469 lines
25 KiB
JSON
469 lines
25 KiB
JSON
{
|
||
"trace_id": "mevf8cflx6G80P",
|
||
"timestamp": "2026-03-23T18:08:16.744Z",
|
||
"model": "gpt-4o-mini",
|
||
"prompt_version": "normalizer_v2",
|
||
"schema_version": "v2",
|
||
"case_id": "BQ-005",
|
||
"user_question_raw": "Есть ли такие зависшие авансы, которые уже давно надо было либо закрыть, либо хотя бы перепроверить руками?",
|
||
"context": {
|
||
"period_hint": null,
|
||
"business_context": null,
|
||
"expected_route": null,
|
||
"case_id": "BQ-005",
|
||
"eval_mode": "single-pass-strict"
|
||
},
|
||
"request_payload_redacted": {
|
||
"model": "gpt-4o-mini",
|
||
"baseUrl": "https://api.openai.com/v1",
|
||
"temperature": 0,
|
||
"maxOutputTokens": 900,
|
||
"promptVersion": "normalizer_v2",
|
||
"systemPrompt": "Ты semantic-normalizer для бухгалтерского ассистента NDC.\nТвоя роль: только нормализация запроса пользователя в строгий JSON-контракт.\n\nЖесткие правила:\n1) Не давай бухгалтерский ответ по сути вопроса.\n2) Возвращай только JSON без markdown и пояснений.\n3) JSON обязан соответствовать переданной schema normalized_query_v1.\n4) Если период не указан, не выдумывай его; отмечай ambiguity.\n5) Для цепочек документов/проводок/оплат поднимай causal и cross-entity признаки.\n6) Для точечного object trace (номер/строка/ref) поднимай needs_exact_object_trace=true.\n7) Используй терминологию NDC.",
|
||
"developerPrompt": "You are a decomposition-first semantic parser for accounting assistant NDC.\nReturn JSON only, strictly matching `normalized_query_v2`.\nDo not answer the accounting question itself.\n\nCore workflow:\n1) Detect if the user message is in company-specific accounting scope.\n2) Split message into meaningful task fragments.\n3) Discard obvious noise fragments.\n4) For each kept fragment, output semantic hints + flags + candidate_labels.\n5) Keep routing as indirect signals only (flags + labels). Final route is chosen by deterministic code.\n\nDomain gating policy:\n- `in_scope` only for requests about current company data and accounting ontology (documents, postings, balances, settlements, period close risks, anomalies, controls).\n- `out_of_scope` for generic accounting theory, legal questions, standards/laws, unrelated business or casual/offtopic chat.\n- `unclear` when intent may be accounting-related but insufficiently grounded in company context.\n\nbusiness_scope mapping:\n- `company_specific_accounting` for in-scope enterprise-data tasks.\n- `generic_accounting` for abstract accounting questions not tied to this company data.\n- `offtopic` for non-accounting content.\n- `unclear` when not enough signal.\n\nFragment policy:\n- Keep user wording in `raw_fragment_text`.\n- Make `normalized_fragment_text` concise and business-readable.\n- Populate `entity_hints/account_hints/document_hints/register_hints` only from message evidence.\n- `candidate_labels` can contain multiple labels. Do not force a single mandatory intent.\n- `fragment_id` must be stable and sequential: F1, F2, F3...\n\nFlag policy:\n- has_multi_entity_scope: requires linking multiple accounting entities/documents/payments/postings.\n- asks_for_chain_explanation: asks to explain “why”, “where chain breaks”, or causal reconstruction.\n- asks_for_ranking_or_top: asks top/ranking/prioritized ordering.\n- asks_for_period_summary: asks broad period/company summary or slice.\n- asks_for_rule_check: asks control/rule compliance checks.\n- asks_for_anomaly_scan: asks suspicious zones/anomaly search.\n- asks_for_exact_object_trace: asks exact object trace (document number/ref/line).\n- asks_for_evidence: asks documentary/proof evidence.\n- mentions_period_close_context: explicit pre-close/close/reporting deadline context.\n\nDiscard policy:\n- Put chatter/noise/meta filler into `discarded_fragments` with reason.\n- Do not discard valid accounting tasks, even if wording is slang-like.\n\nConfidence policy:\n- high only for clear, explicit fragment semantics.\n- medium for normal ambiguity.\n- low for unclear or weakly grounded fragments.\n- scope_confidence reflects whole-message domain certainty.\n\nglobal_notes policy:\n- needs_clarification=true when period/object/scope is missing for reliable execution.\n- clarification_reason should be concrete and short.\n\nStrictness:\n- Output only valid JSON object matching `normalized_query_v2`.\n- No markdown, no prose, no extra fields.",
|
||
"domainPrompt": "Контекст домена: бухгалтерия 1С/NDC.\n\nКлючевые счета:\n- 01, 02, 10, 41, 51, 60, 62, 68.02, 90, 97.\n\nТиповые сущности:\n- контрагент, договор, документ реализации, документ поступления, оплата, проводка, регистр, закрывающий документ.\n\nЛексика causal и сверки (сильные сигналы для cross_entity):\n- \"не бьется\", \"не сходится\", \"не видно\", \"не собралось\", \"повисло\", \"хвост\";\n- \"разложи по документам/оплатам/закрывающим\";\n- \"чем подтверждается\", \"где ошибка в цепочке\", \"что пошло криво\".\n\nЛексика точечного drilldown:\n- \"документ №...\", \"ref\", \"строка проводки\", \"покажи конкретную операцию\", \"точный source-of-record\".\n\nЛексика rule-based контроля:\n- \"проверь настройки\", \"ошибка срока/даты\", \"контроль 97/10/ОС\", \"нарушение правила учета\".\n\nЛексика обзорной аналитики:\n- \"рейтинг\", \"топ рисков\", \"в целом по компании\", \"перед закрытием периода\", \"приоритизация проверок\".\n\nВажное правило:\nЕсли в одном вопросе есть и риск-лексика, и цепочка document/payment/posting, не понижать задачу до чистого `store_feature_risk`.\nПриоритет у causal cross-entity семантики.",
|
||
"fewShotExamples": "Q:\nПо поставщикам на конец месяца где хвосты по взаиморасчетам, разложи где рвется цепочка документов и оплат.\n\nExpected (fragment-level shape):\n{\n \"message_in_scope\": true,\n \"contains_multiple_tasks\": false,\n \"fragments\": [\n {\n \"fragment_id\": \"F1\",\n \"domain_relevance\": \"in_scope\",\n \"business_scope\": \"company_specific_accounting\",\n \"account_hints\": [\"60\"],\n \"flags\": {\n \"has_multi_entity_scope\": true,\n \"asks_for_chain_explanation\": true,\n \"asks_for_ranking_or_top\": false,\n \"asks_for_period_summary\": false,\n \"asks_for_rule_check\": false,\n \"asks_for_anomaly_scan\": true,\n \"asks_for_exact_object_trace\": false,\n \"asks_for_evidence\": true,\n \"mentions_period_close_context\": true\n },\n \"candidate_labels\": [\"cross_entity\", \"anomaly_probe\"]\n }\n ]\n}\n\nQ:\nСкажи как вообще по ФСБУ правильно закрывать 97 счет, и еще проверь у нас по июню подозрительные РБП.\n\nExpected (fragment-level shape):\n{\n \"message_in_scope\": true,\n \"contains_multiple_tasks\": true,\n \"fragments\": [\n {\n \"fragment_id\": \"F1\",\n \"domain_relevance\": \"out_of_scope\",\n \"business_scope\": \"generic_accounting\",\n \"flags\": {\n \"asks_for_rule_check\": true,\n \"asks_for_anomaly_scan\": false\n },\n \"candidate_labels\": []\n },\n {\n \"fragment_id\": \"F2\",\n \"domain_relevance\": \"in_scope\",\n \"business_scope\": \"company_specific_accounting\",\n \"account_hints\": [\"97\"],\n \"flags\": {\n \"asks_for_rule_check\": true,\n \"asks_for_anomaly_scan\": true,\n \"asks_for_period_summary\": false\n },\n \"candidate_labels\": [\"rule_based_account_control\", \"anomaly_probe\"]\n }\n ],\n \"global_notes\": {\n \"needs_clarification\": false\n }\n}\n\nQ:\nДокумент TRX-88 покажи и связанную строку проводки, а еще где по банку повторяется паттерн подозрительных хвостов.\n\nExpected (fragment-level shape):\n{\n \"message_in_scope\": true,\n \"contains_multiple_tasks\": true,\n \"fragments\": [\n {\n \"fragment_id\": \"F1\",\n \"domain_relevance\": \"in_scope\",\n \"business_scope\": \"company_specific_accounting\",\n \"flags\": {\n \"asks_for_exact_object_trace\": true,\n \"asks_for_evidence\": true,\n \"has_multi_entity_scope\": false\n },\n \"candidate_labels\": [\"drilldown_explain\"]\n },\n {\n \"fragment_id\": \"F2\",\n \"domain_relevance\": \"in_scope\",\n \"business_scope\": \"company_specific_accounting\",\n \"flags\": {\n \"asks_for_anomaly_scan\": true,\n \"asks_for_ranking_or_top\": true,\n \"asks_for_period_summary\": true\n },\n \"candidate_labels\": [\"heavy_analytical\", \"anomaly_probe\"]\n }\n ]\n}",
|
||
"userQuestion": "Есть ли такие зависшие авансы, которые уже давно надо было либо закрыть, либо хотя бы перепроверить руками?",
|
||
"context": {
|
||
"eval_label": "eval-aOEoelkQb2",
|
||
"case_id": "BQ-005",
|
||
"eval_mode": "single-pass-strict"
|
||
},
|
||
"retryPolicy": "single-pass-strict",
|
||
"useMock": false
|
||
},
|
||
"raw_model_response": {
|
||
"id": "resp_0c586a71c2fffee00069c18182ff088195b885dc7e69826973",
|
||
"object": "response",
|
||
"created_at": 1774289283,
|
||
"status": "completed",
|
||
"background": false,
|
||
"billing": {
|
||
"payer": "developer"
|
||
},
|
||
"completed_at": 1774289296,
|
||
"error": null,
|
||
"frequency_penalty": 0,
|
||
"incomplete_details": null,
|
||
"instructions": null,
|
||
"max_output_tokens": 900,
|
||
"max_tool_calls": null,
|
||
"model": "gpt-4o-mini-2024-07-18",
|
||
"output": [
|
||
{
|
||
"id": "msg_0c586a71c2fffee00069c18185ec0c81959f82947dd23057fc",
|
||
"type": "message",
|
||
"status": "completed",
|
||
"content": [
|
||
{
|
||
"type": "output_text",
|
||
"annotations": [],
|
||
"logprobs": [],
|
||
"text": "{\n \"schema_version\": \"normalized_query_v2\",\n \"user_message_raw\": \"Есть ли такие зависшие авансы, которые уже давно надо было либо закрыть, либо хотя бы перепроверить руками?\",\n \"message_in_scope\": true,\n \"scope_confidence\": \"high\",\n \"contains_multiple_tasks\": false,\n \"fragments\": [\n {\n \"fragment_id\": \"F1\",\n \"raw_fragment_text\": \"зависшие авансы, которые уже давно надо было либо закрыть, либо хотя бы перепроверить руками\",\n \"normalized_fragment_text\": \"проверка зависших авансов на закрытие или перепроверку\",\n \"domain_relevance\": \"in_scope\",\n \"business_scope\": \"company_specific_accounting\",\n \"entity_hints\": [],\n \"account_hints\": [\"51\", \"60\"],\n \"document_hints\": [],\n \"register_hints\": [],\n \"time_scope\": {\n \"type\": \"missing\",\n \"value\": null,\n \"confidence\": \"low\"\n },\n \"flags\": {\n \"has_multi_entity_scope\": false,\n \"asks_for_chain_explanation\": false,\n \"asks_for_ranking_or_top\": false,\n \"asks_for_period_summary\": false,\n \"asks_for_rule_check\": false,\n \"asks_for_anomaly_scan\": true,\n \"asks_for_exact_object_trace\": false,\n \"asks_for_evidence\": false,\n \"mentions_period_close_context\": false\n },\n \"candidate_labels\": [\"anomaly_probe\"]\n \n \t\t\n \t\t,\n \"confidence\": \"medium\"\n }\n ],\n \"discarded_fragments\": [],\n \"global_notes\": {\n \"needs_clarification\": true,\n \"clarification_reason\": \"не указаны конкретные сроки для проверки зависших авансов\"\n }\n}"
|
||
}
|
||
],
|
||
"role": "assistant"
|
||
}
|
||
],
|
||
"parallel_tool_calls": true,
|
||
"presence_penalty": 0,
|
||
"previous_response_id": null,
|
||
"prompt_cache_key": null,
|
||
"prompt_cache_retention": null,
|
||
"reasoning": {
|
||
"effort": null,
|
||
"summary": null
|
||
},
|
||
"safety_identifier": null,
|
||
"service_tier": "default",
|
||
"store": true,
|
||
"temperature": 0,
|
||
"text": {
|
||
"format": {
|
||
"type": "json_schema",
|
||
"description": null,
|
||
"name": "normalized_query_v2",
|
||
"schema": {
|
||
"$id": "normalized_query_v2",
|
||
"title": "Normalized Query V2",
|
||
"type": "object",
|
||
"additionalProperties": false,
|
||
"required": [
|
||
"schema_version",
|
||
"user_message_raw",
|
||
"message_in_scope",
|
||
"scope_confidence",
|
||
"contains_multiple_tasks",
|
||
"fragments",
|
||
"discarded_fragments",
|
||
"global_notes"
|
||
],
|
||
"properties": {
|
||
"schema_version": {
|
||
"type": "string",
|
||
"const": "normalized_query_v2"
|
||
},
|
||
"user_message_raw": {
|
||
"type": "string",
|
||
"minLength": 1
|
||
},
|
||
"message_in_scope": {
|
||
"type": "boolean"
|
||
},
|
||
"scope_confidence": {
|
||
"type": "string",
|
||
"enum": [
|
||
"high",
|
||
"medium",
|
||
"low"
|
||
]
|
||
},
|
||
"contains_multiple_tasks": {
|
||
"type": "boolean"
|
||
},
|
||
"fragments": {
|
||
"type": "array",
|
||
"items": {
|
||
"type": "object",
|
||
"additionalProperties": false,
|
||
"required": [
|
||
"fragment_id",
|
||
"raw_fragment_text",
|
||
"normalized_fragment_text",
|
||
"domain_relevance",
|
||
"business_scope",
|
||
"entity_hints",
|
||
"account_hints",
|
||
"document_hints",
|
||
"register_hints",
|
||
"time_scope",
|
||
"flags",
|
||
"candidate_labels",
|
||
"confidence"
|
||
],
|
||
"properties": {
|
||
"fragment_id": {
|
||
"type": "string",
|
||
"minLength": 1
|
||
},
|
||
"raw_fragment_text": {
|
||
"type": "string",
|
||
"minLength": 1
|
||
},
|
||
"normalized_fragment_text": {
|
||
"type": "string",
|
||
"minLength": 1
|
||
},
|
||
"domain_relevance": {
|
||
"type": "string",
|
||
"enum": [
|
||
"in_scope",
|
||
"out_of_scope",
|
||
"unclear"
|
||
]
|
||
},
|
||
"business_scope": {
|
||
"type": "string",
|
||
"enum": [
|
||
"company_specific_accounting",
|
||
"generic_accounting",
|
||
"offtopic",
|
||
"unclear"
|
||
]
|
||
},
|
||
"entity_hints": {
|
||
"type": "array",
|
||
"items": {
|
||
"type": "string"
|
||
}
|
||
},
|
||
"account_hints": {
|
||
"type": "array",
|
||
"items": {
|
||
"type": "string"
|
||
}
|
||
},
|
||
"document_hints": {
|
||
"type": "array",
|
||
"items": {
|
||
"type": "string"
|
||
}
|
||
},
|
||
"register_hints": {
|
||
"type": "array",
|
||
"items": {
|
||
"type": "string"
|
||
}
|
||
},
|
||
"time_scope": {
|
||
"type": "object",
|
||
"additionalProperties": false,
|
||
"required": [
|
||
"type",
|
||
"value",
|
||
"confidence"
|
||
],
|
||
"properties": {
|
||
"type": {
|
||
"type": "string",
|
||
"enum": [
|
||
"explicit",
|
||
"inferred",
|
||
"missing"
|
||
]
|
||
},
|
||
"value": {
|
||
"type": [
|
||
"string",
|
||
"null"
|
||
]
|
||
},
|
||
"confidence": {
|
||
"type": "string",
|
||
"enum": [
|
||
"high",
|
||
"medium",
|
||
"low"
|
||
]
|
||
}
|
||
}
|
||
},
|
||
"flags": {
|
||
"type": "object",
|
||
"additionalProperties": false,
|
||
"required": [
|
||
"has_multi_entity_scope",
|
||
"asks_for_chain_explanation",
|
||
"asks_for_ranking_or_top",
|
||
"asks_for_period_summary",
|
||
"asks_for_rule_check",
|
||
"asks_for_anomaly_scan",
|
||
"asks_for_exact_object_trace",
|
||
"asks_for_evidence",
|
||
"mentions_period_close_context"
|
||
],
|
||
"properties": {
|
||
"has_multi_entity_scope": {
|
||
"type": "boolean"
|
||
},
|
||
"asks_for_chain_explanation": {
|
||
"type": "boolean"
|
||
},
|
||
"asks_for_ranking_or_top": {
|
||
"type": "boolean"
|
||
},
|
||
"asks_for_period_summary": {
|
||
"type": "boolean"
|
||
},
|
||
"asks_for_rule_check": {
|
||
"type": "boolean"
|
||
},
|
||
"asks_for_anomaly_scan": {
|
||
"type": "boolean"
|
||
},
|
||
"asks_for_exact_object_trace": {
|
||
"type": "boolean"
|
||
},
|
||
"asks_for_evidence": {
|
||
"type": "boolean"
|
||
},
|
||
"mentions_period_close_context": {
|
||
"type": "boolean"
|
||
}
|
||
}
|
||
},
|
||
"candidate_labels": {
|
||
"type": "array",
|
||
"items": {
|
||
"type": "string",
|
||
"enum": [
|
||
"heavy_analytical",
|
||
"cross_entity",
|
||
"drilldown_explain",
|
||
"rule_based_account_control",
|
||
"anomaly_probe",
|
||
"period_close_risk",
|
||
"ambiguous_human_query",
|
||
"simple_factual"
|
||
]
|
||
}
|
||
},
|
||
"confidence": {
|
||
"type": "string",
|
||
"enum": [
|
||
"high",
|
||
"medium",
|
||
"low"
|
||
]
|
||
}
|
||
}
|
||
}
|
||
},
|
||
"discarded_fragments": {
|
||
"type": "array",
|
||
"items": {
|
||
"type": "object",
|
||
"additionalProperties": false,
|
||
"required": [
|
||
"raw_fragment_text",
|
||
"reason"
|
||
],
|
||
"properties": {
|
||
"raw_fragment_text": {
|
||
"type": "string",
|
||
"minLength": 1
|
||
},
|
||
"reason": {
|
||
"type": "string",
|
||
"minLength": 1
|
||
}
|
||
}
|
||
}
|
||
},
|
||
"global_notes": {
|
||
"type": "object",
|
||
"additionalProperties": false,
|
||
"required": [
|
||
"needs_clarification",
|
||
"clarification_reason"
|
||
],
|
||
"properties": {
|
||
"needs_clarification": {
|
||
"type": "boolean"
|
||
},
|
||
"clarification_reason": {
|
||
"type": [
|
||
"string",
|
||
"null"
|
||
]
|
||
}
|
||
}
|
||
}
|
||
}
|
||
},
|
||
"strict": true
|
||
},
|
||
"verbosity": "medium"
|
||
},
|
||
"tool_choice": "auto",
|
||
"tools": [],
|
||
"top_logprobs": 0,
|
||
"top_p": 1,
|
||
"truncation": "disabled",
|
||
"usage": {
|
||
"input_tokens": 2416,
|
||
"input_tokens_details": {
|
||
"cached_tokens": 0
|
||
},
|
||
"output_tokens": 392,
|
||
"output_tokens_details": {
|
||
"reasoning_tokens": 0
|
||
},
|
||
"total_tokens": 2808
|
||
},
|
||
"user": null,
|
||
"metadata": {}
|
||
},
|
||
"parsed_normalized_json": {
|
||
"schema_version": "normalized_query_v2",
|
||
"user_message_raw": "Есть ли такие зависшие авансы, которые уже давно надо было либо закрыть, либо хотя бы перепроверить руками?",
|
||
"message_in_scope": true,
|
||
"scope_confidence": "high",
|
||
"contains_multiple_tasks": false,
|
||
"fragments": [
|
||
{
|
||
"fragment_id": "F1",
|
||
"raw_fragment_text": "зависшие авансы, которые уже давно надо было либо закрыть, либо хотя бы перепроверить руками",
|
||
"normalized_fragment_text": "проверка зависших авансов на закрытие или перепроверку",
|
||
"domain_relevance": "in_scope",
|
||
"business_scope": "company_specific_accounting",
|
||
"entity_hints": [],
|
||
"account_hints": [
|
||
"51",
|
||
"60"
|
||
],
|
||
"document_hints": [],
|
||
"register_hints": [],
|
||
"time_scope": {
|
||
"type": "missing",
|
||
"value": null,
|
||
"confidence": "low"
|
||
},
|
||
"flags": {
|
||
"has_multi_entity_scope": false,
|
||
"asks_for_chain_explanation": false,
|
||
"asks_for_ranking_or_top": false,
|
||
"asks_for_period_summary": false,
|
||
"asks_for_rule_check": false,
|
||
"asks_for_anomaly_scan": true,
|
||
"asks_for_exact_object_trace": false,
|
||
"asks_for_evidence": false,
|
||
"mentions_period_close_context": false
|
||
},
|
||
"candidate_labels": [
|
||
"anomaly_probe"
|
||
],
|
||
"confidence": "medium"
|
||
}
|
||
],
|
||
"discarded_fragments": [],
|
||
"global_notes": {
|
||
"needs_clarification": true,
|
||
"clarification_reason": "не указаны конкретные сроки для проверки зависших авансов"
|
||
}
|
||
},
|
||
"validation_result": {
|
||
"passed": true,
|
||
"errors": []
|
||
},
|
||
"route_hint_summary": {
|
||
"mode": "deterministic_v2",
|
||
"message_in_scope": true,
|
||
"scope_confidence": "high",
|
||
"planner": {
|
||
"total_fragments": 1,
|
||
"in_scope_fragments": 1,
|
||
"out_of_scope_fragments": 0,
|
||
"discarded_fragments": 0,
|
||
"contains_multiple_tasks": false
|
||
},
|
||
"decisions": [
|
||
{
|
||
"fragment_id": "F1",
|
||
"domain_relevance": "in_scope",
|
||
"business_scope": "company_specific_accounting",
|
||
"candidate_labels": [
|
||
"anomaly_probe"
|
||
],
|
||
"decision_flags": {
|
||
"has_multi_entity_scope": false,
|
||
"asks_for_chain_explanation": false,
|
||
"asks_for_ranking_or_top": false,
|
||
"asks_for_period_summary": false,
|
||
"asks_for_rule_check": false,
|
||
"asks_for_anomaly_scan": true,
|
||
"asks_for_exact_object_trace": false,
|
||
"asks_for_evidence": false,
|
||
"mentions_period_close_context": false
|
||
},
|
||
"route": "store_feature_risk",
|
||
"reason": "Anomaly scan without heavy ranking or causal chain."
|
||
}
|
||
],
|
||
"fallback": {
|
||
"type": "clarification",
|
||
"message": "Могу проверить это в контуре компании, но нужно уточнить период, документ, счет или участок учета."
|
||
}
|
||
},
|
||
"route_hint": "store_feature_risk",
|
||
"confidence": "high",
|
||
"usage": {
|
||
"input_tokens": 2416,
|
||
"output_tokens": 392,
|
||
"total_tokens": 2808
|
||
},
|
||
"latency_ms": 14121,
|
||
"eval_label": "eval-aOEoelkQb2",
|
||
"eval_mode": "single-pass-strict",
|
||
"request_count_for_case": 1
|
||
} |