АДРЕСНЫЙ РЕЖИМ -ADDRESS:Шаг 1 - ЛЛМ ФЕРСТ + feat(address): стабилизация wave1 dynamic resolver контрагентов, follow-up carryover и актуализация docs/tests

2026-04-02 23:11:49 +03:00
parent 53716e548e
commit 88094c09f8
243 changed files with 94296 additions and 181 deletions
@@ -0,0 +1,33 @@
+# 2026-04-02_Address_GateBlock_Probe_WithMetrics
+
+Generated at: 2026-04-02T22:11:14
+Questions file: X:\1C\NDC_1C\docs\ADDRESS\question_sets\temp_gate_block_probe_2026-04-02.json
+Backend URL: http://127.0.0.1:8787/api/assistant/message
+LLM: local / qwen2.5-14b-instruct-1m @ http://127.0.0.1:1234
+Strict policy: semantic
+
+## Totals
+- questions_total: 2
+- ok_200_count: 2
+- semantic_pass_count: 1
+- semantic_pass_rate: 0.5
+- route_pass_count: 1
+- route_pass_rate: 0.5
+- strict_pass_count: 1
+- strict_pass_rate: 0.5
+- factual_count: 1
+- partial_coverage_count: 0
+- clarification_required_count: 1
+- http_error_count: 0
+- llm_decomposition_attempted_count: 1
+- llm_decomposition_applied_count: 0
+- llm_fallback_count: 0
+- llm_fallback_rate: 0.0
+- tool_gate_blocked_count: 0
+- tool_gate_blocked_rate: 0.0
+- avg_elapsed_ms: 8938
+
+## Files
+- run_summary.json
+- full_live_results.json
+- failures_only.json
@@ -0,0 +1,395 @@
+[
+  {
+    "index": 1,
+    "id": "G001",
+    "question": "привет как дела",
+    "session": null,
+    "session_id": "asst-2026-04-02_Address_GateBlock_Probe_WithMetrics-g001",
+    "status_code": 200,
+    "ok": true,
+    "elapsed_ms": 12732,
+    "reply_type": "clarification_required",
+    "trace_id": "y_xkHosQzDZRaL",
+    "assistant_reply": "Коротко: Наиболее вероятный механизм проблемы подтвержден частично и требует первичной проверки. Что сломано:\n- Есть признаки проблемы, но без уточнений по периоду и объекту вывод ненадежен. Почему это похоже на проблему:\n- Сигнал проблемы есть, но механизм подтвержден не полностью. На чем это основано:\n- Опора частичная: часть требований покрыта не полностью.\n- Отдельно не подтверждено или покрыто частично: R1. Что проверить первым:\n- Уточните период проверки, чтобы подтвердить проблему без лишнего шума.\n- Проверьте связку документов и проводок по проблемному участку в указанном периоде.\n- Уточните период проверки (например, июль 2020). Ограничения:\n- Ни одно требование не получило подтвержденного покрытия.\n- Нет explainable-сигналов why_included в результатах выборки.\n- required anchors claim-bound grounding.\n- evidence.\n- Targeted evidence acquisition claim target path.",
+    "assistant_reply_first_line": "Коротко: Наиболее вероятный механизм проблемы подтвержден частично и требует первичной проверки. Что сломано:",
+    "expected_intent": null,
+    "actual_intent": null,
+    "intent_match": true,
+    "expected_mode": "address_query",
+    "actual_mode": null,
+    "mode_match": false,
+    "expected_reply_type": null,
+    "reply_match": true,
+    "semantic_pass": false,
+    "route_pass": false,
+    "route_health": "blocked_clarification",
+    "strict_policy": "semantic",
+    "strict_pass": false,
+    "selected_recipe": null,
+    "missing_required_filters": null,
+    "match_failure_stage": null,
+    "match_failure_reason": null,
+    "rows_fetched": null,
+    "rows_matched": null,
+    "mcp_call_status": null,
+    "limited_reason_category": null,
+    "llm_decomposition_attempted": null,
+    "llm_decomposition_applied": null,
+    "llm_decomposition_reason": null,
+    "llm_canonical_candidate_detected": null,
+    "fallback_rule_hit": null,
+    "tool_gate_decision": null,
+    "tool_gate_reason": null,
+    "predecompose_contract_intent": null,
+    "predecompose_contract_aggregation_profile": null,
+    "predecompose_contract_period_scope": null,
+    "debug_payload": {
+      "trace_id": "y_xkHosQzDZRaL",
+      "prompt_version": "address_query_runtime_v1",
+      "schema_version": "v1",
+      "fallback_type": "clarification",
+      "route_summary": null,
+      "fragments": [],
+      "requirements_extracted": [
+        {
+          "requirement_id": "R1",
+          "source_fragment_id": null,
+          "requirement_text": "привет как дела",
+          "subject_tokens": [],
+          "status": "clarification_needed",
+          "route": null
+        }
+      ],
+      "coverage_report": {
+        "requirements_total": 1,
+        "requirements_covered": 0,
+        "requirements_uncovered": [],
+        "requirements_partially_covered": [],
+        "clarification_needed_for": [
+          "R1"
+        ],
+        "out_of_scope_requirements": []
+      },
+      "routes": [],
+      "retrieval_status": [],
+      "retrieval_results": [],
+      "answer_grounding_check": {
+        "status": "no_grounded_answer",
+        "route_subject_match": true,
+        "missing_requirements": [
+          "R1"
+        ],
+        "reasons": [
+          "Ни одно требование не получило подтвержденного покрытия.",
+          "Нет explainable-сигналов why_included в результатах выборки.",
+          "РќРµРґРѕСЃС‚Р°С‚РѕС‡РЅРѕ РїРѕРєСЂС‹С‚РёСЏ required anchors РґР»СЏ claim-bound grounding.",
+          "РќРµРґРѕСЃС‚Р°С‚РѕС‡РЅРѕ РґРѕРїСѓСЃС‚РёРјРѕРіРѕ evidence РґР»СЏ РѕР±РѕСЃРЅРѕРІР°РЅРЅРѕРіРѕ РѕС‚РІРµС‚Р°.",
+          "Targeted evidence acquisition РЅРµ РґР°Р» РґРѕРїСѓСЃС‚РёРјС‹С… РїРѕРїР°РґР°РЅРёР№ РїРѕ claim target path."
+        ],
+        "why_included_summary": [],
+        "selection_reason_summary": []
+      },
+      "dropped_intent_segments": [],
+      "question_type_class": "why_breaks",
+      "company_anchors": {
+        "contract_numbers": [],
+        "document_numbers": [],
+        "dates": [],
+        "amounts": [],
+        "accounts": [],
+        "periods": [],
+        "document_types": [],
+        "all": []
+      },
+      "business_scope_raw": [],
+      "business_scope_resolved": [
+        "company_specific_accounting"
+      ],
+      "company_grounding_applied": true,
+      "scope_resolution_reason": [
+        "settlement_claim_company_scope_recovery"
+      ],
+      "company_scope_resolution_reason": [
+        "settlement_claim_company_scope_recovery"
+      ],
+      "raw_time_anchor": null,
+      "raw_time_scope": null,
+      "resolved_time_anchor": null,
+      "resolved_primary_period": null,
+      "effective_primary_period": null,
+      "temporal_guard_input": null,
+      "temporal_alignment_status": "conflicting",
+      "temporal_resolution_source": "normalized_time_scope:missing",
+      "temporal_guard_basis": "none",
+      "temporal_guard_applied": false,
+      "temporal_guard_outcome": "passed",
+      "temporal_guard": {
+        "raw_time_anchor": null,
+        "raw_time_scope": null,
+        "resolved_time_anchor": null,
+        "resolved_primary_period": null,
+        "effective_primary_period": null,
+        "temporal_guard_input": null,
+        "temporal_alignment_status": "conflicting",
+        "temporal_resolution_source": "normalized_time_scope:missing",
+        "temporal_guard_basis": "none",
+        "temporal_guard_applied": false,
+        "temporal_guard_outcome": "passed",
+        "primary_period_window": null,
+        "allowed_context_window": null,
+        "controlled_temporal_expansion_enabled": false,
+        "context_expansion_reasons_allowed": [
+          "prehistory",
+          "carryover",
+          "post_period_closure",
+          "long_running_contract_context"
+        ],
+        "normalized_anchor_drift_detected": false,
+        "reason_codes": [
+          "missing_resolved_primary_period"
+        ]
+      },
+      "raw_numeric_tokens": [],
+      "classified_numeric_tokens": [],
+      "rejected_as_non_accounts": [],
+      "resolved_account_anchors": [],
+      "domain_polarity_guard": {
+        "applied": false,
+        "polarity": "not_applicable",
+        "outcome": "not_applicable",
+        "supplier_score": 0,
+        "customer_score": 0,
+        "account_scope": [],
+        "raw_numeric_tokens": [],
+        "classified_numeric_tokens": [],
+        "rejected_as_non_accounts": [],
+        "resolved_account_anchors": [],
+        "rejected_problem_units": 0,
+        "rejected_evidence": 0,
+        "critical_contradiction": false,
+        "reason_codes": []
+      },
+      "claim_anchor_audit": {
+        "claim_type": "prove_settlement_closure_state",
+        "settlement_role": "unknown",
+        "settlement_role_resolution_reason": [
+          "settlement_role_resolved_unknown"
+        ],
+        "polarity_resolution_status": "unknown",
+        "required_anchors": [
+          "period",
+          "account_scope",
+          "counterparty_scope_or_contract",
+          "closure_signal"
+        ],
+        "resolved_anchors": {
+          "period": [],
+          "account_scope": [],
+          "amounts": [],
+          "contract": [],
+          "document_numbers": [],
+          "document_types": [],
+          "counterparty_scope": [],
+          "advance_signal": [],
+          "closure_signal": [],
+          "vat_signal": [],
+          "chain_signal": [],
+          "close_signal": [],
+          "cost_scope": [],
+          "rbp_signal": [],
+          "writeoff_signal": [],
+          "fixed_asset_signal": [],
+          "amortization_signal": [],
+          "expected_fa_set": [],
+          "actual_fa_set": []
+        },
+        "missing_anchors": [
+          "period",
+          "account_scope",
+          "counterparty_scope_or_contract",
+          "closure_signal"
+        ],
+        "claim_anchor_resolution_rate": 0,
+        "primary_period": null,
+        "allowed_context_window": null,
+        "context_expansion_reasons_allowed": [
+          "prehistory",
+          "carryover",
+          "post_period_closure",
+          "long_running_contract_context"
+        ],
+        "reason_codes": [
+          "claim_missing_required_anchors",
+          "claim_anchor_resolution_low",
+          "unresolved_supplier_customer_polarity"
+        ]
+      },
+      "settlement_role": "unknown",
+      "settlement_role_resolution_reason": [
+        "settlement_role_resolved_unknown"
+      ],
+      "polarity_resolution_status": "unknown",
+      "targeted_evidence_acquisition": {
+        "claim_type": "prove_settlement_closure_state",
+        "required_checks": [
+          "payment_document_found",
+          "contract_matched",
+          "settlement_object_matched",
+          "closing_document_found",
+          "register_closure_entry_found",
+          "posting_link_found"
+        ],
+        "check_status": {
+          "payment_document_found": "not_found",
+          "contract_matched": "not_found",
+          "settlement_object_matched": "not_found",
+          "closing_document_found": "not_found",
+          "register_closure_entry_found": "not_found",
+          "posting_link_found": "not_found"
+        },
+        "targeted_item_hits": 0,
+        "targeted_evidence_hits": 0,
+        "targeted_evidence_hit_rate": 0,
+        "targeted_evidence_source_refs": [],
+        "reason_codes": [
+          "targeted_evidence_not_found",
+          "targeted_evidence_hit_rate_low"
+        ]
+      },
+      "evidence_admissibility_gate": {
+        "candidate_evidence_total": 0,
+        "admissible_evidence_count": 0,
+        "rejected_evidence_count": 0,
+        "rejected_item_count": 0,
+        "reject_breakdown": {
+          "wrong_period": 0,
+          "wrong_domain": 0,
+          "wrong_account_scope": 0,
+          "weak_source_mapping": 0,
+          "zero_live_match": 0,
+          "future_dated_or_out_of_window": 0
+        },
+        "category_breakdown": {
+          "hard_evidence": 0,
+          "supporting_signal": 0,
+          "inadmissible_noise": 0
+        },
+        "reason_codes": [
+          "no_admissible_evidence_for_grounded_answer"
+        ]
+      },
+      "eligibility_time_basis": "none",
+      "grounded_answer_eligibility_guard": {
+        "eligible": false,
+        "temporal_passed": true,
+        "eligibility_time_basis": "none",
+        "business_scope_passed": true,
+        "polarity_passed": true,
+        "claim_anchors_passed": false,
+        "claim_anchor_resolution_rate": 0,
+        "missing_required_anchors": 4,
+        "admissible_evidence_count": 0,
+        "critical_contradiction": false,
+        "outcome": "limited_or_insufficient_evidence",
+        "grounding_mode": "limited_or_insufficient_evidence",
+        "reason_codes": [
+          "claim_anchor_coverage_insufficient",
+          "admissible_evidence_count_zero",
+          "targeted_evidence_hit_rate_zero"
+        ]
+      },
+      "problem_centric_answer_applied": false,
+      "problem_units_used_count": 0,
+      "problem_answer_mode": "stage1_policy_v11",
+      "answer_structure_v11": {
+        "schema_version": "answer_structure_v1_1",
+        "answer_summary": "Нужны уточнения: без сужения фокуса надежный вывод невозможен.",
+        "direct_answer": "Есть признаки проблемы, но без уточнений по периоду и объекту вывод ненадежен.",
+        "mechanism_block": {
+          "status": "unresolved",
+          "mechanism_notes": [],
+          "limitation_reason_codes": []
+        },
+        "evidence_block": {
+          "evidence_ids": [],
+          "source_refs": [],
+          "mechanism_notes": [],
+          "coverage_note": "coverage_partial_or_limited"
+        },
+        "uncertainty_block": {
+          "open_uncertainties": [
+            "R1",
+            "missing_anchor:period",
+            "missing_anchor:account",
+            "missing_anchor:document_or_object",
+            "missing_anchor:counterparty"
+          ],
+          "limitations": [
+            "Ни одно требование не получило подтвержденного покрытия.",
+            "Нет explainable-сигналов why_included в результатах выборки.",
+            "РќРµРґРѕСЃС‚Р°С‚РѕС‡РЅРѕ РїРѕРєСЂС‹С‚РёСЏ required anchors РґР»СЏ claim-bound grounding.",
+            "РќРµРґРѕСЃС‚Р°С‚РѕС‡РЅРѕ РґРѕРїСѓСЃС‚РёРјРѕРіРѕ evidence РґР»СЏ РѕР±РѕСЃРЅРѕРІР°РЅРЅРѕРіРѕ РѕС‚РІРµС‚Р°.",
+            "Targeted evidence acquisition РЅРµ РґР°Р» РґРѕРїСѓСЃС‚РёРјС‹С… РїРѕРїР°РґР°РЅРёР№ РїРѕ claim target path."
+          ]
+        },
+        "next_step_block": {
+          "recommended_actions": [
+            "Р”Р°Р№С‚Рµ РЅРµРґРѕСЃС‚Р°СЋС‰РёРµ СЏРєРѕСЂСЏ (РїРµСЂРёРѕРґ/СЃС‡РµС‚/РѕР±СЉРµРєС‚), РёРЅР°С‡Рµ СЃРёР»СЊРЅС‹Р№ factual РІС‹РІРѕРґ РЅРµРІРѕР·РјРѕР¶РµРЅ."
+          ],
+          "clarification_questions": [
+            "Уточните период проверки (например, июль 2020).",
+            "РЈС‚РѕС‡РЅРёС‚Рµ СЃС‡РµС‚ РёР»Рё РіСЂСѓРїРїСѓ СЃС‡РµС‚РѕРІ (РЅР°РїСЂРёРјРµСЂ, 19, 60, 62).",
+            "РЈРєР°Р¶РёС‚Рµ РґРѕРєСѓРјРµРЅС‚/GUID/РєРѕРЅРєСЂРµС‚РЅС‹Р№ РѕР±СЉРµРєС‚ РґР»СЏ С‚СЂР°СЃСЃРёСЂРѕРІРєРё.",
+            "РЈРєР°Р¶РёС‚Рµ РєРѕРЅС‚СЂР°РіРµРЅС‚Р° РёР»Рё РіСЂСѓРїРїСѓ РєРѕРЅС‚СЂР°РіРµРЅС‚РѕРІ.",
+            "Р—Р°РєСЂРѕР№С‚Рµ СѓС‚РѕС‡РЅРµРЅРёСЏ РґР»СЏ С‚СЂРµР±РѕРІР°РЅРёР№: R1."
+          ]
+        }
+      },
+      "investigation_state_snapshot": {
+        "schema_version": "investigation_state_v1",
+        "session_id": "asst-2026-04-02_Address_GateBlock_Probe_WithMetrics-g001",
+        "status": "active",
+        "turn_index": 1,
+        "updated_at": "2026-04-02T19:11:09.280Z",
+        "question_id": "msg-wm3KZbU0j2",
+        "question_scope_id": "s:привет как дела",
+        "scope_origin": "underspecified",
+        "focus": {
+          "domain": null,
+          "period": null,
+          "primary_accounts": [],
+          "active_query_subject": "привет как дела"
+        },
+        "narrowing_status": "unknown",
+        "evidence_refs": [],
+        "open_uncertainties": [
+          "clarify:R1"
+        ],
+        "last_answer_mode": "clarification_required",
+        "followup_context": {
+          "previous_question_id": null,
+          "last_user_message": "привет как дела",
+          "referenced_requirement_ids": [
+            "R1"
+          ],
+          "active_domain": null,
+          "active_requirement_ids": [
+            "R1"
+          ],
+          "uncovered_requirement_ids": [
+            "R1"
+          ],
+          "last_problem_unit_id": null,
+          "settlement_next_actions": [],
+          "evidence_summary": [],
+          "question_scope_id": "s:привет как дела",
+          "scope_origin": "underspecified"
+        },
+        "query_mode_hint": "investigation_candidate"
+      },
+      "normalized": null
+    },
+    "error_code": null,
+    "error_message": null
+  }
+]
@@ -0,0 +1,6 @@
+# Response Audit: 2026-04-02_Address_GateBlock_Probe_WithMetrics
+
+| id | strict | route_health | reply_type | intent | limited_reason | question | assistant_first_line |
+|---|---|---|---|---|---|---|---|
+| G001 | False | blocked_clarification | clarification_required | None | None | привет как дела | Коротко: Наиболее вероятный механизм проблемы подтвержден частично и требует первичной проверки. Что сломано: |
+| G002 | True | ok_or_factual | factual | counterparty_activity_lifecycle | None | выведи список заказчиков за все время | Собран профиль активности заказчиков (bank-doc activity aggregate). |
@@ -0,0 +1,90 @@
+{
+  "run_id": "2026-04-02_Address_GateBlock_Probe_WithMetrics",
+  "generated_at": "2026-04-02T22:11:14",
+  "source_questions_file": "X:\\1C\\NDC_1C\\docs\\ADDRESS\\question_sets\\temp_gate_block_probe_2026-04-02.json",
+  "backend_url": "http://127.0.0.1:8787/api/assistant/message",
+  "llm_provider": "local",
+  "llm_model": "qwen2.5-14b-instruct-1m",
+  "llm_base_url": "http://127.0.0.1:1234",
+  "strict_policy": "semantic",
+  "totals": {
+    "questions_total": 2,
+    "ok_200_count": 2,
+    "semantic_pass_count": 1,
+    "semantic_pass_rate": 0.5,
+    "route_pass_count": 1,
+    "route_pass_rate": 0.5,
+    "strict_pass_count": 1,
+    "strict_pass_rate": 0.5,
+    "factual_count": 1,
+    "partial_coverage_count": 0,
+    "clarification_required_count": 1,
+    "http_error_count": 0,
+    "llm_decomposition_attempted_count": 1,
+    "llm_decomposition_applied_count": 0,
+    "llm_fallback_count": 0,
+    "llm_fallback_rate": 0.0,
+    "tool_gate_blocked_count": 0,
+    "tool_gate_blocked_rate": 0.0,
+    "avg_elapsed_ms": 8938
+  },
+  "distributions": {
+    "reply_type": {
+      "clarification_required": 1,
+      "factual": 1
+    },
+    "actual_intent": {
+      "None": 1,
+      "counterparty_activity_lifecycle": 1
+    },
+    "actual_mode": {
+      "None": 1,
+      "address_query": 1
+    },
+    "mcp_call_status": {
+      "None": 1,
+      "matched_non_empty": 1
+    },
+    "limited_reason_category": {},
+    "route_health": {
+      "blocked_clarification": 1,
+      "ok_or_factual": 1
+    },
+    "tool_gate_decision": {
+      "run_address_lane": 1
+    },
+    "tool_gate_reason": {
+      "address_mode_classifier_detected": 1
+    }
+  },
+  "address_llm_predecompose_metrics": {
+    "overall": {
+      "llm_attempted": 1,
+      "llm_applied": 0,
+      "fallback_used": 0,
+      "fallback_rate": 0.0,
+      "tool_gate_blocked": 0,
+      "gate_block_rate": 0.0
+    },
+    "by_intent": {
+      "unknown": {
+        "total": 1,
+        "llm_attempted": 0,
+        "llm_applied": 0,
+        "fallback_used": 0,
+        "tool_gate_blocked": 0,
+        "fallback_rate": 0.0,
+        "gate_block_rate": 0.0
+      },
+      "counterparty_activity_lifecycle": {
+        "total": 1,
+        "llm_attempted": 1,
+        "llm_applied": 0,
+        "fallback_used": 0,
+        "tool_gate_blocked": 0,
+        "fallback_rate": 0.0,
+        "gate_block_rate": 0.0
+      }
+    }
+  }
+}