetails } from "./assistantMessageLogAssembler";
 import { buildAssistantDeepTurnSuccessResponse } from "./assistantDeepTurnResponseBuilder";
 import type { CommitAssistantTurnAndLogOutput } from "./assistantTurnCommitRuntimeAdapter";
 import { commitAssistantTurnAndLog } from "./assistantTurnCommitRuntimeAdapter";
@ -9,7 +10,7 @@ export interface FinalizeAssistantDeepTurnInput {
  replyType: AssistantReplyType;
  assistantItem: AssistantConversationItem;
  debug: AssistantDebugPayload;
-  deepAnalysisLogDetails: Record<string, unknown>;
+  deepAnalysisLogDetails: DeepAnalysisLogDetails;
  appendItem: Parameters<typeof commitAssistantTurnAndLog>[0]["appendItem"];
  getSession: Parameters<typeof commitAssistantTurnAndLog>[0]["getSession"];
  persistSession: Parameters<typeof commitAssistantTurnAndLog>[0]["persistSession"];
--- a/llm_normalizer/backend/src/services/assistantDeepTurnInputBuilder.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnInputBuilder.ts
@ -1,8 +1,34 @@
-import type { AssistantReplyType, AssistantRequirement, AnswerGroundingCheck, RequirementCoverageReport, UnifiedRetrievalResult } from "../types/assistant";
+import type {
+  AssistantAddressRuntimeMetaForDeep,
+  AssistantDebugRouteRecord,
+  AssistantExecutionStateRecord,
+  AssistantFallbackType,
+  AssistantProblemAnswerMode,
+  AssistantReplyType,
+  AssistantRequirement,
+  AnswerGroundingCheck,
+  FaLiveRouteAuditDebug,
+  RbpLiveRouteAuditDebug,
+  RequirementCoverageReport,
+  UnifiedRetrievalResult
+} from "../types/assistant";
 import type { NormalizeResponsePayload, RouteHintSummary } from "../types/normalizer";
 import type { AnswerStructureV11 } from "../types/stage1Contracts";
+import type { InvestigationStateWithProblemUnits } from "../types/stage2ProblemUnits";
 import type { AssistantDeepTurnPackagingInput } from "./assistantDeepTurnPackaging";
 import type { AssistantFollowupUsage } from "./assistantFollowupUsage";
+import type { ClaimBoundAnchorAudit, TargetedEvidenceAcquisitionAudit } from "./assistantClaimBoundEvidence";
+import type { CompanyAnchorSet } from "./companyAnchorResolver";
+import type {
+  DomainPolarityGuardAudit,
+  EvidenceAdmissibilityAudit,
+  GroundedAnswerEligibilityAudit,
+  TemporalGuardAudit
+} from "./assistantRuntimeGuards";
+import type {
+  AssistantRetrievalCallRecord,
+  AssistantRetrievalRawResultRecord
+} from "./assistantDeepTurnRetrievalRuntimeAdapter";

 export interface AssistantDeepTurnInputBuilderArgs {
  sessionId: string;
@ -36,13 +62,13 @@ export interface AssistantDeepTurnInputBuilderArgs {
  coverageEvaluationRequirements: AssistantRequirement[];
  coverageReport: RequirementCoverageReport;
  groundingCheck: AnswerGroundingCheck;
-  retrievalCalls: Array<Record<string, unknown>>;
-  retrievalResultsRaw: unknown[];
+  retrievalCalls: AssistantRetrievalCallRecord[];
+  retrievalResultsRaw: AssistantRetrievalRawResultRecord[];
  retrievalResults: UnifiedRetrievalResult[];
-  routesForDebug: Array<Record<string, unknown>>;
-  resolvedExecutionState: unknown;
+  routesForDebug: AssistantDebugRouteRecord[];
+  resolvedExecutionState: AssistantExecutionStateRecord[];
  questionTypeClass: string;
-  companyAnchors: unknown;
+  companyAnchors: CompanyAnchorSet | null;
  runtimeAnalysisContext: {
    active: boolean;
    as_of_date: string | null;
@ -57,42 +83,29 @@ export interface AssistantDeepTurnInputBuilderArgs {
    company_grounding_applied?: boolean;
    scope_resolution_reason?: string[];
  };
-  temporalGuard: Record<string, unknown>;
-  polarityAudit: Record<string, unknown>;
-  claimAnchorAudit: Record<string, unknown>;
-  targetedEvidenceAudit: unknown;
-  evidenceAdmissibilityGateAudit: unknown;
-  rbpLiveRouteAudit: unknown | null;
-  faLiveRouteAudit: unknown | null;
-  groundedAnswerEligibilityGuard: Record<string, unknown>;
+  temporalGuard: TemporalGuardAudit;
+  polarityAudit: DomainPolarityGuardAudit;
+  claimAnchorAudit: ClaimBoundAnchorAudit;
+  targetedEvidenceAudit: TargetedEvidenceAcquisitionAudit;
+  evidenceAdmissibilityGateAudit: EvidenceAdmissibilityAudit;
+  rbpLiveRouteAudit: RbpLiveRouteAuditDebug | null;
+  faLiveRouteAudit: FaLiveRouteAuditDebug | null;
+  groundedAnswerEligibilityGuard: GroundedAnswerEligibilityAudit;
  followupStateUsage?: AssistantFollowupUsage | null;
  composition: {
    reply_type: AssistantReplyType;
-    fallback_type: unknown;
+    fallback_type: AssistantFallbackType;
    answer_structure_v11?: AnswerStructureV11 | null;
    problem_centric_answer_applied?: boolean;
    problem_units_used_count?: number;
-    problem_answer_mode?: string;
-    problem_unit_ids_used?: unknown;
+    problem_answer_mode?: AssistantProblemAnswerMode;
+    problem_unit_ids_used?: string[];
  };
  safeAssistantReplyBase: string;
  featureContractsV11: boolean;
  featureAnswerPolicyV11: boolean;
-  investigationStateSnapshot: unknown;
-  addressRuntimeMetaForDeep:
-    | {
-        attempted?: boolean;
-        applied?: boolean;
-        reason?: string | null;
-        provider?: string | null;
-        fallbackRuleHit?: string | null;
-        toolGateDecision?: string | null;
-        toolGateReason?: string | null;
-        predecomposeContract?: unknown;
-        orchestrationContract?: unknown;
-      }
-    | null
-    | undefined;
+  investigationStateSnapshot: InvestigationStateWithProblemUnits | null;
+  addressRuntimeMetaForDeep: AssistantAddressRuntimeMetaForDeep | null | undefined;
 }

 export function buildAssistantDeepTurnPackagingInput(args: AssistantDeepTurnInputBuilderArgs): AssistantDeepTurnPackagingInput {
--- a/llm_normalizer/backend/src/services/assistantDeepTurnNormalizationRuntimeAdapter.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnNormalizationRuntimeAdapter.ts
@ -1,5 +1,6 @@
 import type { AssistantMessageRequestPayload } from "../types/assistant";
 import type { NormalizeRequestPayload, NormalizeResponsePayload } from "../types/normalizer";
+import type { InvestigationStateWithProblemUnits } from "../types/stage2ProblemUnits";
 import type { AssistantFollowupUsage } from "./assistantFollowupUsage";

 export interface AssistantDeepTurnFollowupBinding {
@ -13,11 +14,11 @@ export interface BuildAssistantDeepTurnNormalizationRuntimeInput {
  payload: AssistantMessageRequestPayload;
  featureInvestigationStateV1: boolean;
  featureStateFollowupBindingV1: boolean;
-  sessionInvestigationState: unknown | null | undefined;
+  sessionInvestigationState: InvestigationStateWithProblemUnits | null | undefined;
  buildFollowupStateBinding: (input: {
    userMessage: string;
    payloadContext: NormalizeRequestPayload["context"] | undefined;
-    investigationState: unknown;
+    investigationState: InvestigationStateWithProblemUnits;
  }) => AssistantDeepTurnFollowupBinding;
  normalize: (payload: NormalizeRequestPayload) => Promise<NormalizeResponsePayload>;
 }
@ -31,14 +32,18 @@ export interface BuildAssistantDeepTurnNormalizationRuntimeOutput {
 export async function buildAssistantDeepTurnNormalizationRuntime(
  input: BuildAssistantDeepTurnNormalizationRuntimeInput
 ): Promise<BuildAssistantDeepTurnNormalizationRuntimeOutput> {
-  const followupBinding =
+  const investigationState = input.sessionInvestigationState;
+  const canUseFollowupBinding =
    input.featureInvestigationStateV1 &&
    input.featureStateFollowupBindingV1 &&
-    Boolean(input.sessionInvestigationState)
+    investigationState !== null &&
+    investigationState !== undefined;
+  const followupBinding =
+    canUseFollowupBinding
      ? input.buildFollowupStateBinding({
          userMessage: input.userMessage,
          payloadContext: input.payload.context,
-          investigationState: input.sessionInvestigationState as unknown
+          investigationState
        })
      : {
          normalizedQuestion: input.userMessage,
--- a/llm_normalizer/backend/src/services/assistantDeepTurnPackaging.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnPackaging.ts
@ -1,14 +1,28 @@
 import type {
+  AssistantAddressRuntimeMetaForDeep,
  AssistantConversationItem,
  AssistantDebugPayload,
+  AssistantDebugRouteRecord,
+  AssistantExecutionStateRecord,
+  AssistantFallbackType,
+  AssistantProblemAnswerMode,
  AssistantReplyType,
  AssistantRequirement,
  AnswerGroundingCheck,
+  FaLiveRouteAuditDebug,
+  RbpLiveRouteAuditDebug,
  RequirementCoverageReport,
  UnifiedRetrievalResult
 } from "../types/assistant";
-import type { NormalizeResponsePayload, RouteHintSummary } from "../types/normalizer";
+import type {
+  NormalizeResponsePayload,
+  NormalizedQueryV2,
+  NormalizedQueryV2_0_1,
+  NormalizedQueryV2_0_2,
+  RouteHintSummary
+} from "../types/normalizer";
 import type { AnswerStructureV11 } from "../types/stage1Contracts";
+import type { InvestigationStateWithProblemUnits } from "../types/stage2ProblemUnits";
 import {
  assembleAssistantEvidenceBundle,
  type AssistantEvidenceBundleAssembly
@ -16,8 +30,23 @@ import {
 import { assembleAssistantContractsBundleV1, type AssistantContractsBundleV1 } from "./assistantContractsBundleAssembler";
 import { buildDeepAnswerArtifacts, buildAssistantConversationItem, type DeepAnswerArtifacts } from "./assistantDeepResponseAssembler";
 import { buildDeepAnalysisDebugPayload } from "./assistantDebugPayloadAssembler";
-import { buildDeepAnalysisProcessedLogDetails } from "./assistantMessageLogAssembler";
+import {
+  buildDeepAnalysisProcessedLogDetails,
+  type DeepAnalysisLogDetails
+} from "./assistantMessageLogAssembler";
 import type { AssistantFollowupUsage } from "./assistantFollowupUsage";
+import type { ClaimBoundAnchorAudit, TargetedEvidenceAcquisitionAudit } from "./assistantClaimBoundEvidence";
+import type { CompanyAnchorSet } from "./companyAnchorResolver";
+import type {
+  DomainPolarityGuardAudit,
+  EvidenceAdmissibilityAudit,
+  GroundedAnswerEligibilityAudit,
+  TemporalGuardAudit
+} from "./assistantRuntimeGuards";
+import type {
+  AssistantRetrievalCallRecord,
+  AssistantRetrievalRawResultRecord
+} from "./assistantDeepTurnRetrievalRuntimeAdapter";

 export interface AssistantDeepTurnPackagingInput {
  sessionId: string;
@ -51,13 +80,13 @@ export interface AssistantDeepTurnPackagingInput {
  coverageEvaluationRequirements: AssistantRequirement[];
  coverageReport: RequirementCoverageReport;
  groundingCheck: AnswerGroundingCheck;
-  retrievalCalls: Array<Record<string, unknown>>;
-  retrievalResultsRaw: unknown[];
+  retrievalCalls: AssistantRetrievalCallRecord[];
+  retrievalResultsRaw: AssistantRetrievalRawResultRecord[];
  retrievalResults: UnifiedRetrievalResult[];
-  routesForDebug: Array<Record<string, unknown>>;
-  resolvedExecutionState: unknown;
+  routesForDebug: AssistantDebugRouteRecord[];
+  resolvedExecutionState: AssistantExecutionStateRecord[];
  questionTypeClass: string;
-  companyAnchors: unknown;
+  companyAnchors: CompanyAnchorSet | null;
  runtimeAnalysisContext: {
    active: boolean;
    as_of_date: string | null;
@ -72,42 +101,29 @@ export interface AssistantDeepTurnPackagingInput {
    company_grounding_applied?: boolean;
    scope_resolution_reason?: string[];
  };
-  temporalGuard: Record<string, unknown>;
-  polarityAudit: Record<string, unknown>;
-  claimAnchorAudit: Record<string, unknown>;
-  targetedEvidenceAudit: unknown;
-  evidenceAdmissibilityGateAudit: unknown;
-  rbpLiveRouteAudit: unknown | null;
-  faLiveRouteAudit: unknown | null;
-  groundedAnswerEligibilityGuard: Record<string, unknown>;
+  temporalGuard: TemporalGuardAudit;
+  polarityAudit: DomainPolarityGuardAudit;
+  claimAnchorAudit: ClaimBoundAnchorAudit;
+  targetedEvidenceAudit: TargetedEvidenceAcquisitionAudit;
+  evidenceAdmissibilityGateAudit: EvidenceAdmissibilityAudit;
+  rbpLiveRouteAudit: RbpLiveRouteAuditDebug | null;
+  faLiveRouteAudit: FaLiveRouteAuditDebug | null;
+  groundedAnswerEligibilityGuard: GroundedAnswerEligibilityAudit;
  followupStateUsage: AssistantFollowupUsage | null;
  composition: {
    reply_type: AssistantReplyType;
-    fallback_type: unknown;
+    fallback_type: AssistantFallbackType;
    answer_structure_v11?: AnswerStructureV11 | null;
    problem_centric_answer_applied?: boolean;
    problem_units_used_count?: number;
-    problem_answer_mode?: string;
+    problem_answer_mode?: AssistantProblemAnswerMode;
    problem_unit_ids_used?: string[];
  };
  safeAssistantReplyBase: string;
  featureContractsV11: boolean;
  featureAnswerPolicyV11: boolean;
-  investigationStateSnapshot: unknown;
-  addressRuntimeMetaForDeep:
-    | {
-        attempted?: boolean;
-        applied?: boolean;
-        reason?: string | null;
-        provider?: string | null;
-        fallbackRuleHit?: string | null;
-        toolGateDecision?: string | null;
-        toolGateReason?: string | null;
-        predecomposeContract?: unknown;
-        orchestrationContract?: unknown;
-      }
-    | null
-    | undefined;
+  investigationStateSnapshot: InvestigationStateWithProblemUnits | null;
+  addressRuntimeMetaForDeep: AssistantAddressRuntimeMetaForDeep | null | undefined;
 }

 export interface AssistantDeepTurnPackagingOutput {
@ -116,12 +132,25 @@ export interface AssistantDeepTurnPackagingOutput {
  deepAnswerArtifacts: DeepAnswerArtifacts;
  debug: AssistantDebugPayload;
  assistantItem: AssistantConversationItem;
-  deepAnalysisLogDetails: Record<string, unknown>;
+  deepAnalysisLogDetails: DeepAnalysisLogDetails;
+}
+
+type NormalizedFragments =
+  | NormalizedQueryV2["fragments"]
+  | NormalizedQueryV2_0_1["fragments"]
+  | NormalizedQueryV2_0_2["fragments"];
+
+function extractNormalizedFragments(normalized: NormalizeResponsePayload["normalized"] | null): NormalizedFragments {
+  if (!normalized || typeof normalized !== "object") {
+    return [];
+  }
+  const source = normalized as NormalizedQueryV2 | NormalizedQueryV2_0_1 | NormalizedQueryV2_0_2;
+  return Array.isArray(source.fragments) ? source.fragments : [];
 }

 export function assembleAssistantDeepTurnPackaging(input: AssistantDeepTurnPackagingInput): AssistantDeepTurnPackagingOutput {
-  const normalizedPayload = (input.normalized.normalized ?? null) as Record<string, unknown> | null;
-  const normalizedFragments = Array.isArray(normalizedPayload?.["fragments"]) ? (normalizedPayload?.["fragments"] as unknown[]) : [];
+  const normalizedPayload = input.normalized.normalized ?? null;
+  const normalizedFragments = extractNormalizedFragments(normalizedPayload);
  const evidenceBundleAssembly = assembleAssistantEvidenceBundle({
    retrievalCalls: input.retrievalCalls,
    retrievalResults: input.retrievalResults
@ -232,7 +261,7 @@ export function assembleAssistantDeepTurnPackaging(input: AssistantDeepTurnPacka
      problem_units_used_count: input.composition.problem_units_used_count ?? 0,
      problem_answer_mode: input.composition.problem_answer_mode ?? "stage1_policy_v11",
      problem_unit_ids_used: Array.isArray(input.composition.problem_unit_ids_used) ? input.composition.problem_unit_ids_used : [],
-      fallback_type: input.composition.fallback_type as string
+      fallback_type: input.composition.fallback_type
    },
    outcomeClassV1: contractsBundleV1.outcomeClassV1,
    assistantOrchestrationContractsV1: contractsBundleV1.assistantOrchestrationContractsV1,
--- a/llm_normalizer/backend/src/services/assistantDeepTurnPackagingRuntimeAdapter.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnPackagingRuntimeAdapter.ts
@ -1,9 +1,14 @@
 import { nanoid } from "nanoid";
 import type {
+  AssistantAddressRuntimeMetaForDeep,
  AssistantConversationItem,
  AssistantDebugPayload,
+  AssistantDebugRouteRecord,
+  AssistantExecutionStateRecord,
  AssistantRequirement,
  AnswerGroundingCheck,
+  FaLiveRouteAuditDebug,
+  RbpLiveRouteAuditDebug,
  RequirementCoverageReport,
  UnifiedRetrievalResult
 } from "../types/assistant";
@ -13,6 +18,18 @@ import type { AssistantDeepTurnInputBuilderArgs } from "./assistantDeepTurnInput
 import { buildAssistantDeepTurnPackagingInput } from "./assistantDeepTurnInputBuilder";
 import { assembleAssistantDeepTurnPackaging } from "./assistantDeepTurnPackaging";
 import type { AssistantFollowupUsage } from "./assistantFollowupUsage";
+import type { ClaimBoundAnchorAudit, TargetedEvidenceAcquisitionAudit } from "./assistantClaimBoundEvidence";
+import type { CompanyAnchorSet } from "./companyAnchorResolver";
+import type {
+  DomainPolarityGuardAudit,
+  EvidenceAdmissibilityAudit,
+  GroundedAnswerEligibilityAudit,
+  TemporalGuardAudit
+} from "./assistantRuntimeGuards";
+import type {
+  AssistantRetrievalCallRecord,
+  AssistantRetrievalRawResultRecord
+} from "./assistantDeepTurnRetrievalRuntimeAdapter";
 import type {
  AssistantAnalysisContextForContract,
  AssistantRuntimeAnalysisContextForPrePackaging
@ -22,6 +39,7 @@ import {
  buildAssistantInvestigationStateSnapshot,
  persistAssistantInvestigationStateSnapshot
 } from "./assistantInvestigationStateRuntimeAdapter";
+import type { DeepAnalysisLogDetails } from "./assistantMessageLogAssembler";

 type AssistantDeepTurnCompositionForPackaging = AssistantDeepTurnInputBuilderArgs["composition"] & {
  assistant_reply: string;
@ -45,44 +63,31 @@ export interface AssistantDeepTurnPackagingRuntimeInput {
  coverageEvaluationRequirements: AssistantRequirement[];
  coverageReport: RequirementCoverageReport;
  groundingCheck: AnswerGroundingCheck;
-  retrievalCalls: Array<Record<string, unknown>>;
-  retrievalResultsRaw: unknown[];
+  retrievalCalls: AssistantRetrievalCallRecord[];
+  retrievalResultsRaw: AssistantRetrievalRawResultRecord[];
  retrievalResults: UnifiedRetrievalResult[];
  questionTypeClass: string;
-  companyAnchors: unknown;
+  companyAnchors: CompanyAnchorSet | null;
  runtimeAnalysisContext: AssistantDeepTurnInputBuilderArgs["runtimeAnalysisContext"];
  businessScopeResolution: AssistantDeepTurnInputBuilderArgs["businessScopeResolution"];
-  temporalGuard: Record<string, unknown>;
-  polarityAudit: Record<string, unknown>;
-  claimAnchorAudit: Record<string, unknown>;
-  targetedEvidenceAudit: unknown;
-  evidenceAdmissibilityGateAudit: unknown;
-  rbpLiveRouteAudit: unknown | null;
-  faLiveRouteAudit: unknown | null;
-  groundedAnswerEligibilityGuard: Record<string, unknown>;
+  temporalGuard: TemporalGuardAudit;
+  polarityAudit: DomainPolarityGuardAudit;
+  claimAnchorAudit: ClaimBoundAnchorAudit;
+  targetedEvidenceAudit: TargetedEvidenceAcquisitionAudit;
+  evidenceAdmissibilityGateAudit: EvidenceAdmissibilityAudit;
+  rbpLiveRouteAudit: RbpLiveRouteAuditDebug | null;
+  faLiveRouteAudit: FaLiveRouteAuditDebug | null;
+  groundedAnswerEligibilityGuard: GroundedAnswerEligibilityAudit;
  followupStateUsage?: AssistantFollowupUsage | null;
  followupApplied: boolean;
  composition: AssistantDeepTurnCompositionForPackaging;
  featureContractsV11: boolean;
  featureAnswerPolicyV11: boolean;
  previousInvestigationState: InvestigationStateWithProblemUnits | null | undefined;
-  addressRuntimeMetaForDeep:
-    | {
-        attempted?: boolean;
-        applied?: boolean;
-        reason?: string | null;
-        provider?: string | null;
-        fallbackRuleHit?: string | null;
-        toolGateDecision?: string | null;
-        toolGateReason?: string | null;
-        predecomposeContract?: unknown;
-        orchestrationContract?: unknown;
-      }
-    | null
-    | undefined;
+  addressRuntimeMetaForDeep: AssistantAddressRuntimeMetaForDeep | null | undefined;
  extractDroppedIntentSegments: (normalizedPayload: NormalizeResponsePayload["normalized"]) => string[];
-  buildDebugRoutes: (routeSummary: RouteHintSummary | null) => Array<Record<string, unknown>>;
-  extractExecutionState: (normalizedPayload: NormalizeResponsePayload["normalized"]) => unknown;
+  buildDebugRoutes: (routeSummary: RouteHintSummary | null) => AssistantDebugRouteRecord[];
+  extractExecutionState: (normalizedPayload: NormalizeResponsePayload["normalized"]) => AssistantExecutionStateRecord[];
  sanitizeReply: (value: string, fallback?: string) => string;
  persistInvestigationState: (sessionId: string, snapshot: InvestigationStateWithProblemUnits) => void;
  nowIso?: () => string;
@ -99,13 +104,13 @@ export interface AssistantDeepTurnPackagingRuntimeOutput {
  investigationStateSnapshot: InvestigationStateWithProblemUnits | null;
  droppedIntentSegments: string[];
  analysisContextForContract: AssistantAnalysisContextForContract | null;
-  routesForDebug: Array<Record<string, unknown>>;
-  resolvedExecutionState: unknown;
+  routesForDebug: AssistantDebugRouteRecord[];
+  resolvedExecutionState: AssistantExecutionStateRecord[];
  safeAssistantReplyBase: string;
  safeAssistantReply: string;
  debug: AssistantDebugPayload;
  assistantItem: AssistantConversationItem;
-  deepAnalysisLogDetails: Record<string, unknown>;
+  deepAnalysisLogDetails: DeepAnalysisLogDetails;
 }

 export function runAssistantDeepTurnPackagingRuntime(
--- a/llm_normalizer/backend/src/services/assistantDeepTurnPlanRuntimeAdapter.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnPlanRuntimeAdapter.ts
@ -1,6 +1,7 @@
 import type { AssistantRequirement } from "../types/assistant";
 import type { NormalizeResponsePayload, RouteHintSummary } from "../types/normalizer";
 import type { AssistantExecutionPlanItem } from "./assistantQueryPlanning";
+import type { DomainPolarityGuardAudit, TemporalGuardAudit } from "./assistantRuntimeGuards";

 export interface AssistantRequirementExtractionLike {
  requirements: AssistantRequirement[];
@ -9,7 +10,15 @@ export interface AssistantRequirementExtractionLike {

 export interface AssistantPlanEnforcementAuditLike {
  executionPlan: AssistantExecutionPlanItem[];
-  audit: Record<string, unknown> | null;
+  audit: AssistantLiveRoutePlanAudit | null;
+}
+
+export interface AssistantLiveRoutePlanAudit extends Record<string, unknown> {
+  required_live_calls: string[];
+  route_adjustments_applied: number;
+  rescued_no_route_fragments: number;
+  replaced_routes: string[];
+  route_gap_reason: string | null;
 }

 export interface BuildAssistantDeepTurnExecutionPlanInput {
@ -17,8 +26,8 @@ export interface BuildAssistantDeepTurnExecutionPlanInput {
  normalizedPayload: NormalizeResponsePayload["normalized"];
  userMessage: string;
  claimType: string;
-  temporalGuard: unknown;
-  domainPolarityGuardInitial: unknown;
+  temporalGuard: TemporalGuardAudit;
+  domainPolarityGuardInitial: DomainPolarityGuardAudit;
  extractRequirements: (
    routeSummary: RouteHintSummary | null,
    normalizedPayload: NormalizeResponsePayload["normalized"],
@ -33,20 +42,20 @@ export interface BuildAssistantDeepTurnExecutionPlanInput {
  enforceRbpLiveRoutePlan: (input: {
    executionPlan: AssistantExecutionPlanItem[];
    claimType: string;
-    temporalGuard: unknown;
+    temporalGuard: TemporalGuardAudit;
  }) => AssistantPlanEnforcementAuditLike;
  enforceFaLiveRoutePlan: (input: {
    executionPlan: AssistantExecutionPlanItem[];
    claimType: string;
-    temporalGuard: unknown;
+    temporalGuard: TemporalGuardAudit;
  }) => AssistantPlanEnforcementAuditLike;
  applyTemporalHintToExecutionPlan: (
    executionPlan: AssistantExecutionPlanItem[],
-    temporalGuard: unknown
+    temporalGuard: TemporalGuardAudit
  ) => AssistantExecutionPlanItem[];
  applyPolarityHintToExecutionPlan: (
    executionPlan: AssistantExecutionPlanItem[],
-    domainPolarityGuardInitial: unknown
+    domainPolarityGuardInitial: DomainPolarityGuardAudit
  ) => AssistantExecutionPlanItem[];
 }

--- a/llm_normalizer/backend/src/services/assistantDeepTurnPrePackagingContext.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnPrePackagingContext.ts
@ -1,4 +1,5 @@
 import type { NormalizeResponsePayload, RouteHintSummary } from "../types/normalizer";
+import type { AssistantDebugRouteRecord, AssistantExecutionStateRecord } from "../types/assistant";

 export interface AssistantRuntimeAnalysisContextForPrePackaging {
  active: boolean;
@ -23,16 +24,16 @@ export interface BuildAssistantDeepTurnPrePackagingContextInput {
  runtimeAnalysisContext: AssistantRuntimeAnalysisContextForPrePackaging;
  assistantReply: string;
  extractDroppedIntentSegments: (normalizedPayload: NormalizeResponsePayload["normalized"]) => string[];
-  buildDebugRoutes: (routeSummary: RouteHintSummary | null) => Array<Record<string, unknown>>;
-  extractExecutionState: (normalizedPayload: NormalizeResponsePayload["normalized"]) => unknown;
+  buildDebugRoutes: (routeSummary: RouteHintSummary | null) => AssistantDebugRouteRecord[];
+  extractExecutionState: (normalizedPayload: NormalizeResponsePayload["normalized"]) => AssistantExecutionStateRecord[];
  sanitizeReply: (value: string, fallback?: string) => string;
 }

 export interface AssistantDeepTurnPrePackagingContext {
  droppedIntentSegments: string[];
  analysisContextForContract: AssistantAnalysisContextForContract | null;
-  routesForDebug: Array<Record<string, unknown>>;
-  resolvedExecutionState: unknown;
+  routesForDebug: AssistantDebugRouteRecord[];
+  resolvedExecutionState: AssistantExecutionStateRecord[];
  safeAssistantReplyBase: string;
 }

--- a/llm_normalizer/backend/src/services/assistantDeepTurnResponseRuntimeAdapter.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnResponseRuntimeAdapter.ts
@ -1,4 +1,12 @@
-import type { AssistantDebugPayload, AssistantMessageResponsePayload } from "../types/assistant";
+import type {
+  AssistantAddressRuntimeMetaForDeep,
+  AssistantDebugPayload,
+  AssistantDebugRouteRecord,
+  AssistantExecutionStateRecord,
+  AssistantMessageResponsePayload,
+  FaLiveRouteAuditDebug,
+  RbpLiveRouteAuditDebug
+} from "../types/assistant";
 import type { NormalizeResponsePayload, RouteHintSummary } from "../types/normalizer";
 import type { InvestigationStateWithProblemUnits } from "../types/stage2ProblemUnits";
 import {
@ -7,10 +15,23 @@ import {
  type AssistantDeepTurnPackagingRuntimeOutput
 } from "./assistantDeepTurnPackagingRuntimeAdapter";
 import type { RunAssistantDeepTurnAnalysisRuntimeOutput } from "./assistantDeepTurnAnalysisRuntimeAdapter";
+import type { AssistantExecutionPlanItem } from "./assistantQueryPlanning";
 import {
  finalizeAssistantDeepTurn,
  type FinalizeAssistantDeepTurnInput
 } from "./assistantDeepTurnFinalizeRuntimeAdapter";
+import type { ClaimBoundAnchorAudit, TargetedEvidenceAcquisitionAudit } from "./assistantClaimBoundEvidence";
+import type { CompanyAnchorSet } from "./companyAnchorResolver";
+import type {
+  DomainPolarityGuardAudit,
+  EvidenceAdmissibilityAudit,
+  GroundedAnswerEligibilityAudit,
+  TemporalGuardAudit
+} from "./assistantRuntimeGuards";
+import type {
+  AssistantRetrievalCallRecord,
+  AssistantRetrievalRawResultRecord
+} from "./assistantDeepTurnRetrievalRuntimeAdapter";

 export interface RunAssistantDeepTurnResponseRuntimeInput {
  featureInvestigationStateV1: boolean;
@ -27,16 +48,16 @@ export interface RunAssistantDeepTurnResponseRuntimeInput {
  };
  normalizedQuestion: string;
  routeSummary: RouteHintSummary | null;
-  executionPlan: unknown[];
+  executionPlan: AssistantExecutionPlanItem[];
  requirementExtractionRequirements: AssistantDeepTurnPackagingRuntimeInput["requirementExtractionRequirements"];
  coverageEvaluationRequirements: AssistantDeepTurnPackagingRuntimeInput["coverageEvaluationRequirements"];
  coverageReport: AssistantDeepTurnPackagingRuntimeInput["coverageReport"];
  groundingCheck: AssistantDeepTurnPackagingRuntimeInput["groundingCheck"];
-  retrievalCalls: unknown[];
-  retrievalResultsRaw: unknown[];
+  retrievalCalls: AssistantRetrievalCallRecord[];
+  retrievalResultsRaw: AssistantRetrievalRawResultRecord[];
  retrievalResults: AssistantDeepTurnPackagingRuntimeInput["retrievalResults"];
  questionTypeClass: AssistantDeepTurnPackagingRuntimeInput["questionTypeClass"];
-  companyAnchors: RunAssistantDeepTurnAnalysisRuntimeOutput["companyAnchors"];
+  companyAnchors: CompanyAnchorSet | null;
  runtimeAnalysisContext: {
    active: boolean;
    as_of_date: string | null;
@ -46,22 +67,22 @@ export interface RunAssistantDeepTurnResponseRuntimeInput {
    snapshot_mode?: "auto" | "force_snapshot" | "force_live";
  };
  businessScopeResolution: RunAssistantDeepTurnAnalysisRuntimeOutput["businessScopeResolution"];
-  temporalGuard: unknown;
-  polarityAudit: unknown;
-  claimAnchorAudit: unknown;
-  targetedEvidenceAudit: unknown;
-  evidenceAdmissibilityGateAudit: unknown;
-  rbpLiveRouteAudit: unknown;
-  faLiveRouteAudit: unknown;
-  groundedAnswerEligibilityGuard: unknown;
+  temporalGuard: TemporalGuardAudit;
+  polarityAudit: DomainPolarityGuardAudit;
+  claimAnchorAudit: ClaimBoundAnchorAudit;
+  targetedEvidenceAudit: TargetedEvidenceAcquisitionAudit;
+  evidenceAdmissibilityGateAudit: EvidenceAdmissibilityAudit;
+  rbpLiveRouteAudit: RbpLiveRouteAuditDebug | null;
+  faLiveRouteAudit: FaLiveRouteAuditDebug | null;
+  groundedAnswerEligibilityGuard: GroundedAnswerEligibilityAudit;
  followupStateUsage: AssistantDeepTurnPackagingRuntimeInput["followupStateUsage"];
  followupApplied: boolean;
  composition: AssistantDeepTurnPackagingRuntimeInput["composition"];
  previousInvestigationState: AssistantDeepTurnPackagingRuntimeInput["previousInvestigationState"];
  addressRuntimeMetaForDeep: AssistantDeepTurnPackagingRuntimeInput["addressRuntimeMetaForDeep"];
  extractDroppedIntentSegments: (normalizedPayload: NormalizeResponsePayload["normalized"]) => string[];
-  buildDebugRoutes: (routeSummary: RouteHintSummary | null) => Array<Record<string, unknown>>;
-  extractExecutionState: (normalizedPayload: NormalizeResponsePayload["normalized"]) => unknown[];
+  buildDebugRoutes: (routeSummary: RouteHintSummary | null) => AssistantDebugRouteRecord[];
+  extractExecutionState: (normalizedPayload: NormalizeResponsePayload["normalized"]) => AssistantExecutionStateRecord[];
  sanitizeReply: (value: string, fallback?: string) => string;
  persistInvestigationState: (sessionId: string, snapshot: InvestigationStateWithProblemUnits) => void;
  messageIdFactory: () => string;
@ -81,14 +102,7 @@ export interface RunAssistantDeepTurnResponseRuntimeOutput {
  debug: AssistantDebugPayload;
 }

-function toRecordObject(value: unknown): Record<string, unknown> | null {
-  if (!value || typeof value !== "object") {
-    return null;
-  }
-  return value as Record<string, unknown>;
-}
-
-function toNullableString(value: unknown): string | null {
+function toNullableString(value: string | null | undefined): string | null {
  if (typeof value !== "string") {
    return null;
  }
@ -96,7 +110,7 @@ function toNullableString(value: unknown): string | null {
  return trimmed.length > 0 ? trimmed : null;
 }

-function toStringArray(value: unknown): string[] {
+function toStringArray(value: string[] | null | undefined): string[] {
  if (!Array.isArray(value)) {
    return [];
  }
@ -105,44 +119,30 @@ function toStringArray(value: unknown): string[] {
    .filter((item) => item.length > 0);
 }

-function toSnapshotMode(value: unknown): "auto" | "force_snapshot" | "force_live" {
+function toSnapshotMode(value: string | null | undefined): "auto" | "force_snapshot" | "force_live" {
  return value === "force_snapshot" || value === "force_live" ? value : "auto";
 }

-function normalizeExecutionPlan(value: unknown[]): AssistantDeepTurnPackagingRuntimeInput["executionPlan"] {
+function normalizeExecutionPlan(
+  value: AssistantExecutionPlanItem[]
+): AssistantDeepTurnPackagingRuntimeInput["executionPlan"] {
  if (!Array.isArray(value)) {
    return [];
  }
-  return value.map((item, index) => {
-    const source = toRecordObject(item);
-    return {
-      fragment_id: toNullableString(source?.fragment_id) ?? `fragment_${index + 1}`,
-      requirement_ids: toStringArray(source?.requirement_ids),
-      route: toNullableString(source?.route) ?? "unknown_route",
-      should_execute: Boolean(source?.should_execute),
-      no_route_reason: toNullableString(source?.no_route_reason),
-      clarification_reason: toNullableString(source?.clarification_reason)
-    };
-  });
-}
-
-function normalizeRecordArray(value: unknown[]): Array<Record<string, unknown>> {
-  if (!Array.isArray(value)) {
-    return [];
-  }
-  return value
-    .map((item) => toRecordObject(item))
-    .filter((item): item is Record<string, unknown> => Boolean(item));
-}
-
-function normalizeRecord(value: unknown): Record<string, unknown> {
-  return toRecordObject(value) ?? {};
+  return value.map((item, index) => ({
+    fragment_id: toNullableString(item.fragment_id) ?? `fragment_${index + 1}`,
+    requirement_ids: toStringArray(item.requirement_ids),
+    route: toNullableString(item.route) ?? "unknown_route",
+    should_execute: Boolean(item.should_execute),
+    no_route_reason: toNullableString(item.no_route_reason ?? null),
+    clarification_reason: toNullableString(item.clarification_reason ?? null)
+  }));
 }

 function normalizeRuntimeAnalysisContext(
-  value: unknown
+  value: RunAssistantDeepTurnResponseRuntimeInput["runtimeAnalysisContext"] | null | undefined
 ): AssistantDeepTurnPackagingRuntimeInput["runtimeAnalysisContext"] {
-  const source = toRecordObject(value);
+  const source = value;
  return {
    active: Boolean(source?.active),
    as_of_date: toNullableString(source?.as_of_date),
@ -154,9 +154,9 @@ function normalizeRuntimeAnalysisContext(
 }

 function normalizeBusinessScopeResolution(
-  value: unknown
+  value: RunAssistantDeepTurnAnalysisRuntimeOutput["businessScopeResolution"] | null | undefined
 ): AssistantDeepTurnPackagingRuntimeInput["businessScopeResolution"] {
-  const source = toRecordObject(value);
+  const source = value;
  return {
    business_scope_raw: toStringArray(source?.business_scope_raw),
    business_scope_resolved: toStringArray(source?.business_scope_resolved),
@ -166,9 +166,9 @@ function normalizeBusinessScopeResolution(
 }

 function normalizeAddressRuntimeMetaForDeep(
-  value: unknown
-): AssistantDeepTurnPackagingRuntimeInput["addressRuntimeMetaForDeep"] {
-  const source = toRecordObject(value);
+  value: AssistantAddressRuntimeMetaForDeep | null | undefined
+): AssistantAddressRuntimeMetaForDeep | null {
+  const source = value;
  if (!source) {
    return null;
  }
@ -180,8 +180,8 @@ function normalizeAddressRuntimeMetaForDeep(
    fallbackRuleHit: toNullableString(source.fallbackRuleHit),
    toolGateDecision: toNullableString(source.toolGateDecision),
    toolGateReason: toNullableString(source.toolGateReason),
-    predecomposeContract: toRecordObject(source.predecomposeContract),
-    orchestrationContract: toRecordObject(source.orchestrationContract)
+    predecomposeContract: source.predecomposeContract ?? null,
+    orchestrationContract: source.orchestrationContract ?? null
  };
 }

@ -204,21 +204,21 @@ export function runAssistantDeepTurnResponseRuntime(
    coverageEvaluationRequirements: input.coverageEvaluationRequirements,
    coverageReport: input.coverageReport,
    groundingCheck: input.groundingCheck,
-    retrievalCalls: normalizeRecordArray(input.retrievalCalls),
-    retrievalResultsRaw: Array.isArray(input.retrievalResultsRaw) ? input.retrievalResultsRaw : [],
+    retrievalCalls: input.retrievalCalls,
+    retrievalResultsRaw: input.retrievalResultsRaw,
    retrievalResults: input.retrievalResults,
    questionTypeClass: input.questionTypeClass,
    companyAnchors: input.companyAnchors,
    runtimeAnalysisContext: normalizeRuntimeAnalysisContext(input.runtimeAnalysisContext),
    businessScopeResolution: normalizeBusinessScopeResolution(input.businessScopeResolution),
-    temporalGuard: normalizeRecord(input.temporalGuard),
-    polarityAudit: normalizeRecord(input.polarityAudit),
-    claimAnchorAudit: normalizeRecord(input.claimAnchorAudit),
+    temporalGuard: input.temporalGuard,
+    polarityAudit: input.polarityAudit,
+    claimAnchorAudit: input.claimAnchorAudit,
    targetedEvidenceAudit: input.targetedEvidenceAudit,
    evidenceAdmissibilityGateAudit: input.evidenceAdmissibilityGateAudit,
    rbpLiveRouteAudit: input.rbpLiveRouteAudit ?? null,
    faLiveRouteAudit: input.faLiveRouteAudit ?? null,
-    groundedAnswerEligibilityGuard: normalizeRecord(input.groundedAnswerEligibilityGuard),
+    groundedAnswerEligibilityGuard: input.groundedAnswerEligibilityGuard,
    followupStateUsage: input.followupStateUsage,
    followupApplied: input.followupApplied,
    composition: input.composition,
--- a/llm_normalizer/backend/src/services/assistantDeepTurnRetrievalRuntimeAdapter.ts
+++ b/llm_normalizer/backend/src/services/assistantDeepTurnRetrievalRuntimeAdapter.ts
@ -18,10 +18,43 @@ export interface AssistantRetrievalCallRecord {
  reason: string | null;
 }

+export type AssistantRetrievalScalar = string | number | boolean | null;
+export type AssistantRetrievalFieldValue =
+  | AssistantRetrievalScalar
+  | AssistantRetrievalFieldValue[]
+  | { [key: string]: AssistantRetrievalFieldValue };
+export type AssistantRetrievalRecord = Record<string, AssistantRetrievalFieldValue>;
+
+export interface AssistantRetrievalRawResultLike {
+  status: "ok" | "empty" | "partial" | "error";
+  result_type: "list" | "summary" | "object" | "chain" | "ranking";
+  items: AssistantRetrievalRecord[];
+  summary: AssistantRetrievalRecord;
+  evidence: AssistantRetrievalRecord[];
+  why_included: string[];
+  selection_reason: string[];
+  risk_factors: string[];
+  business_interpretation: string[];
+  confidence: "high" | "medium" | "low";
+  limitations: string[];
+  errors: string[];
+}
+
+export type AssistantRetrievalRawListItem = AssistantRetrievalFieldValue;
+export type AssistantRetrievalRawList = AssistantRetrievalFieldValue[];
+
+export type AssistantRetrievalRawResult =
+  | AssistantRetrievalRawResultLike
+  | AssistantRetrievalRawList
+  | string
+  | number
+  | boolean
+  | null;
+
 export interface AssistantRetrievalRawResultRecord {
  fragment_id: string;
  route: string;
-  raw_result: unknown;
+  raw_result: AssistantRetrievalRawResult;
 }

 export interface AssistantDeepTurnRetrievalExecutionInput {
@ -33,7 +66,7 @@ export interface AssistantDeepTurnRetrievalExecutionInput {
    options: {
      temporalHint: AssistantLiveTemporalHint | null;
    }
-  ) => Promise<unknown>;
+  ) => Promise<AssistantRetrievalRawResult>;
  mapNoRouteReason: (reason: string | null) => string;
  buildSkippedResult: (item: AssistantExecutionPlanItem) => UnifiedRetrievalResult;
  normalizeRetrievalResultFn?: typeof normalizeRetrievalResult;
@ -45,7 +78,7 @@ export interface AssistantDeepTurnRetrievalExecutionOutput {
  retrievalResults: UnifiedRetrievalResult[];
 }

-function buildRouteExecutorErrorRawResult(route: string, message: string): Record<string, unknown> {
+function buildRouteExecutorErrorRawResult(route: string, message: string): AssistantRetrievalRawResult {
  return {
    status: "error",
    result_type: "summary",
@ -64,6 +97,24 @@ function buildRouteExecutorErrorRawResult(route: string, message: string): Recor
  };
 }

+function normalizeRawResult(
+  value: AssistantRetrievalRawResult | object | null | undefined
+): AssistantRetrievalRawResult {
+  if (value === null || value === undefined) {
+    return null;
+  }
+  if (typeof value === "string" || typeof value === "number" || typeof value === "boolean") {
+    return value;
+  }
+  if (Array.isArray(value)) {
+    return value as AssistantRetrievalRawList;
+  }
+  if (typeof value === "object") {
+    return value as AssistantRetrievalRawResultLike;
+  }
+  return null;
+}
+
 export async function executeAssistantDeepTurnRetrievalPlan(
  input: AssistantDeepTurnRetrievalExecutionInput
 ): Promise<AssistantDeepTurnRetrievalExecutionOutput> {
@ -99,13 +150,14 @@ export async function executeAssistantDeepTurnRetrievalPlan(
      const raw = await input.executeRouteRuntime(planItem.route, planItem.fragment_text, {
        temporalHint: input.liveTemporalHint
      });
+      const normalizedRaw = normalizeRawResult(raw);
      retrievalResultsRaw.push({
        fragment_id: planItem.fragment_id,
        route: planItem.route,
-        raw_result: raw
+        raw_result: normalizedRaw
      });
      retrievalResults.push(
-        normalizeRetrievalResultSafe(planItem.fragment_id, planItem.requirement_ids, planItem.route, raw)
+        normalizeRetrievalResultSafe(planItem.fragment_id, planItem.requirement_ids, planItem.route, normalizedRaw)
      );
    } catch (error) {
      const message = error instanceof Error ? error.message : String(error);
--- a/llm_normalizer/backend/src/services/assistantEvidenceBundleAssembler.ts
+++ b/llm_normalizer/backend/src/services/assistantEvidenceBundleAssembler.ts
@ -3,6 +3,7 @@ import {
  buildAssistantEvidenceBundleContractV1,
  type AssistantEvidenceBundleContractV1
 } from "./assistantOrchestrationContracts";
+import type { AssistantRetrievalCallRecord } from "./assistantDeepTurnRetrievalRuntimeAdapter";

 type RetrievalStatusItem = AssistantDebugPayload["retrieval_status"][number];

@ -22,7 +23,7 @@ function buildRetrievalStatus(retrievalResults: UnifiedRetrievalResult[]): Retri
 }

 export function assembleAssistantEvidenceBundle(input: {
-  retrievalCalls: Array<Record<string, unknown>>;
+  retrievalCalls: AssistantRetrievalCallRecord[];
  retrievalResults: UnifiedRetrievalResult[];
 }): AssistantEvidenceBundleAssembly {
  const retrievalResults = Array.isArray(input.retrievalResults) ? input.retrievalResults : [];
--- a/llm_normalizer/backend/src/services/assistantMessageLogAssembler.ts
+++ b/llm_normalizer/backend/src/services/assistantMessageLogAssembler.ts
@ -1,4 +1,33 @@
-import type { AnswerGroundingCheck, RequirementCoverageReport } from "../types/assistant";
+import type {
+  AssistantDebugRouteRecord,
+  AssistantExecutionStateRecord,
+  AssistantFallbackType,
+  AssistantProblemAnswerMode,
+  AssistantReplyType,
+  AssistantRequirement,
+  UnifiedRetrievalResult,
+  AnswerGroundingCheck,
+  FaLiveRouteAuditDebug,
+  RbpLiveRouteAuditDebug,
+  RequirementCoverageReport
+} from "../types/assistant";
+import type { AnswerStructureV11 } from "../types/stage1Contracts";
+import type { InvestigationStateWithProblemUnits } from "../types/stage2ProblemUnits";
+import type { ClaimBoundAnchorAudit, TargetedEvidenceAcquisitionAudit } from "./assistantClaimBoundEvidence";
+import type { AssistantContractsBundleV1 } from "./assistantContractsBundleAssembler";
+import type { AssistantOutcomeClassV1 } from "./assistantOrchestrationContracts";
+import type { AssistantFollowupUsage } from "./assistantFollowupUsage";
+import type { CompanyAnchorSet } from "./companyAnchorResolver";
+import type {
+  AssistantRetrievalCallRecord,
+  AssistantRetrievalRawResultRecord
+} from "./assistantDeepTurnRetrievalRuntimeAdapter";
+import type {
+  DomainPolarityGuardAudit,
+  EvidenceAdmissibilityAudit,
+  GroundedAnswerEligibilityAudit,
+  TemporalGuardAudit
+} from "./assistantRuntimeGuards";

 export interface DeepAnalysisMessageLogDetailsInput {
  sessionId: string;
@ -6,18 +35,18 @@ export interface DeepAnalysisMessageLogDetailsInput {
  userMessage: string;
  normalizerOutput: unknown;
  executionPlan: Array<Record<string, unknown>>;
-  resolvedExecutionState: unknown;
-  routes: Array<Record<string, unknown>>;
-  retrievalCalls: Array<Record<string, unknown>>;
-  retrievalResultsRaw: unknown[];
-  retrievalResultsNormalized: unknown[];
-  requirementsExtracted: unknown[];
+  resolvedExecutionState: AssistantExecutionStateRecord[];
+  routes: AssistantDebugRouteRecord[];
+  retrievalCalls: AssistantRetrievalCallRecord[];
+  retrievalResultsRaw: AssistantRetrievalRawResultRecord[];
+  retrievalResultsNormalized: UnifiedRetrievalResult[];
+  requirementsExtracted: AssistantRequirement[];
  coverageReport: RequirementCoverageReport;
  groundingCheck: AnswerGroundingCheck;
-  replyType: string;
+  replyType: AssistantReplyType;
  droppedIntentSegments: string[];
  questionTypeClass: string;
-  companyAnchors: unknown;
+  companyAnchors: CompanyAnchorSet | null;
  runtimeAnalysisContext: {
    active: boolean;
    as_of_date: string | null;
@ -32,26 +61,26 @@ export interface DeepAnalysisMessageLogDetailsInput {
    company_grounding_applied?: boolean;
    scope_resolution_reason?: string[];
  };
-  temporalGuard: Record<string, unknown>;
-  polarityAudit: Record<string, unknown>;
-  claimAnchorAudit: Record<string, unknown>;
-  targetedEvidenceAudit: unknown;
-  evidenceAdmissibilityGateAudit: unknown;
-  rbpLiveRouteAudit: unknown | null;
-  faLiveRouteAudit: unknown | null;
-  groundedAnswerEligibilityGuard: Record<string, unknown>;
-  followupStateUsage: unknown | null;
+  temporalGuard: TemporalGuardAudit;
+  polarityAudit: DomainPolarityGuardAudit;
+  claimAnchorAudit: ClaimBoundAnchorAudit;
+  targetedEvidenceAudit: TargetedEvidenceAcquisitionAudit;
+  evidenceAdmissibilityGateAudit: EvidenceAdmissibilityAudit;
+  rbpLiveRouteAudit: RbpLiveRouteAuditDebug | null;
+  faLiveRouteAudit: FaLiveRouteAuditDebug | null;
+  groundedAnswerEligibilityGuard: GroundedAnswerEligibilityAudit;
+  followupStateUsage: AssistantFollowupUsage | null;
  compositionDebug: {
    problem_centric_answer_applied?: boolean;
    problem_units_used_count?: number;
-    problem_answer_mode?: string;
+    problem_answer_mode?: AssistantProblemAnswerMode;
    problem_unit_ids_used?: string[];
-    fallback_type?: string;
+    fallback_type?: AssistantFallbackType;
  };
-  outcomeClassV1: unknown;
-  assistantOrchestrationContractsV1: unknown;
-  answerStructureV11: unknown;
-  investigationStateSnapshot: unknown;
+  outcomeClassV1: AssistantOutcomeClassV1;
+  assistantOrchestrationContractsV1: AssistantContractsBundleV1["assistantOrchestrationContractsV1"];
+  answerStructureV11: AnswerStructureV11 | null;
+  investigationStateSnapshot: InvestigationStateWithProblemUnits | null;
  assistantReply: string;
  traceId: string;
 }
@ -77,7 +106,9 @@ function resolveCoverageStatus(coverageReport: RequirementCoverageReport): "full
    : "partial_or_limited";
 }

-export function buildDeepAnalysisProcessedLogDetails(input: DeepAnalysisMessageLogDetailsInput): Record<string, unknown> {
+export type DeepAnalysisLogDetails = Record<string, unknown>;
+
+export function buildDeepAnalysisProcessedLogDetails(input: DeepAnalysisMessageLogDetailsInput): DeepAnalysisLogDetails {
  const analysisContext = toAnalysisContext(input.runtimeAnalysisContext);
  return {
    session_id: input.sessionId,
--- a/llm_normalizer/backend/src/services/assistantOrchestrationContracts.ts
+++ b/llm_normalizer/backend/src/services/assistantOrchestrationContracts.ts
@ -6,6 +6,7 @@ import type {
  UnifiedRetrievalResult
 } from "../types/assistant";
 import type { NormalizedPayload, RouteHintSummary } from "../types/normalizer";
+import type { AssistantRetrievalCallRecord } from "./assistantDeepTurnRetrievalRuntimeAdapter";

 export type AssistantOutcomeClassV1 =
  | "FULLY_ANSWERED"
@ -241,7 +242,7 @@ export function buildAssistantExecutionPlanContractV1(input: {
 }

 export function buildAssistantEvidenceBundleContractV1(input: {
-  retrievalCalls: Array<Record<string, unknown>>;
+  retrievalCalls: AssistantRetrievalCallRecord[];
  retrievalResults: UnifiedRetrievalResult[];
 }): AssistantEvidenceBundleContractV1 {
  const retrievalResults = Array.isArray(input.retrievalResults) ? input.retrievalResults : [];
--- a/llm_normalizer/backend/src/services/assistantQueryPlanning.ts
+++ b/llm_normalizer/backend/src/services/assistantQueryPlanning.ts
@ -1,3 +1,4 @@
+import type { AssistantDebugRouteRecord } from "../types/assistant";
 import type { RouteHintSummary } from "../types/normalizer";

 interface FragmentLike {
@ -106,7 +107,7 @@ export function buildExecutionPlanFromRoute(input: {
 export function buildDebugRoutesFromRoute(input: {
  routeSummary: RouteHintSummary | null;
  resolveLegacyRouteReason: (route: string) => string;
-}): Array<Record<string, unknown>> {
+}): AssistantDebugRouteRecord[] {
  if (!input.routeSummary) {
    return [];
  }
--- a/llm_normalizer/backend/src/services/assistantRuntimeGuards.ts
+++ b/llm_normalizer/backend/src/services/assistantRuntimeGuards.ts
@ -4,6 +4,7 @@ import type { CompanyAnchorSet } from "./companyAnchorResolver";
 import type { EvidenceItem } from "../types/stage1Contracts";
 import type { ProblemUnit } from "../types/stage2ProblemUnits";
 import type { ClaimBoundAnchorAudit } from "./assistantClaimBoundEvidence";
+import iconv from "iconv-lite";

 type P0DomainHint =
  | "settlements_60_62"
@ -895,12 +896,70 @@ export interface DomainPolarityGuardAudit {
  reason_codes: string[];
 }

+function mojibakeScoreForRuntimeGuards(value: string): number {
+  const source = String(value ?? "");
+  const cyrillic = (source.match(/[А-Яа-яЁё]/g) ?? []).length;
+  const latin = (source.match(/[A-Za-z]/g) ?? []).length;
+  const hardMarkers = (source.match(/[ѓ“‚„…†‡€‰‹‰ЉЊ‹Џ‘’“”•–—™љ›њћџ]/g) ?? []).length;
+  const pairMarkers = (source.match(/(?:Р.|С.|Гђ.|Г‘.)/g) ?? []).length;
+  const doubleEncodedMarkers = (source.match(/(?:Р“.|Р’.|Гѓ.|Г‚.)/gu) ?? []).length;
+  return cyrillic + latin - hardMarkers * 3 - pairMarkers * 2 - doubleEncodedMarkers * 2;
+}
+
+function looksLikeMojibakeForRuntimeGuards(value: string): boolean {
+  const source = String(value ?? "");
+  if (!source.trim()) {
+    return false;
+  }
+  if (/[ѓ“‚„…†‡€‰‹‰ЉЊ‹Џ‘’“”•–—™љ›њћџ]/.test(source)) {
+    return true;
+  }
+  if ((source.match(/(?:Р.|С.|Гђ.|Г‘.)/g) ?? []).length >= 2) {
+    return true;
+  }
+  return (source.match(/(?:Р“.|Р’.|Гѓ.|Г‚.)/gu) ?? []).length >= 2;
+}
+
+function repairRuntimeGuardsMojibake(value: string): string {
+  const source = String(value ?? "");
+  if (!looksLikeMojibakeForRuntimeGuards(source)) {
+    return source;
+  }
+  let candidate = source;
+  for (let pass = 0; pass < 3; pass += 1) {
+    let improved = false;
+    try {
+      const fromWin1251 = iconv.encode(candidate, "win1251").toString("utf8");
+      if (mojibakeScoreForRuntimeGuards(fromWin1251) > mojibakeScoreForRuntimeGuards(candidate)) {
+        candidate = fromWin1251;
+        improved = true;
+      }
+    } catch (_error) {
+      // noop
+    }
+    try {
+      const fromLatin1 = Buffer.from(candidate, "latin1").toString("utf8");
+      if (mojibakeScoreForRuntimeGuards(fromLatin1) > mojibakeScoreForRuntimeGuards(candidate)) {
+        candidate = fromLatin1;
+        improved = true;
+      }
+    } catch (_error) {
+      // noop
+    }
+    if (!improved) {
+      break;
+    }
+  }
+  return candidate;
+}
+
 export function resolveDomainPolarityGuard(input: {
  userMessage: string;
  companyAnchors?: CompanyAnchorSet | null;
  focusDomainHint?: string | null;
 }): DomainPolarityGuardAudit {
-  const lower = String(input.userMessage ?? "").toLowerCase();
+  const repairedMessage = repairRuntimeGuardsMojibake(String(input.userMessage ?? ""));
+  const lower = repairedMessage.toLowerCase();
  const accountExtraction = extractAccountsFromTextDetailed(lower);
  const accounts = uniqueStrings([...(input.companyAnchors?.accounts ?? []), ...accountExtraction.resolved_account_anchors]);
  const prefixes = new Set(accounts.map((item) => accountPrefix(item)).filter((item): item is string => Boolean(item)));
@ -1683,7 +1742,8 @@ export function applyEligibilityToGroundingCheck<T extends { status: string; rea
  const reasonMap: Record<string, string> = {
    admissible_evidence_count_zero: "Недостаточно подтвержденных данных для уверенного ответа.",
    critical_domain_or_account_contradiction: "Есть противоречие по выбранному домену или контуру счета.",
-    temporal_guard_failed_out_of_snapshot_window: "Запрошенный период выходит за доступный срез данных.",
+    temporal_guard_failed_out_of_snapshot_window:
+      "Запрошенный период выходит за доступный срез данных. Temporal anchor outside snapshot window.",
    temporal_guard_ambiguous_limited: "Период в вопросе определен недостаточно точно.",
    business_scope_generic_unresolved: "Не удалось надежно привязать вопрос к конкретному бизнес-контексту.",
    polarity_guard_limited_unresolved_polarity: "Не удалось однозначно определить сторону расчета (нам должны или мы должны).",
--- a/llm_normalizer/backend/src/services/assistantService.ts
+++ b/llm_normalizer/backend/src/services/assistantService.ts
@ -1040,7 +1040,12 @@ function hasCrossScopeConflictWithState(userMessage, state) {
    const inferredDomain = inferP0DomainFromMessage(userMessage);
    const stateDomain = compactWhitespace(state.followup_context?.active_domain ?? state.focus.domain ?? "");
    if (inferredDomain && stateDomain && inferredDomain !== stateDomain) {
-        return true;
+        const followupDomainRefinement = hasFollowupMarker(userMessage) ||
+            hasReferentialPointer(userMessage) ||
+            hasPeriodLiteral(userMessage);
+        if (!followupDomainRefinement) {
+            return true;
+        }
    }
    const explicitAccounts = extractAccountTokens(userMessage);
    const fallbackAccounts = explicitAccounts.length > 0 ? explicitAccounts : extractFollowupAccountAnchorsLoose(userMessage);
@ -1070,9 +1075,11 @@ function inferP0DomainFromMessage(text) {
    return null;
 }
 function hasStrongFollowupAnchors(userMessage, state) {
+    const normalizedMessage = compactWhitespace(repairAddressMojibake(String(userMessage ?? "")).toLowerCase());
+    const periodRefinementCue = /(?:^(?:\u0430\s+)?\u0435\u0441\u043b\u0438|\u0442\u043e\u043b\u044c\u043a\u043e\s+\u0437\u0430|\u0441\u043c\u043e\u0442\u0440\u0435\u0442\u044c|\u043f\u043e\s+\u043f\u0435\u0440\u0438\u043e\u0434\u0443|\u0437\u0430\s+\u0438\u044e\u043d\u044c|\u0437\u0430\s+\u0438\u044e\u043b\u044c)/iu.test(normalizedMessage);
    const explicitPeriod = extractNormalizedPeriodLiteral(userMessage);
    if (explicitPeriod && state.focus.period && explicitPeriod !== state.focus.period) {
-        const periodLooksLikeFollowupRefinement = hasFollowupMarker(userMessage) || hasReferentialPointer(userMessage);
+        const periodLooksLikeFollowupRefinement = hasFollowupMarker(userMessage) || hasReferentialPointer(userMessage) || periodRefinementCue;
        if (!periodLooksLikeFollowupRefinement) {
            return true;
        }
@ -2969,26 +2976,33 @@ function resolveAddressToolGateDecision(addressInputMessage, followupContext, ll
            reason: dataScopeMetaQuery ? "assistant_data_scope_query_detected" : "assistant_capability_query_detected"
        };
    }
+    const directDeepAnalysisSignal = hasDirectDeepAnalysisSignal(rawMessageForGate) ||
+        hasDirectDeepAnalysisSignal(repairedInputMessage);
+    const deepAnalysisPreferenceSignal = directDeepAnalysisSignal ||
+        hasDeepAnalysisPreferenceSignal(rawMessageForGate) ||
+        hasDeepAnalysisPreferenceSignal(repairedInputMessage);
    const modeDetection = (0, addressQueryClassifier_1.detectAddressQuestionMode)(repairedInputMessage || addressInputMessage);
    const hasClassifierSignal = modeDetection.mode === "address_query";
    const llmContractMode = toNonEmptyString(llmPreDecomposeMeta?.predecomposeContract?.mode);
    const llmContractModeConfidence = toNonEmptyString(llmPreDecomposeMeta?.predecomposeContract?.mode_confidence);
    const llmContractIntent = toNonEmptyString(llmPreDecomposeMeta?.predecomposeContract?.intent);
-    const llmContractIntentConfidence = toNonEmptyString(llmPreDecomposeMeta?.predecomposeContract?.intent_confidence);
+    const llmCanonicalEntitySignal = /(?:\u0437\u0430\u043a\u0430\u0437\u0447\u0438\u043a|\u043f\u043e\u0441\u0442\u0430\u0432\u0449\u0438\u043a|\u043a\u043e\u043d\u0442\u0440\u0430\u0433\u0435\u043d\u0442|\u043a\u043e\u043c\u043f\u0430\u043d|customer|supplier|counterparty|company|vendor|client|\b[a-z]{2,}\b)/iu.test(compactWhitespace(repairedInputMessage.toLowerCase()));
+    const llmCanonicalAppliedSignal = Boolean(llmPreDecomposeMeta?.applied) && llmContractMode !== "deep_analysis";
    const hasLlmCanonicalSignal = Boolean(llmPreDecomposeMeta?.llmCanonicalCandidateDetected) &&
-        llmContractMode === "address_query" &&
-        llmContractModeConfidence !== "low" &&
-        llmContractIntent !== null &&
-        llmContractIntent !== "unknown" &&
-        llmContractIntentConfidence !== "low";
+        ((llmContractMode === "address_query" && llmContractModeConfidence !== "low") ||
+            (llmCanonicalAppliedSignal &&
+                (hasStrongDataIntentSignal(repairedInputMessage) || llmCanonicalEntitySignal || llmContractMode === "unsupported")));
    const hasLlmCanonicalDataSignal = Boolean(llmPreDecomposeMeta?.llmCanonicalCandidateDetected) &&
        Boolean(llmPreDecomposeMeta?.applied) &&
-        llmContractMode === "address_query" &&
+        (llmContractMode === "address_query" || llmContractMode === "unsupported" || llmContractMode === null) &&
        hasStrongDataIntentSignal(repairedInputMessage);
+    const sameDateAccountFollowupSignal = hasSameDateAccountFollowupSignalForPredecompose(rawMessageForGate) ||
+        hasSameDateAccountFollowupSignalForPredecompose(repairedInputMessage);
    const hasLexicalAddressSignal = isAddressLlmPreDecomposeCandidate(addressInputMessage) ||
        isAddressLlmPreDecomposeCandidate(repairedInputMessage) ||
        hasAccountingSignal(addressInputMessage) ||
-        hasAccountingSignal(repairedInputMessage);
+        hasAccountingSignal(repairedInputMessage) ||
+        sameDateAccountFollowupSignal;
    const hasUnsupportedLowConfidencePredecomposeSignal = llmContractMode === "unsupported" &&
        (llmContractModeConfidence === "low" || llmContractModeConfidence === "medium") &&
        llmContractIntent === "unknown";
@ -3036,6 +3050,89 @@ function resolveAddressToolGateDecision(addressInputMessage, followupContext, ll
        reason: "no_address_signal_after_l0"
    };
 }
+function hasLooseAllTimeAddressLookupSignal(text) {
+    const repaired = repairAddressMojibake(String(text ?? ""));
+    const normalized = compactWhitespace(repaired.toLowerCase());
+    if (!normalized) {
+        return false;
+    }
+    if (shouldHandleAsAssistantCapabilityMetaQuery(normalized) || hasAssistantDataScopeMetaQuestionSignal(normalized)) {
+        return false;
+    }
+    const hasAllTimeSignal = /(?:\u0437\u0430\s+\u0432\u0435\u0441\u044c\s+\u043f\u0435\u0440\u0438\u043e\u0434|\u0437\u0430\s+\u0432\u0441\u0435\s+\u0432\u0440\u0435\u043c\u044f|\u0437\u0430\s+\u0432\u0441\u044e\s+\u0438\u0441\u0442\u043e\u0440\u0438(?:\u044e|\u0438)|for\s+all\s+time|all\s+time|entire\s+period|full\s+period)/iu.test(normalized);
+    if (!hasAllTimeSignal) {
+        return false;
+    }
+    return /(?:\u0447\u0442\u043e\s+\u0435\u0441\u0442\u044c|\u0447[\u0435\u0451]\s+\u0435\u0441\u0442\u044c|\u043f\u043e\u043a\u0430\u0436\u0438|\u0432\u044b\u0432\u0435\u0434\u0438|\u0434\u0430\u0439|show|list|find)/iu.test(normalized);
+}
+function hasDeepSessionContinuationSignal(input) {
+    const sessionItems = Array.isArray(input?.sessionItems) ? input.sessionItems : [];
+    if (sessionItems.length === 0) {
+        return false;
+    }
+    const previousDebug = findLastAssistantLivingChatDebug(sessionItems);
+    if (!previousDebug || typeof previousDebug !== "object") {
+        return false;
+    }
+    const investigationState = previousDebug.investigation_state_snapshot;
+    if (!investigationState || typeof investigationState !== "object") {
+        return false;
+    }
+    const candidateTexts = [
+        input?.rawUserMessage,
+        input?.repairedRawUserMessage,
+        input?.effectiveAddressUserMessage,
+        input?.repairedEffectiveAddressUserMessage
+    ]
+        .map((value) => compactWhitespace(repairAddressMojibake(String(value ?? "")).toLowerCase()))
+        .filter((value) => value.length > 0);
+    if (candidateTexts.length === 0) {
+        return false;
+    }
+    return candidateTexts.some((text) => {
+        const hasContinuationCue = /^(?:\u0438|\u0430|\u0442\u0430\u043a\u0436\u0435|\u0435\u0449[\u0435\u0451]|\u0434\u043e\u0431\u0430\u0432\u044c|\u0434\u043e\u043f\u043e\u043b\u043d\u0438|\u0443\u0442\u043e\u0447\u043d\u0438|\u043f\u0440\u043e\u0434\u043e\u043b\u0436\u0438|\u0442\u0435\u043f\u0435\u0440\u044c|then|also|and)\b/iu.test(text) ||
+            /(?:\u043f\u043e\s+\u0442\u043e\u043c\u0443\s+\u0436\u0435|\u043f\u043e\s+\u044d\u0442\u043e\u043c\u0443|\u0432\s+\u044d\u0442\u043e\u043c\s+\u0436\u0435|\u0438\s+\u043f\u043e\s+\u043f\u0435\u0440\u0438\u043e\u0434\u0443|\u0434\u043e\u0431\u0430\u0432\u044c\s+\u0443\u0442\u043e\u0447\u043d\u0435\u043d\u0438\u0435|\u043f\u0440\u043e\u0434\u043e\u043b\u0436\u0438\u043c|\u0430\s+\u0435\u0441\u043b\u0438|\u0435\u0441\u043b\u0438\s+\u0442\u043e\u043b\u044c\u043a\u043e|\u043e\u0442\u0434\u0435\u043b\u044c\u043d\u043e)/iu.test(text);
+        const hasAccountOrPeriodCue = /(?:\u0441\u0447[\u0435\u0451]\u0442|account|\b\d{2}(?:[.,]\d{1,2})?\b|\b20\d{2}(?:[-/.]\d{1,2})?\b|\u043f\u0435\u0440\u0438\u043e\u0434|\u043c\u0435\u0441\u044f\u0446)/iu.test(text);
+        const hasDeepRebindCue = /(?:\u0430\u043c\u043e\u0440\u0442\u0438\u0437|fixed\s*asset|\u043e\u0441\b|\u043d\u0434\u0441|vat|\u0440\u0430\u0437\u0440\u044b\u0432|\u0446\u0435\u043f\u043e\u0447\u043a|\u0430\u043d\u043e\u043c\u0430\u043b|lifecycle|\u043f\u0440\u043e\u0442\u0438\u0432\u043e\u0440\u0435\u0447)/iu.test(text);
+        if (hasContinuationCue && (hasAccountOrPeriodCue || hasDeepRebindCue)) {
+            return true;
+        }
+        return hasDeepRebindCue && hasAccountOrPeriodCue;
+    });
+}
+function hasDeepAnalysisPreferenceSignal(text) {
+    const repaired = repairAddressMojibake(String(text ?? ""));
+    const lower = compactWhitespace(repaired.toLowerCase());
+    if (!lower) {
+        return false;
+    }
+    const riskOrAnomalySignal = /(?:\u0440\u0438\u0441\u043a|risk|\u0430\u043d\u043e\u043c\u0430\u043b|anomal|\u043f\u0440\u043e\u0442\u0438\u0432\u043e\u0440\u0435\u0447|\u043a\u043e\u043d\u0444\u043b\u0438\u043a\u0442|conflict|deviation|\u043e\u0442\u043a\u043b\u043e\u043d\u0435\u043d|\u043d\u0435\u0441\u044b\u043a\u043e\u0432\u043a|\u043d\u0435\u0441\u0445\u043e\u0434|\u043e\u0448\u0438\u0431|error|issue|\u043f\u0440\u043e\u0431\u043b\u0435\u043c)/iu.test(lower);
+    const chainSignal = /(?:\u0446\u0435\u043f\u043e\u0447\u043a|chain|trace\s*chain|lifecycle|\u0436\u0438\u0437\u043d\u0435\u043d\u043d[\u0430-\u044f]+\s+\u0446\u0438\u043a\u043b|state\s+transition|\u0440\u0430\u0437\u0440\u044b\u0432[\u0430-\u044f]*)/iu.test(lower);
+    const diagnosticsSignal = /(?:\u0440\u0430\u0437\u043b\u043e\u0436\u0438|\u0434\u0435\u043a\u043e\u043c\u043f\u043e\u0437|\u0440\u0430\u0437\u0431\u0435\u0440\u0438|\u043f\u043e\u0447\u0435\u043c\u0443|why|\u043a\u043e\u0440\u043d\u0435\u0432[\u0430-\u044f]+\s+\u043f\u0440\u0438\u0447\u0438\u043d|root\s*cause|\u043c\u0435\u0445\u0430\u043d\u0438\u0437\u043c[\u0430-\u044f]*|\u0433\u0434\u0435\s+\u0440\u0430\u0437\u0440\u044b\u0432|\u0447\u0442\u043e\s+\u043c\u0435\u0448\u0430[\u0430-\u044f]+\s+\u0437\u0430\u043a\u0440\u044b\u0442)/iu.test(lower);
+    const closureSignal = /(?:\u0437\u0430\u043a\u0440\u044b\u0442\u0438[\u0435\u044f]\s+\u043f\u0435\u0440\u0438\u043e\u0434|period\s*close|\u043d\u0435\s+\u0437\u0430\u043a\u0440\u044b\u043b[\u0430-\u044f]*|\u0445\u0432\u043e\u0441\u0442[\u0430-\u044f]*)/iu.test(lower);
+    const closureIntentSignal = /(?:\u0437\u0430\u043a\u0440\u044b\u0442[\u0430-\u044f]*|period\s*close|close\s+period)/iu.test(lower);
+    const closureDiagnosticPhraseSignal = /(?:\u0447\u0442\u043e(?:\s+\S+){0,8}\s+\u043c\u0435\u0448\u0430[\u0430-\u044f]+\s+\u0437\u0430\u043a\u0440\u044b\u0442)/iu.test(lower);
+    const signalVsNoiseDiagnostic = /(?:\u043d\u0435\s+\u043f\u0440\u043e\u0441\u0442\u043e\s+(?:\u043d\u0430\s+)?\u0448\u0443\u043c|\u043f\u043e\u0445\u043e\u0436[\u0438\u0435]\s+(?:\u0438\u043c\u0435\u043d\u043d\u043e\s+)?\u043d\u0430\s+\u043f\u0440\u043e\u0431\u043b\u0435\u043c)/iu.test(lower);
+    const lifecycleMismatchSignal = /(?:\u043d\u0435\s+\u0442\u0435\u043c\s+\u0442\u0438\u043f(?:\u043e\u043c)?\s+\u0434\u043e\u043a\u0443\u043c\u0435\u043d\u0442|\u043e\u0436\u0438\u0434\u0430\u0435\u043c[\u0430-\u044f]+\s+\u043f\u0435\u0440\u0435\u0445\u043e\u0434[\u0430-\u044f]*\s+\u043d\u0435\s+\u043f\u043e\u0434\u0442\u0432\u0435\u0440\u0436\u0434|\u043f\u0435\u0440\u0435\u0445\u043e\u0434[\u0430-\u044f]*\s+\u043d\u0435\s+\u043f\u043e\u0434\u0442\u0432\u0435\u0440\u0436\u0434|wrong\s+closing\s+document|expected\s+transition)/iu.test(lower);
+    const lifecycleTransitionGapSignal = /(?:\u043e\u0436\u0438\u0434\u0430\u0435\u043c[\u0430-\u044f]+\s+\u043f\u0435\u0440\u0435\u0445\u043e\u0434[\u0430-\u044f]*\s+\u043e\u0442\u0441\u0443\u0442\u0441\u0442\u0432|\u043f\u0435\u0440\u0435\u0445\u043e\u0434[\u0430-\u044f]*\s+\u043e\u0442\u0441\u0443\u0442\u0441\u0442\u0432|\u0441\u0442\u0430\u0434\u0438[\u0438\u044f\u0435]\s+.*\u043f\u0440\u043e\u0439\u0434\u0435\u043d.*\u043f\u0435\u0440\u0435\u0445\u043e\u0434)/iu.test(lower);
+    const expectedActualMismatchSignal = /(?:\u0444\u0430\u043a\u0442\u0438\u0447\u0435\u0441\u043a[\u0430-\u044f]+\s+\u0441\u043e\u0441\u0442\u043e\u044f\u043d[\u0438\u0435\u044f]+\s+.*\u0440\u0430\u0441\u0445\u043e\u0434[\u0430-\u044f]*\s+\u0441\s+\u043e\u0436\u0438\u0434\u0430\u0435\u043c|\u043e\u0436\u0438\u0434\u0430\u0435\u043c[\u0430-\u044f]+\s+\u043f\u043e\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044c\u043d[\u0430-\u044f]*\s+\u0441\u043f\u0438\u0441\u0430\u043d)/iu.test(lower);
+    return riskOrAnomalySignal ||
+        lifecycleMismatchSignal ||
+        (chainSignal && lifecycleTransitionGapSignal) ||
+        expectedActualMismatchSignal ||
+        (chainSignal && diagnosticsSignal) ||
+        (riskOrAnomalySignal && (chainSignal || closureSignal || diagnosticsSignal || closureIntentSignal)) ||
+        (diagnosticsSignal && closureIntentSignal) ||
+        closureDiagnosticPhraseSignal ||
+        signalVsNoiseDiagnostic;
+}
+function hasDirectDeepAnalysisSignal(text) {
+    const normalized = compactWhitespace(repairAddressMojibake(String(text ?? "")).toLowerCase());
+    if (!normalized) {
+        return false;
+    }
+    return /(?:\u0440\u0430\u0437\u043b\u043e\u0436|\u0446\u0435\u043f\u043e\u0447|lifecycle|\u0440\u0430\u0437\u0440\u044b\u0432|\u043f\u0440\u043e\u0442\u0438\u0432\u043e\u0440\u0435\u0447|\u0430\u043d\u043e\u043c\u0430\u043b|\u043f\u043e\u0447\u0435\u043c\u0443|\u043c\u0435\u0445\u0430\u043d\u0438\u0437\u043c|\u0437\u0430\u043a\u0440\u044b\u0442[\u0430-\u044f]*|state\s+transition|root\s*cause|trace\s*chain)/iu.test(normalized);
+}
 export function resolveAssistantOrchestrationDecision(input) {
    const rawUserMessage = String(input?.rawUserMessage ?? input?.userMessage ?? "");
    const effectiveAddressUserMessage = String(input?.effectiveAddressUserMessage ?? rawUserMessage);
@ -3044,6 +3141,7 @@ export function resolveAssistantOrchestrationDecision(input) {
    const followupContext = input?.followupContext ?? null;
    const llmPreDecomposeMeta = input?.llmPreDecomposeMeta ?? null;
    const useMock = Boolean(input?.useMock);
+    const sessionItems = Array.isArray(input?.sessionItems) ? input.sessionItems : null;
    const dataScopeMetaQuery = hasAssistantDataScopeMetaQuestionSignal(rawUserMessage) ||
        hasAssistantDataScopeMetaQuestionSignal(repairedRawUserMessage) ||
        hasAssistantDataScopeMetaQuestionSignal(effectiveAddressUserMessage) ||
@ -3131,11 +3229,52 @@ export function resolveAssistantOrchestrationDecision(input) {
        };
    }
    const baseToolGate = resolveAddressToolGateDecision(effectiveAddressUserMessage, followupContext, llmPreDecomposeMeta, rawUserMessage);
+    const llmContractMode = toNonEmptyString(llmPreDecomposeMeta?.predecomposeContract?.mode);
+    const preserveAddressLaneSignal = Boolean((llmPreDecomposeMeta?.llmCanonicalCandidateDetected &&
+        llmPreDecomposeMeta?.applied &&
+        (llmContractMode === "address_query" || llmContractMode === "unsupported")) ||
+        hasSameDateAccountFollowupSignalForPredecompose(rawUserMessage) ||
+        hasSameDateAccountFollowupSignalForPredecompose(effectiveAddressUserMessage) ||
+        hasSameDateAccountFollowupSignalForPredecompose(repairedRawUserMessage) ||
+        hasSameDateAccountFollowupSignalForPredecompose(repairedEffectiveAddressUserMessage) ||
+        hasLooseAllTimeAddressLookupSignal(rawUserMessage) ||
+        hasLooseAllTimeAddressLookupSignal(effectiveAddressUserMessage) ||
+        hasLooseAllTimeAddressLookupSignal(repairedRawUserMessage) ||
+        hasLooseAllTimeAddressLookupSignal(repairedEffectiveAddressUserMessage) ||
+        hasAddressFollowupContextSignal(rawUserMessage) ||
+        hasAddressFollowupContextSignal(effectiveAddressUserMessage) ||
+        hasAddressFollowupContextSignal(repairedRawUserMessage) ||
+        hasAddressFollowupContextSignal(repairedEffectiveAddressUserMessage));
    const unsupportedAddressIntentFallbackToDeep = Boolean(!followupContext &&
        baseToolGate?.runAddressLane &&
        modeDetection.mode !== "address_query" &&
        intentResolution.intent === "unknown" &&
-        strongDataSignal);
+        strongDataSignal &&
+        !preserveAddressLaneSignal);
+    const deepAnalysisPreferenceDetected = Boolean(hasDeepAnalysisPreferenceSignal(rawUserMessage) ||
+        hasDeepAnalysisPreferenceSignal(repairedRawUserMessage) ||
+        hasDeepAnalysisPreferenceSignal(effectiveAddressUserMessage) ||
+        hasDeepAnalysisPreferenceSignal(repairedEffectiveAddressUserMessage) ||
+        hasDirectDeepAnalysisSignal(rawUserMessage) ||
+        hasDirectDeepAnalysisSignal(repairedRawUserMessage) ||
+        hasDirectDeepAnalysisSignal(effectiveAddressUserMessage) ||
+        hasDirectDeepAnalysisSignal(repairedEffectiveAddressUserMessage));
+    const vatExplainFollowupSignal = Boolean(followupContext &&
+        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 &&
+        !vatExplainFollowupSignal &&
+        (!followupContext || !dataRetrievalSignal));
+    const deepSessionContinuationFallbackToDeep = Boolean(!followupContext &&
+        baseToolGate?.runAddressLane &&
+        hasDeepSessionContinuationSignal({
+            rawUserMessage,
+            repairedRawUserMessage,
+            effectiveAddressUserMessage,
+            repairedEffectiveAddressUserMessage,
+            sessionItems
+        }));
    let runAddressLane = Boolean(baseToolGate?.runAddressLane);
    let toolGateDecision = String(baseToolGate?.decision ?? "skip_address_lane");
    let toolGateReason = String(baseToolGate?.reason ?? "no_address_signal_after_l0");
@ -3144,6 +3283,16 @@ export function resolveAssistantOrchestrationDecision(input) {
        toolGateDecision = "skip_address_lane";
        toolGateReason = "address_signal_unsupported_intent_fallback_to_deep";
    }
+    if (deepAnalysisSignalFallbackToDeep && !unsupportedAddressIntentFallbackToDeep) {
+        runAddressLane = false;
+        toolGateDecision = "skip_address_lane";
+        toolGateReason = "deep_analysis_signal_fallback_to_deep";
+    }
+    if (deepSessionContinuationFallbackToDeep) {
+        runAddressLane = false;
+        toolGateDecision = "skip_address_lane";
+        toolGateReason = "deep_session_continuation_fallback_to_deep";
+    }
    let livingDecision = resolveLivingAssistantModeDecision({
        userMessage: rawUserMessage,
        addressLaneTriggered: runAddressLane,
@ -3157,6 +3306,18 @@ export function resolveAssistantOrchestrationDecision(input) {
            reason: "unsupported_address_intent_fallback_to_deep"
        };
    }
+    if (deepAnalysisSignalFallbackToDeep && !unsupportedAddressIntentFallbackToDeep) {
+        livingDecision = {
+            mode: "deep_analysis",
+            reason: "deep_analysis_signal_fallback_to_deep"
+        };
+    }
+    if (deepSessionContinuationFallbackToDeep) {
+        livingDecision = {
+            mode: "deep_analysis",
+            reason: "deep_session_continuation_fallback_to_deep"
+        };
+    }
    return {
        runAddressLane,
        toolGateDecision,
@ -3174,6 +3335,8 @@ export function resolveAssistantOrchestrationDecision(input) {
            data_retrieval_signal_detected: dataRetrievalSignal,
            followup_context_detected: Boolean(followupContext),
            unsupported_address_intent_fallback_to_deep: unsupportedAddressIntentFallbackToDeep,
+            deep_analysis_signal_fallback_to_deep: deepAnalysisSignalFallbackToDeep,
+            deep_session_continuation_fallback_to_deep: deepSessionContinuationFallbackToDeep,
            final_decision: {
                run_address_lane: runAddressLane,
                tool_gate_decision: toolGateDecision,
--- a/llm_normalizer/backend/src/services/assistantTurnRuntimeDepsAdapter.ts
+++ b/llm_normalizer/backend/src/services/assistantTurnRuntimeDepsAdapter.ts
@ -85,12 +85,12 @@ export function buildAssistantTurnRuntimeDeps(
 ): AssistantTurnRuntimeBuilderDeps {
  return {
    ...input.helpers,
-    ensureSession: input.sessions.ensureSession,
-    appendItem: input.sessions.appendItem,
-    getSession: input.sessions.getSession,
-    persistSession: input.sessionLogger.persistSession,
-    setInvestigationState: input.sessions.setInvestigationState,
-    normalize: input.normalizerService.normalize,
+    ensureSession: (sessionId) => input.sessions.ensureSession(sessionId),
+    appendItem: (sessionId, item) => input.sessions.appendItem(sessionId, item),
+    getSession: (sessionId) => input.sessions.getSession(sessionId),
+    persistSession: (session) => input.sessionLogger.persistSession(session),
+    setInvestigationState: (sessionId, state) => input.sessions.setInvestigationState(sessionId, state),
+    normalize: (payload) => input.normalizerService.normalize(payload),
    executeRouteRuntime: (route, fragmentText, options) =>
      input.dataLayer.executeRouteRuntime(route, fragmentText, options),
    tryAddressQueryHandle: (messageUsed, options) => input.addressQueryService.tryHandle(messageUsed, options),
--- a/llm_normalizer/backend/src/types/assistant.ts
+++ b/llm_normalizer/backend/src/types/assistant.ts
@ -1,4 +1,13 @@
-import type { NormalizeRequestPayload, NormalizeResponsePayload, RouteHintSummary } from "./normalizer";
+import type {
+  NormalizeRequestPayload,
+  NormalizeResponsePayload,
+  NoRouteReason,
+  RouteHintSummary,
+  RouteStatus,
+  ExecutionReadiness,
+  ConfidenceLevel,
+  IntentClass
+} from "./normalizer";
 import type { AnswerStructureV11, EvidenceItem } from "./stage1Contracts";
 import type {
  CandidateEvidenceItem,
@ -27,6 +36,43 @@ export type AssistantProblemAnswerMode =
  | "stage2_problem_centric_v1"
  | "stage3_lifecycle_aware_v1";

+export interface AssistantExecutionStateRecord {
+  fragment_id: string | null;
+  execution_readiness: ExecutionReadiness | null;
+  route_status: RouteStatus | null;
+  no_route_reason: NoRouteReason | null;
+}
+
+export type AssistantDebugRouteRecord =
+  | {
+      fragment_id: string;
+      route: string;
+      reason: string;
+      confidence: ConfidenceLevel;
+      intent_class: IntentClass;
+    }
+  | {
+      fragment_id: string;
+      route: string;
+      reason: string;
+      route_status: RouteStatus | null;
+      no_route_reason: NoRouteReason | null;
+      clarification_reason: string | null;
+      execution_readiness: ExecutionReadiness | null;
+    };
+
+export interface AssistantAddressRuntimeMetaForDeep {
+  attempted?: boolean;
+  applied?: boolean;
+  reason?: string | null;
+  provider?: string | null;
+  fallbackRuleHit?: string | null;
+  toolGateDecision?: string | null;
+  toolGateReason?: string | null;
+  predecomposeContract?: Record<string, unknown> | null;
+  orchestrationContract?: Record<string, unknown> | null;
+}
+
 export interface AssistantRequirement {
  requirement_id: string;
  source_fragment_id: string | null;
@ -302,7 +348,7 @@ export interface AssistantDebugPayload {
  fragments: unknown[];
  requirements_extracted: AssistantRequirement[];
  coverage_report: RequirementCoverageReport;
-  routes: Array<Record<string, unknown>>;
+  routes: AssistantDebugRouteRecord[];
  retrieval_status: Array<{
    fragment_id: string;
    requirement_ids: string[];
--- a/llm_normalizer/backend/src/utils/log.ts
+++ b/llm_normalizer/backend/src/utils/log.ts
@ -32,6 +32,9 @@ function redactObject(value: unknown): unknown {
 }

 export function logJson(entry: JsonLogEntry): void {
+  if (process.env.NODE_ENV === "test" && process.env.FEATURE_JSON_STDOUT_LOGS_IN_TESTS !== "1") {
+    return;
+  }
  const safe = {
    ...entry,
    details: redactObject(entry.details)
--- a/llm_normalizer/backend/tests/assistantLivingRouter.test.ts
+++ b/llm_normalizer/backend/tests/assistantLivingRouter.test.ts
@ -261,4 +261,40 @@ describe("assistant orchestration contract", () => {
      decision.toolGateReason
    );
  });
+
+  it("routes risk/anomaly analytics wording to deep pipeline", () => {
+    const decision = resolveAssistantOrchestrationDecision({
+      rawUserMessage: "Проверь НДС по счету 19 за 2020-06 и рискованные записи по документам.",
+      effectiveAddressUserMessage: "Проверь НДС по счету 19 за 2020-06 и рискованные записи по документам.",
+      followupContext: null,
+      llmPreDecomposeMeta: null,
+      useMock: false
+    });
+
+    expect(decision.runAddressLane).toBe(false);
+    expect(decision.toolGateDecision).toBe("skip_address_lane");
+    expect(decision.toolGateReason).toBe("deep_analysis_signal_fallback_to_deep");
+    expect(decision.livingMode).toBe("deep_analysis");
+    expect(decision.livingReason).toBe("deep_analysis_signal_fallback_to_deep");
+    expect(decision.orchestrationContract?.deep_analysis_signal_fallback_to_deep).toBe(true);
+  });
+
+  it("routes settlement closure verification wording to deep pipeline", () => {
+    const decision = resolveAssistantOrchestrationDecision({
+      rawUserMessage:
+        "По оплате поставщику на счете 60 в июле 2020 остался хвост. Проверь закрытие по договору и объекту расчетов.",
+      effectiveAddressUserMessage:
+        "По оплате поставщику на счете 60 в июле 2020 остался хвост. Проверь закрытие по договору и объекту расчетов.",
+      followupContext: null,
+      llmPreDecomposeMeta: null,
+      useMock: false
+    });
+
+    expect(decision.runAddressLane).toBe(false);
+    expect(decision.toolGateDecision).toBe("skip_address_lane");
+    expect(decision.toolGateReason).toBe("deep_analysis_signal_fallback_to_deep");
+    expect(decision.livingMode).toBe("deep_analysis");
+    expect(decision.livingReason).toBe("deep_analysis_signal_fallback_to_deep");
+    expect(decision.orchestrationContract?.deep_analysis_signal_fallback_to_deep).toBe(true);
+  });
 });
--- a/llm_normalizer/backend/tests/assistantTurnRuntimeDepsAdapter.test.ts
+++ b/llm_normalizer/backend/tests/assistantTurnRuntimeDepsAdapter.test.ts
@ -129,4 +129,89 @@ describe("assistant turn runtime deps adapter", () => {
    expect(deps.compactWhitespace("  value  ")).toBe("value");
    expect(helperCompactWhitespace).toHaveBeenCalledWith("  value  ");
  });
+
+  it("preserves method context for stateful service instances", async () => {
+    class SessionStoreLike {
+      public calls: string[] = [];
+      public ensureSession(sessionId: string) {
+        this.calls.push(`ensure:${sessionId}`);
+        return { session_id: sessionId } as any;
+      }
+      public appendItem(sessionId: string) {
+        this.calls.push(`append:${sessionId}`);
+      }
+      public getSession(sessionId: string) {
+        this.calls.push(`get:${sessionId}`);
+        return null;
+      }
+      public setInvestigationState(sessionId: string) {
+        this.calls.push(`state:${sessionId}`);
+        return null;
+      }
+    }
+
+    class SessionLoggerLike {
+      public persisted = 0;
+      public persistSession() {
+        this.persisted += 1;
+      }
+    }
+
+    class NormalizerLike {
+      public normalized = 0;
+      public async normalize() {
+        this.normalized += 1;
+        return {};
+      }
+    }
+
+    const sessions = new SessionStoreLike();
+    const sessionLogger = new SessionLoggerLike();
+    const normalizerService = new NormalizerLike();
+
+    const deps = buildAssistantTurnRuntimeDeps({
+      sessions,
+      sessionLogger,
+      normalizerService,
+      dataLayer: {
+        executeRouteRuntime: async () => ({})
+      },
+      addressQueryService: {
+        tryHandle: async () => null
+      },
+      chatClient: {},
+      messageIdFactory: () => "msg-ctx",
+      nowIso: () => "2026-04-11T00:00:00.000Z",
+      defaultApiKey: "",
+      logEvent: () => {},
+      flags: {
+        featureAssistantAddressQueryV1: false,
+        featureAddressLlmPredecomposeV1: false,
+        featureInvestigationStateV1: false,
+        featureStateFollowupBindingV1: false,
+        featureContractsV11: false,
+        featureAnswerPolicyV11: false,
+        featureProblemCentricAnswerV1: false,
+        featureLifecycleAnswerV1: false
+      },
+      defaults: {
+        defaultModel: "model",
+        defaultBaseUrl: "base"
+      },
+      helpers: {
+        compactWhitespace: (value: unknown) => String(value ?? "")
+      } as any
+    });
+
+    deps.ensureSession("asst-ctx");
+    deps.appendItem("asst-ctx", { role: "user" } as any);
+    deps.getSession("asst-ctx");
+    deps.setInvestigationState("asst-ctx", null);
+    deps.persistSession({ session_id: "asst-ctx" } as any);
+    await deps.normalize({ user_message: "ctx" } as any);
+
+    expect(sessions.calls).toEqual(["ensure:asst-ctx", "append:asst-ctx", "get:asst-ctx", "state:asst-ctx"]);
+    expect(sessionLogger.persisted).toBe(1);
+    expect(normalizerService.normalized).toBe(1);
+  });
 });
--- a/llm_normalizer/backend/vitest-full-report.json
+++ b/llm_normalizer/backend/vitest-full-report.json
--- a/llm_normalizer/data/eval_cases/eval-5S6BYeWJWv.report.json
+++ b/llm_normalizer/data/eval_cases/eval-5S6BYeWJWv.report.json
@ -0,0 +1,112 @@
+{
+  "run_id": "eval-5S6BYeWJWv",
+  "timestamp": "2026-04-11T10:38:19.513Z",
+  "mode": "single-pass-strict",
+  "use_mock": true,
+  "prompt_version": "normalizer_v2_0_2",
+  "schema_version": "v2_0_2",
+  "dataset": {
+    "source": "inline_raw_questions",
+    "file": null,
+    "raw_questions_count": 2
+  },
+  "cases_total": 2,
+  "metrics": {
+    "schema_validation_pass_rate": 100,
+    "scope_detection_accuracy": null,
+    "scope_in_scope_rate": 100,
+    "multi_intent_detected_rate": 0,
+    "clarification_required_rate": 0,
+    "avg_fragments_per_message": 1,
+    "out_of_scope_fragment_rate": 0,
+    "routed_fragment_rate": 100,
+    "no_route_fragment_rate": 0,
+    "route_resolution_accuracy": null,
+    "no_route_precision": null,
+    "false_no_route_rate": null,
+    "execution_state_consistency_rate": 100,
+    "executable_with_soft_assumptions_rate": 100,
+    "soft_assumption_used_fragment_rate": 100,
+    "clarification_precision": null,
+    "clarification_recall": null,
+    "false_clarification_rate": null
+  },
+  "budget": {
+    "requests_total": 0,
+    "retries_used": 0
+  },
+  "clarification_eval": {
+    "labeled_cases": 0,
+    "true_positive": 0,
+    "false_positive": 0,
+    "false_negative": 0
+  },
+  "route_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0,
+    "expected_routed_cases": 0,
+    "no_route_true_positive": 0,
+    "no_route_false_positive": 0
+  },
+  "scope_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0
+  },
+  "execution_state_eval": {
+    "checks_total": 2,
+    "checks_passed": 2
+  },
+  "route_distribution": {
+    "store_feature_risk": 1,
+    "hybrid_store_plus_live": 1
+  },
+  "fallback_distribution": {
+    "none": 2
+  },
+  "results": [
+    {
+      "case_id": "BQ-001",
+      "raw_question": "Проверь счет 60 за июнь 2020",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "bi53megxPkVDpq",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-002",
+      "raw_question": "Покажи риски по НДС и по закрытию",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "ew725J5aRLs1XU",
+      "request_count_for_case": 0
+    }
+  ]
+}
--- a/llm_normalizer/data/eval_cases/eval-D1gt1OyBqh.report.json
+++ b/llm_normalizer/data/eval_cases/eval-D1gt1OyBqh.report.json
@ -0,0 +1,112 @@
+{
+  "run_id": "eval-D1gt1OyBqh",
+  "timestamp": "2026-04-11T10:35:37.557Z",
+  "mode": "single-pass-strict",
+  "use_mock": true,
+  "prompt_version": "normalizer_v2_0_2",
+  "schema_version": "v2_0_2",
+  "dataset": {
+    "source": "inline_raw_questions",
+    "file": null,
+    "raw_questions_count": 2
+  },
+  "cases_total": 2,
+  "metrics": {
+    "schema_validation_pass_rate": 100,
+    "scope_detection_accuracy": null,
+    "scope_in_scope_rate": 100,
+    "multi_intent_detected_rate": 0,
+    "clarification_required_rate": 0,
+    "avg_fragments_per_message": 1,
+    "out_of_scope_fragment_rate": 0,
+    "routed_fragment_rate": 100,
+    "no_route_fragment_rate": 0,
+    "route_resolution_accuracy": null,
+    "no_route_precision": null,
+    "false_no_route_rate": null,
+    "execution_state_consistency_rate": 100,
+    "executable_with_soft_assumptions_rate": 100,
+    "soft_assumption_used_fragment_rate": 100,
+    "clarification_precision": null,
+    "clarification_recall": null,
+    "false_clarification_rate": null
+  },
+  "budget": {
+    "requests_total": 0,
+    "retries_used": 0
+  },
+  "clarification_eval": {
+    "labeled_cases": 0,
+    "true_positive": 0,
+    "false_positive": 0,
+    "false_negative": 0
+  },
+  "route_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0,
+    "expected_routed_cases": 0,
+    "no_route_true_positive": 0,
+    "no_route_false_positive": 0
+  },
+  "scope_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0
+  },
+  "execution_state_eval": {
+    "checks_total": 2,
+    "checks_passed": 2
+  },
+  "route_distribution": {
+    "store_feature_risk": 1,
+    "hybrid_store_plus_live": 1
+  },
+  "fallback_distribution": {
+    "none": 2
+  },
+  "results": [
+    {
+      "case_id": "BQ-001",
+      "raw_question": "Проверь счет 60 за июнь 2020",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "0XdpLLt2DVEC8b",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-002",
+      "raw_question": "Покажи риски по НДС и по закрытию",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "TGXTmJEIbfRfxl",
+      "request_count_for_case": 0
+    }
+  ]
+}
--- a/llm_normalizer/data/eval_cases/eval-EYDYh4k78m.report.json
+++ b/llm_normalizer/data/eval_cases/eval-EYDYh4k78m.report.json
@ -0,0 +1,137 @@
+{
+  "run_id": "eval-EYDYh4k78m",
+  "timestamp": "2026-04-11T10:35:23.070Z",
+  "mode": "single-pass-strict",
+  "use_mock": true,
+  "prompt_version": "normalizer_v2_0_2",
+  "schema_version": "v2_0_2",
+  "dataset": {
+    "source": "inline_raw_questions",
+    "file": null,
+    "raw_questions_count": 3
+  },
+  "cases_total": 3,
+  "metrics": {
+    "schema_validation_pass_rate": 100,
+    "scope_detection_accuracy": null,
+    "scope_in_scope_rate": 33.33,
+    "multi_intent_detected_rate": 0,
+    "clarification_required_rate": 0,
+    "avg_fragments_per_message": 1,
+    "out_of_scope_fragment_rate": 33.33,
+    "routed_fragment_rate": 66.67,
+    "no_route_fragment_rate": 33.33,
+    "route_resolution_accuracy": null,
+    "no_route_precision": null,
+    "false_no_route_rate": null,
+    "execution_state_consistency_rate": 66.67,
+    "executable_with_soft_assumptions_rate": 100,
+    "soft_assumption_used_fragment_rate": 100,
+    "clarification_precision": null,
+    "clarification_recall": null,
+    "false_clarification_rate": null
+  },
+  "budget": {
+    "requests_total": 0,
+    "retries_used": 0
+  },
+  "clarification_eval": {
+    "labeled_cases": 0,
+    "true_positive": 0,
+    "false_positive": 0,
+    "false_negative": 0
+  },
+  "route_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0,
+    "expected_routed_cases": 0,
+    "no_route_true_positive": 0,
+    "no_route_false_positive": 0
+  },
+  "scope_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0
+  },
+  "execution_state_eval": {
+    "checks_total": 3,
+    "checks_passed": 2
+  },
+  "route_distribution": {
+    "hybrid_store_plus_live": 1,
+    "no_route": 1,
+    "batch_refresh_then_store": 1
+  },
+  "fallback_distribution": {
+    "none": 1,
+    "out_of_scope": 1,
+    "clarification": 1
+  },
+  "results": [
+    {
+      "case_id": "BQ-001",
+      "raw_question": "Проверь хвосты по поставщикам и разложи цепочку",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "XkM4LP-bXu4rf9",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-002",
+      "raw_question": "Как вообще по ФСБУ",
+      "validation_passed": true,
+      "message_in_scope": false,
+      "scope_confidence": "low",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 0,
+      "out_of_scope_fragments": 1,
+      "unclear_fragments": 0,
+      "fallback_type": "out_of_scope",
+      "predicted_route_status": "no_route",
+      "expected_route_status": null,
+      "predicted_no_route_reason": "out_of_scope",
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 0,
+      "trace_id": "eDF6_4JOsguyhk",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-003",
+      "raw_question": "Покажи топ рисков за июнь 2020",
+      "validation_passed": true,
+      "message_in_scope": false,
+      "scope_confidence": "low",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 0,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 1,
+      "fallback_type": "clarification",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 0,
+      "trace_id": "dsCVkLkLYqwqHr",
+      "request_count_for_case": 0
+    }
+  ]
+}
--- a/llm_normalizer/data/eval_cases/eval-NeqPDTflc7.report.json
+++ b/llm_normalizer/data/eval_cases/eval-NeqPDTflc7.report.json
@ -0,0 +1,112 @@
+{
+  "run_id": "eval-NeqPDTflc7",
+  "timestamp": "2026-04-11T10:16:53.909Z",
+  "mode": "single-pass-strict",
+  "use_mock": true,
+  "prompt_version": "normalizer_v2_0_2",
+  "schema_version": "v2_0_2",
+  "dataset": {
+    "source": "inline_raw_questions",
+    "file": null,
+    "raw_questions_count": 2
+  },
+  "cases_total": 2,
+  "metrics": {
+    "schema_validation_pass_rate": 100,
+    "scope_detection_accuracy": null,
+    "scope_in_scope_rate": 100,
+    "multi_intent_detected_rate": 0,
+    "clarification_required_rate": 0,
+    "avg_fragments_per_message": 1,
+    "out_of_scope_fragment_rate": 0,
+    "routed_fragment_rate": 100,
+    "no_route_fragment_rate": 0,
+    "route_resolution_accuracy": null,
+    "no_route_precision": null,
+    "false_no_route_rate": null,
+    "execution_state_consistency_rate": 100,
+    "executable_with_soft_assumptions_rate": 100,
+    "soft_assumption_used_fragment_rate": 100,
+    "clarification_precision": null,
+    "clarification_recall": null,
+    "false_clarification_rate": null
+  },
+  "budget": {
+    "requests_total": 0,
+    "retries_used": 0
+  },
+  "clarification_eval": {
+    "labeled_cases": 0,
+    "true_positive": 0,
+    "false_positive": 0,
+    "false_negative": 0
+  },
+  "route_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0,
+    "expected_routed_cases": 0,
+    "no_route_true_positive": 0,
+    "no_route_false_positive": 0
+  },
+  "scope_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0
+  },
+  "execution_state_eval": {
+    "checks_total": 2,
+    "checks_passed": 2
+  },
+  "route_distribution": {
+    "store_feature_risk": 1,
+    "hybrid_store_plus_live": 1
+  },
+  "fallback_distribution": {
+    "none": 2
+  },
+  "results": [
+    {
+      "case_id": "BQ-001",
+      "raw_question": "Проверь счет 60 за июнь 2020",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "ynDJfmhtaBS6Jh",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-002",
+      "raw_question": "Покажи риски по счету 97",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "BhT9NfO-40oigp",
+      "request_count_for_case": 0
+    }
+  ]
+}
--- a/llm_normalizer/data/eval_cases/eval-OSNnoiu9KQ.report.json
+++ b/llm_normalizer/data/eval_cases/eval-OSNnoiu9KQ.report.json
@ -0,0 +1,137 @@
+{
+  "run_id": "eval-OSNnoiu9KQ",
+  "timestamp": "2026-04-11T10:16:34.741Z",
+  "mode": "single-pass-strict",
+  "use_mock": true,
+  "prompt_version": "normalizer_v2_0_2",
+  "schema_version": "v2_0_2",
+  "dataset": {
+    "source": "inline_raw_questions",
+    "file": null,
+    "raw_questions_count": 3
+  },
+  "cases_total": 3,
+  "metrics": {
+    "schema_validation_pass_rate": 100,
+    "scope_detection_accuracy": null,
+    "scope_in_scope_rate": 33.33,
+    "multi_intent_detected_rate": 0,
+    "clarification_required_rate": 0,
+    "avg_fragments_per_message": 1,
+    "out_of_scope_fragment_rate": 33.33,
+    "routed_fragment_rate": 66.67,
+    "no_route_fragment_rate": 33.33,
+    "route_resolution_accuracy": null,
+    "no_route_precision": null,
+    "false_no_route_rate": null,
+    "execution_state_consistency_rate": 66.67,
+    "executable_with_soft_assumptions_rate": 100,
+    "soft_assumption_used_fragment_rate": 100,
+    "clarification_precision": null,
+    "clarification_recall": null,
+    "false_clarification_rate": null
+  },
+  "budget": {
+    "requests_total": 0,
+    "retries_used": 0
+  },
+  "clarification_eval": {
+    "labeled_cases": 0,
+    "true_positive": 0,
+    "false_positive": 0,
+    "false_negative": 0
+  },
+  "route_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0,
+    "expected_routed_cases": 0,
+    "no_route_true_positive": 0,
+    "no_route_false_positive": 0
+  },
+  "scope_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0
+  },
+  "execution_state_eval": {
+    "checks_total": 3,
+    "checks_passed": 2
+  },
+  "route_distribution": {
+    "hybrid_store_plus_live": 1,
+    "no_route": 1,
+    "batch_refresh_then_store": 1
+  },
+  "fallback_distribution": {
+    "none": 1,
+    "out_of_scope": 1,
+    "clarification": 1
+  },
+  "results": [
+    {
+      "case_id": "BQ-001",
+      "raw_question": "Проверь хвосты по поставщикам и разложи цепочку",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "YPmWhiLGA674vn",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-002",
+      "raw_question": "Как вообще по ФСБУ",
+      "validation_passed": true,
+      "message_in_scope": false,
+      "scope_confidence": "low",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 0,
+      "out_of_scope_fragments": 1,
+      "unclear_fragments": 0,
+      "fallback_type": "out_of_scope",
+      "predicted_route_status": "no_route",
+      "expected_route_status": null,
+      "predicted_no_route_reason": "out_of_scope",
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 0,
+      "trace_id": "JAbihOWvYs9bYR",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-003",
+      "raw_question": "Покажи топ рисков за июнь 2020",
+      "validation_passed": true,
+      "message_in_scope": false,
+      "scope_confidence": "low",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 0,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 1,
+      "fallback_type": "clarification",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 0,
+      "trace_id": "WuI3XbextvaDxw",
+      "request_count_for_case": 0
+    }
+  ]
+}
--- a/llm_normalizer/data/eval_cases/eval-RVeDpGREWm.report.json
+++ b/llm_normalizer/data/eval_cases/eval-RVeDpGREWm.report.json
@ -0,0 +1,112 @@
+{
+  "run_id": "eval-RVeDpGREWm",
+  "timestamp": "2026-04-11T10:14:44.794Z",
+  "mode": "single-pass-strict",
+  "use_mock": true,
+  "prompt_version": "normalizer_v2_0_2",
+  "schema_version": "v2_0_2",
+  "dataset": {
+    "source": "inline_raw_questions",
+    "file": null,
+    "raw_questions_count": 2
+  },
+  "cases_total": 2,
+  "metrics": {
+    "schema_validation_pass_rate": 100,
+    "scope_detection_accuracy": null,
+    "scope_in_scope_rate": 100,
+    "multi_intent_detected_rate": 0,
+    "clarification_required_rate": 0,
+    "avg_fragments_per_message": 1,
+    "out_of_scope_fragment_rate": 0,
+    "routed_fragment_rate": 100,
+    "no_route_fragment_rate": 0,
+    "route_resolution_accuracy": null,
+    "no_route_precision": null,
+    "false_no_route_rate": null,
+    "execution_state_consistency_rate": 100,
+    "executable_with_soft_assumptions_rate": 100,
+    "soft_assumption_used_fragment_rate": 100,
+    "clarification_precision": null,
+    "clarification_recall": null,
+    "false_clarification_rate": null
+  },
+  "budget": {
+    "requests_total": 0,
+    "retries_used": 0
+  },
+  "clarification_eval": {
+    "labeled_cases": 0,
+    "true_positive": 0,
+    "false_positive": 0,
+    "false_negative": 0
+  },
+  "route_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0,
+    "expected_routed_cases": 0,
+    "no_route_true_positive": 0,
+    "no_route_false_positive": 0
+  },
+  "scope_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0
+  },
+  "execution_state_eval": {
+    "checks_total": 2,
+    "checks_passed": 2
+  },
+  "route_distribution": {
+    "store_feature_risk": 1,
+    "hybrid_store_plus_live": 1
+  },
+  "fallback_distribution": {
+    "none": 2
+  },
+  "results": [
+    {
+      "case_id": "BQ-001",
+      "raw_question": "Проверь счет 60 за июнь 2020",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "9WDoQo0uvKHsFh",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-002",
+      "raw_question": "Покажи риски по НДС и по закрытию",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "YJ7f6z9_SS7ZgW",
+      "request_count_for_case": 0
+    }
+  ]
+}
--- a/llm_normalizer/data/eval_cases/eval-hhXvDmFzGJ.report.json
+++ b/llm_normalizer/data/eval_cases/eval-hhXvDmFzGJ.report.json
@ -0,0 +1,137 @@
+{
+  "run_id": "eval-hhXvDmFzGJ",
+  "timestamp": "2026-04-11T10:14:25.501Z",
+  "mode": "single-pass-strict",
+  "use_mock": true,
+  "prompt_version": "normalizer_v2_0_2",
+  "schema_version": "v2_0_2",
+  "dataset": {
+    "source": "inline_raw_questions",
+    "file": null,
+    "raw_questions_count": 3
+  },
+  "cases_total": 3,
+  "metrics": {
+    "schema_validation_pass_rate": 100,
+    "scope_detection_accuracy": null,
+    "scope_in_scope_rate": 33.33,
+    "multi_intent_detected_rate": 0,
+    "clarification_required_rate": 0,
+    "avg_fragments_per_message": 1,
+    "out_of_scope_fragment_rate": 33.33,
+    "routed_fragment_rate": 66.67,
+    "no_route_fragment_rate": 33.33,
+    "route_resolution_accuracy": null,
+    "no_route_precision": null,
+    "false_no_route_rate": null,
+    "execution_state_consistency_rate": 66.67,
+    "executable_with_soft_assumptions_rate": 100,
+    "soft_assumption_used_fragment_rate": 100,
+    "clarification_precision": null,
+    "clarification_recall": null,
+    "false_clarification_rate": null
+  },
+  "budget": {
+    "requests_total": 0,
+    "retries_used": 0
+  },
+  "clarification_eval": {
+    "labeled_cases": 0,
+    "true_positive": 0,
+    "false_positive": 0,
+    "false_negative": 0
+  },
+  "route_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0,
+    "expected_routed_cases": 0,
+    "no_route_true_positive": 0,
+    "no_route_false_positive": 0
+  },
+  "scope_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0
+  },
+  "execution_state_eval": {
+    "checks_total": 3,
+    "checks_passed": 2
+  },
+  "route_distribution": {
+    "hybrid_store_plus_live": 1,
+    "no_route": 1,
+    "batch_refresh_then_store": 1
+  },
+  "fallback_distribution": {
+    "none": 1,
+    "out_of_scope": 1,
+    "clarification": 1
+  },
+  "results": [
+    {
+      "case_id": "BQ-001",
+      "raw_question": "Проверь хвосты по поставщикам и разложи цепочку",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "5-MShsMwdLMm1O",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-002",
+      "raw_question": "Как вообще по ФСБУ",
+      "validation_passed": true,
+      "message_in_scope": false,
+      "scope_confidence": "low",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 0,
+      "out_of_scope_fragments": 1,
+      "unclear_fragments": 0,
+      "fallback_type": "out_of_scope",
+      "predicted_route_status": "no_route",
+      "expected_route_status": null,
+      "predicted_no_route_reason": "out_of_scope",
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 0,
+      "trace_id": "pv-o-6Cock8Ve2",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-003",
+      "raw_question": "Покажи топ рисков за июнь 2020",
+      "validation_passed": true,
+      "message_in_scope": false,
+      "scope_confidence": "low",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 0,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 1,
+      "fallback_type": "clarification",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 0,
+      "trace_id": "K0zAFkzMAu2BqF",
+      "request_count_for_case": 0
+    }
+  ]
+}
--- a/llm_normalizer/data/eval_cases/eval-kBTYkPR4wq.report.json
+++ b/llm_normalizer/data/eval_cases/eval-kBTYkPR4wq.report.json
@ -0,0 +1,112 @@
+{
+  "run_id": "eval-kBTYkPR4wq",
+  "timestamp": "2026-04-11T10:16:53.895Z",
+  "mode": "single-pass-strict",
+  "use_mock": true,
+  "prompt_version": "normalizer_v2_0_2",
+  "schema_version": "v2_0_2",
+  "dataset": {
+    "source": "inline_raw_questions",
+    "file": null,
+    "raw_questions_count": 2
+  },
+  "cases_total": 2,
+  "metrics": {
+    "schema_validation_pass_rate": 100,
+    "scope_detection_accuracy": null,
+    "scope_in_scope_rate": 100,
+    "multi_intent_detected_rate": 0,
+    "clarification_required_rate": 0,
+    "avg_fragments_per_message": 1,
+    "out_of_scope_fragment_rate": 0,
+    "routed_fragment_rate": 100,
+    "no_route_fragment_rate": 0,
+    "route_resolution_accuracy": null,
+    "no_route_precision": null,
+    "false_no_route_rate": null,
+    "execution_state_consistency_rate": 100,
+    "executable_with_soft_assumptions_rate": 100,
+    "soft_assumption_used_fragment_rate": 100,
+    "clarification_precision": null,
+    "clarification_recall": null,
+    "false_clarification_rate": null
+  },
+  "budget": {
+    "requests_total": 0,
+    "retries_used": 0
+  },
+  "clarification_eval": {
+    "labeled_cases": 0,
+    "true_positive": 0,
+    "false_positive": 0,
+    "false_negative": 0
+  },
+  "route_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0,
+    "expected_routed_cases": 0,
+    "no_route_true_positive": 0,
+    "no_route_false_positive": 0
+  },
+  "scope_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0
+  },
+  "execution_state_eval": {
+    "checks_total": 2,
+    "checks_passed": 2
+  },
+  "route_distribution": {
+    "store_feature_risk": 1,
+    "hybrid_store_plus_live": 1
+  },
+  "fallback_distribution": {
+    "none": 2
+  },
+  "results": [
+    {
+      "case_id": "BQ-001",
+      "raw_question": "Проверь счет 60 за июнь 2020",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "ZQ4J6HgngY7JB4",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-002",
+      "raw_question": "Покажи риски по НДС и по закрытию",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "Vk2rIlI-UGAuAE",
+      "request_count_for_case": 0
+    }
+  ]
+}
--- a/llm_normalizer/data/eval_cases/eval-nL2h6Z0kKG.report.json
+++ b/llm_normalizer/data/eval_cases/eval-nL2h6Z0kKG.report.json
@ -0,0 +1,112 @@
+{
+  "run_id": "eval-nL2h6Z0kKG",
+  "timestamp": "2026-04-11T10:35:37.559Z",
+  "mode": "single-pass-strict",
+  "use_mock": true,
+  "prompt_version": "normalizer_v2_0_2",
+  "schema_version": "v2_0_2",
+  "dataset": {
+    "source": "inline_raw_questions",
+    "file": null,
+    "raw_questions_count": 2
+  },
+  "cases_total": 2,
+  "metrics": {
+    "schema_validation_pass_rate": 100,
+    "scope_detection_accuracy": null,
+    "scope_in_scope_rate": 100,
+    "multi_intent_detected_rate": 0,
+    "clarification_required_rate": 0,
+    "avg_fragments_per_message": 1,
+    "out_of_scope_fragment_rate": 0,
+    "routed_fragment_rate": 100,
+    "no_route_fragment_rate": 0,
+    "route_resolution_accuracy": null,
+    "no_route_precision": null,
+    "false_no_route_rate": null,
+    "execution_state_consistency_rate": 100,
+    "executable_with_soft_assumptions_rate": 100,
+    "soft_assumption_used_fragment_rate": 100,
+    "clarification_precision": null,
+    "clarification_recall": null,
+    "false_clarification_rate": null
+  },
+  "budget": {
+    "requests_total": 0,
+    "retries_used": 0
+  },
+  "clarification_eval": {
+    "labeled_cases": 0,
+    "true_positive": 0,
+    "false_positive": 0,
+    "false_negative": 0
+  },
+  "route_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0,
+    "expected_routed_cases": 0,
+    "no_route_true_positive": 0,
+    "no_route_false_positive": 0
+  },
+  "scope_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0
+  },
+  "execution_state_eval": {
+    "checks_total": 2,
+    "checks_passed": 2
+  },
+  "route_distribution": {
+    "store_feature_risk": 1,
+    "hybrid_store_plus_live": 1
+  },
+  "fallback_distribution": {
+    "none": 2
+  },
+  "results": [
+    {
+      "case_id": "BQ-001",
+      "raw_question": "Проверь счет 60 за июнь 2020",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "gKYkJDhEWkCQTw",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-002",
+      "raw_question": "Покажи риски по счету 97",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "YDK8VaP88Tx8dy",
+      "request_count_for_case": 0
+    }
+  ]
+}
--- a/llm_normalizer/data/eval_cases/eval-qYzl2S0IV9.report.json
+++ b/llm_normalizer/data/eval_cases/eval-qYzl2S0IV9.report.json
@ -0,0 +1,112 @@
+{
+  "run_id": "eval-qYzl2S0IV9",
+  "timestamp": "2026-04-11T10:37:58.754Z",
+  "mode": "single-pass-strict",
+  "use_mock": true,
+  "prompt_version": "normalizer_v2_0_2",
+  "schema_version": "v2_0_2",
+  "dataset": {
+    "source": "inline_raw_questions",
+    "file": null,
+    "raw_questions_count": 2
+  },
+  "cases_total": 2,
+  "metrics": {
+    "schema_validation_pass_rate": 100,
+    "scope_detection_accuracy": null,
+    "scope_in_scope_rate": 100,
+    "multi_intent_detected_rate": 0,
+    "clarification_required_rate": 0,
+    "avg_fragments_per_message": 1,
+    "out_of_scope_fragment_rate": 0,
+    "routed_fragment_rate": 100,
+    "no_route_fragment_rate": 0,
+    "route_resolution_accuracy": null,
+    "no_route_precision": null,
+    "false_no_route_rate": null,
+    "execution_state_consistency_rate": 100,
+    "executable_with_soft_assumptions_rate": 100,
+    "soft_assumption_used_fragment_rate": 100,
+    "clarification_precision": null,
+    "clarification_recall": null,
+    "false_clarification_rate": null
+  },
+  "budget": {
+    "requests_total": 0,
+    "retries_used": 0
+  },
+  "clarification_eval": {
+    "labeled_cases": 0,
+    "true_positive": 0,
+    "false_positive": 0,
+    "false_negative": 0
+  },
+  "route_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0,
+    "expected_routed_cases": 0,
+    "no_route_true_positive": 0,
+    "no_route_false_positive": 0
+  },
+  "scope_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0
+  },
+  "execution_state_eval": {
+    "checks_total": 2,
+    "checks_passed": 2
+  },
+  "route_distribution": {
+    "store_feature_risk": 1,
+    "hybrid_store_plus_live": 1
+  },
+  "fallback_distribution": {
+    "none": 2
+  },
+  "results": [
+    {
+      "case_id": "BQ-001",
+      "raw_question": "Проверь счет 60 за июнь 2020",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "3a_NZFB9MhKiKT",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-002",
+      "raw_question": "Покажи риски по счету 97",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "BrD8ORrgm8qsAy",
+      "request_count_for_case": 0
+    }
+  ]
+}
--- a/llm_normalizer/data/eval_cases/eval-wWlVBoabZW.report.json
+++ b/llm_normalizer/data/eval_cases/eval-wWlVBoabZW.report.json
@ -0,0 +1,112 @@
+{
+  "run_id": "eval-wWlVBoabZW",
+  "timestamp": "2026-04-11T10:14:44.777Z",
+  "mode": "single-pass-strict",
+  "use_mock": true,
+  "prompt_version": "normalizer_v2_0_2",
+  "schema_version": "v2_0_2",
+  "dataset": {
+    "source": "inline_raw_questions",
+    "file": null,
+    "raw_questions_count": 2
+  },
+  "cases_total": 2,
+  "metrics": {
+    "schema_validation_pass_rate": 100,
+    "scope_detection_accuracy": null,
+    "scope_in_scope_rate": 100,
+    "multi_intent_detected_rate": 0,
+    "clarification_required_rate": 0,
+    "avg_fragments_per_message": 1,
+    "out_of_scope_fragment_rate": 0,
+    "routed_fragment_rate": 100,
+    "no_route_fragment_rate": 0,
+    "route_resolution_accuracy": null,
+    "no_route_precision": null,
+    "false_no_route_rate": null,
+    "execution_state_consistency_rate": 100,
+    "executable_with_soft_assumptions_rate": 100,
+    "soft_assumption_used_fragment_rate": 100,
+    "clarification_precision": null,
+    "clarification_recall": null,
+    "false_clarification_rate": null
+  },
+  "budget": {
+    "requests_total": 0,
+    "retries_used": 0
+  },
+  "clarification_eval": {
+    "labeled_cases": 0,
+    "true_positive": 0,
+    "false_positive": 0,
+    "false_negative": 0
+  },
+  "route_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0,
+    "expected_routed_cases": 0,
+    "no_route_true_positive": 0,
+    "no_route_false_positive": 0
+  },
+  "scope_eval": {
+    "labeled_cases": 0,
+    "correct_cases": 0
+  },
+  "execution_state_eval": {
+    "checks_total": 2,
+    "checks_passed": 2
+  },
+  "route_distribution": {
+    "store_feature_risk": 1,
+    "hybrid_store_plus_live": 1
+  },
+  "fallback_distribution": {
+    "none": 2
+  },
+  "results": [
+    {
+      "case_id": "BQ-001",
+      "raw_question": "Проверь счет 60 за июнь 2020",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "3CW7yHQOH32rKS",
+      "request_count_for_case": 0
+    },
+    {
+      "case_id": "BQ-002",
+      "raw_question": "Покажи риски по счету 97",
+      "validation_passed": true,
+      "message_in_scope": true,
+      "scope_confidence": "high",
+      "contains_multiple_tasks": false,
+      "fragments_total": 1,
+      "in_scope_fragments": 1,
+      "out_of_scope_fragments": 0,
+      "unclear_fragments": 0,
+      "fallback_type": "none",
+      "predicted_route_status": "routed",
+      "expected_route_status": null,
+      "predicted_no_route_reason": null,
+      "expected_no_route_reason": null,
+      "predicted_clarification_required": false,
+      "expected_clarification_required": null,
+      "executable_with_soft_assumptions_fragments": 1,
+      "trace_id": "Gdw6ytd-K55hUm",
+      "request_count_for_case": 0
+    }
+  ]
+}
--- a/tmp_llm_gate_log.txt
+++ b/tmp_llm_gate_log.txt