Stage_04_Wave_15_Question_Type_Contract_First_Check_Relevance

2026-03-28 14:09:08 +03:00
parent 8b84f5e989
commit 014ff65188
8 changed files with 952 additions and 115 deletions
@@ -1,4 +1,4 @@
-import { describe, expect, it } from "vitest";
+import { describe, expect, it } from "vitest";
 import { composeAssistantAnswer } from "../src/services/answerComposer";
 import type { AnswerGroundingCheck, RequirementCoverageReport, UnifiedRetrievalResult } from "../src/types/assistant";
 import type { ProblemUnit } from "../src/types/stage2ProblemUnits";
@@ -116,7 +116,7 @@ function buildRetrieval(input: {
            {
              source_entity: "Document",
              source_id: "DOC-1",
-              display_name: "Счет № 4 от 07.07.20",
+              display_name: "РЎС‡РµС‚ в„– 4 РѕС‚ 07.07.20",
              account_context: accountScope,
              graph_domain_scope: domainScope,
              relation_pattern_hits: relationPatterns,
@@ -177,7 +177,7 @@ function buildRetrieval(input: {
              limitation: null,
              payload: {
                notes: input.notes ?? [],
-                contract: "договор № 01/19-ПТ",
+                contract: "РґРѕРіРѕРІРѕСЂ в„– 01/19-РџРў",
                amount: "276 873,60",
                date: "07.07.20"
              }
@@ -245,21 +245,21 @@ function composeCase(input: {
    focusDomainHint: input.focusDomainHint,
    questionTypeHint: input.questionType,
    companyAnchors: {
-      contract_numbers: ["договор № 01/19-ПТ"],
-      document_numbers: ["документ № 4"],
+      contract_numbers: ["РґРѕРіРѕРІРѕСЂ в„– 01/19-РџРў"],
+      document_numbers: ["РґРѕРєСѓРјРµРЅС‚ в„– 4"],
      dates: ["07.07.20", "13.07.20"],
      amounts: ["276 873,60"],
      accounts: ["62.02"],
-      periods: ["июль 2020"],
+      periods: ["РёСЋР»СЊ 2020"],
      document_types: ["payment", "invoice"],
      all: [
-        "договор № 01/19-ПТ",
-        "документ № 4",
+        "РґРѕРіРѕРІРѕСЂ в„– 01/19-РџРў",
+        "РґРѕРєСѓРјРµРЅС‚ в„– 4",
        "07.07.20",
        "13.07.20",
        "276 873,60",
        "account:62.02",
-        "period:июль 2020"
+        "period:РёСЋР»СЊ 2020"
      ]
    },
    enableAnswerPolicyV11: true,
@@ -279,7 +279,7 @@ describe("wave13 domain fit + question-type fit + company-anchor grounding", ()
    });
    const output = composeCase({
      userMessage:
-        "Почему по договору № 01/19-ПТ от 09.01.2019 оплата 276 873,60 есть, а 62.01/62.02 все равно не сходятся?",
+        "РџРѕС‡РµРјСѓ РїРѕ РґРѕРіРѕРІРѕСЂСѓ в„– 01/19-РџРў РѕС‚ 09.01.2019 РѕРїР»Р°С‚Р° 276 873,60 РµСЃС‚СЊ, Р° 62.01/62.02 РІСЃРµ СЂР°РІРЅРѕ РЅРµ СЃС…РѕРґСЏС‚СЃСЏ?",
      questionType: "why_breaks",
      focusDomainHint: "vat_document_register_book",
      retrievalResults: [
@@ -294,8 +294,8 @@ describe("wave13 domain fit + question-type fit + company-anchor grounding", ()
      ]
    });

-    expect(output.assistant_reply).toMatch(/расчет|62\.01|62\.02|зачет|зачёт/i);
-    expect(output.assistant_reply).not.toMatch(/переход от документа к регистру и книге|цепочке ндс/i);
+    expect(output.assistant_reply).toMatch(/СЂР°СЃС‡РµС‚|62\.01|62\.02|Р·Р°С‡РµС‚|Р·Р°С‡С‘С‚/i);
+    expect(output.assistant_reply).not.toMatch(/РїРµСЂРµС…РѕРґ РѕС‚ РґРѕРєСѓРјРµРЅС‚Р° Рє СЂРµРіРёСЃС‚СЂСѓ Рё РєРЅРёРіРµ|С†РµРїРѕС‡РєРµ РЅРґСЃ/i);
  });

  it("question_type_where_break_is_must_produce_localization_style_line", () => {
@@ -308,13 +308,13 @@ describe("wave13 domain fit + question-type fit + company-anchor grounding", ()
    });
    const output = composeCase({
      userMessage:
-        "Где именно разрыв по договору № 01/19-ПТ: в договоре, объекте расчетов или в связке документов?",
+        "Р“РґРµ РёРјРµРЅРЅРѕ СЂР°Р·СЂС‹РІ РїРѕ РґРѕРіРѕРІРѕСЂСѓ в„– 01/19-РџРў: РІ РґРѕРіРѕРІРѕСЂРµ, РѕР±СЉРµРєС‚Рµ СЂР°СЃС‡РµС‚РѕРІ РёР»Рё РІ СЃРІСЏР·РєРµ РґРѕРєСѓРјРµРЅС‚РѕРІ?",
      questionType: "where_break_is",
      focusDomainHint: "settlements_60_62",
      retrievalResults: [buildRetrieval({ requirementId: "R1", status: "ok", units: [settlementUnit] })]
    });

-    expect(output.assistant_reply).toMatch(/локализ|узел разрыва|где именно/i);
+    expect(output.assistant_reply).toMatch(/локализ|узел разрыва|где/i);
  });

  it("question_type_prove_or_guess_must_explicitly_separate_proven_vs_hypothesis", () => {
@@ -327,7 +327,7 @@ describe("wave13 domain fit + question-type fit + company-anchor grounding", ()
    });
    const output = composeCase({
      userMessage:
-        "По НДС это доказано по данным или это только гипотеза? На чем основано утверждение?",
+        "РџРѕ РќР”РЎ СЌС‚Рѕ РґРѕРєР°Р·Р°РЅРѕ РїРѕ РґР°РЅРЅС‹Рј РёР»Рё СЌС‚Рѕ С‚РѕР»СЊРєРѕ РіРёРїРѕС‚РµР·Р°? РќР° С‡РµРј РѕСЃРЅРѕРІР°РЅРѕ СѓС‚РІРµСЂР¶РґРµРЅРёРµ?",
      questionType: "prove_or_guess",
      focusDomainHint: "vat_document_register_book",
      retrievalResults: [
@@ -346,26 +346,26 @@ describe("wave13 domain fit + question-type fit + company-anchor grounding", ()
      }
    });

-    expect(output.assistant_reply).toMatch(/доказан|гипотез|ограничени/i);
+    expect(output.assistant_reply).toMatch(/доказ|гипотез|огранич/i);
  });

  it("anchor_usage_lines_must_be_present_when_company_anchors_are_used", () => {
    const output = composeCase({
      userMessage:
-        "Оплата по счету № 4 от 07.07.20 на 276 873,60 пришла 13 июля. Что проверить первым по 62.02?",
+        "РћРїР»Р°С‚Р° РїРѕ СЃС‡РµС‚Сѓ в„– 4 РѕС‚ 07.07.20 РЅР° 276 873,60 РїСЂРёС€Р»Р° 13 РёСЋР»СЏ. Р§С‚Рѕ РїСЂРѕРІРµСЂРёС‚СЊ РїРµСЂРІС‹Рј РїРѕ 62.02?",
      questionType: "what_to_check_first",
      focusDomainHint: "settlements_60_62",
      retrievalResults: [buildRetrieval({ requirementId: "R1", status: "ok" })]
    });

-    expect(output.assistant_reply).toMatch(/якоря вопроса/i);
-    expect(output.assistant_reply).toMatch(/договор|07\.07\.20|276 873,60|62\.02/i);
+    expect(output.assistant_reply).toMatch(/якоря вопроса|в опоре использованы|якоря из вопроса/i);
+    expect(output.assistant_reply).toMatch(/РґРѕРіРѕРІРѕСЂ|07\.07\.20|276 873,60|62\.02/i);
  });

  it("anchor_usage_must_be_honest_when_part_of_anchors_not_confirmed", () => {
    const output = composeCase({
      userMessage:
-        "Почему по договору № 01/19-ПТ не сходится 62.02 в июле 2020, если была оплата 276 873,60?",
+        "РџРѕС‡РµРјСѓ РїРѕ РґРѕРіРѕРІРѕСЂСѓ в„– 01/19-РџРў РЅРµ СЃС…РѕРґРёС‚СЃСЏ 62.02 РІ РёСЋР»Рµ 2020, РµСЃР»Рё Р±С‹Р»Р° РѕРїР»Р°С‚Р° 276 873,60?",
      questionType: "what_is_it_grounded_on",
      focusDomainHint: "settlements_60_62",
      retrievalResults: [
@@ -380,19 +380,19 @@ describe("wave13 domain fit + question-type fit + company-anchor grounding", ()
      ]
    });

-    expect(output.assistant_reply).toMatch(/без прямого подтверждения|ограничени/i);
+    expect(output.assistant_reply).toMatch(/без прямого подтверждения|огранич/i);
  });

  it("answers_for_different_question_types_must_not_collapse_to_same_generic_pattern", () => {
    const baseRetrieval = [buildRetrieval({ requirementId: "R1", status: "ok" })];
    const whereOutput = composeCase({
-      userMessage: "Где именно разрыв по 62.01/62.02?",
+      userMessage: "Р“РґРµ РёРјРµРЅРЅРѕ СЂР°Р·СЂС‹РІ РїРѕ 62.01/62.02?",
      questionType: "where_break_is",
      focusDomainHint: "settlements_60_62",
      retrievalResults: baseRetrieval
    });
    const checkFirstOutput = composeCase({
-      userMessage: "Что проверить первым по 62.01/62.02?",
+      userMessage: "Р§С‚Рѕ РїСЂРѕРІРµСЂРёС‚СЊ РїРµСЂРІС‹Рј РїРѕ 62.01/62.02?",
      questionType: "what_to_check_first",
      focusDomainHint: "settlements_60_62",
      retrievalResults: baseRetrieval
@@ -400,6 +400,9 @@ describe("wave13 domain fit + question-type fit + company-anchor grounding", ()

    expect(whereOutput.assistant_reply).not.toEqual(checkFirstOutput.assistant_reply);
    expect(whereOutput.assistant_reply).toMatch(/локализ|разрыв/i);
-    expect(checkFirstOutput.assistant_reply).toMatch(/первый маршрут проверки|начните с первого пункта/i);
+    expect(checkFirstOutput.assistant_reply).toMatch(
+      /маршрут первых проверок|проверьте пункты по порядку|начните с первого пункта/i
+    );
  });
 });
+
@@ -0,0 +1,334 @@
+import { describe, expect, it } from "vitest";
+import { composeAssistantAnswer } from "../src/services/answerComposer";
+import type { AnswerGroundingCheck, RequirementCoverageReport, UnifiedRetrievalResult } from "../src/types/assistant";
+import type { ProblemUnit } from "../src/types/stage2ProblemUnits";
+import type { QuestionTypeClass } from "../src/services/questionTypeResolver";
+
+function buildRouteSummary() {
+  return {
+    mode: "deterministic_v2" as const,
+    message_in_scope: true,
+    scope_confidence: "high" as const,
+    planner: {
+      total_fragments: 1,
+      in_scope_fragments: 1,
+      out_of_scope_fragments: 0,
+      discarded_fragments: 0,
+      contains_multiple_tasks: false
+    },
+    decisions: [],
+    fallback: {
+      type: "none" as const,
+      message: null
+    }
+  };
+}
+
+function buildCoverage(input?: Partial<RequirementCoverageReport>): RequirementCoverageReport {
+  return {
+    requirements_total: 1,
+    requirements_covered: 1,
+    requirements_uncovered: [],
+    requirements_partially_covered: [],
+    clarification_needed_for: [],
+    out_of_scope_requirements: [],
+    ...input
+  };
+}
+
+function buildGrounding(input?: Partial<AnswerGroundingCheck>): AnswerGroundingCheck {
+  return {
+    status: "grounded",
+    route_subject_match: true,
+    missing_requirements: [],
+    reasons: [],
+    why_included_summary: ["wave15-test"],
+    selection_reason_summary: ["wave15-test"],
+    ...input
+  };
+}
+
+function buildProblemUnit(input: {
+  id: string;
+  type: ProblemUnit["problem_unit_type"];
+  account: string;
+  defect: string;
+  lifecycleDomain: ProblemUnit["lifecycle_domain"];
+}): ProblemUnit {
+  return {
+    schema_version: "problem_unit_v0_1",
+    problem_unit_id: input.id,
+    problem_unit_type: input.type,
+    title: "Wave15 problem unit",
+    mechanism_summary: `Mechanism candidate: ${input.defect}.`,
+    business_defect_class: input.defect,
+    severity: {
+      score: 0.74,
+      grade: "high"
+    },
+    confidence: {
+      score: 0.66,
+      grade: "medium"
+    },
+    lifecycle_domain: input.lifecycleDomain,
+    affected_entities: ["Document:DOC-1"],
+    affected_documents: ["Document:DOC-1"],
+    affected_postings: ["Posting:POST-1"],
+    affected_accounts: [input.account],
+    affected_counterparties: ["Counterparty:CP-1"],
+    affected_contracts: ["Contract:CTR-1"],
+    failed_expected_edge: input.defect,
+    period_impact: {
+      is_period_sensitive: true,
+      impact_class: "close_risk"
+    },
+    evidence_pack: ["cand-1"],
+    entity_backlinks: [{ entity: "Document", id: "DOC-1" }],
+    snapshot_limitations: []
+  };
+}
+
+function buildRetrieval(input: {
+  requirementId: string;
+  status: UnifiedRetrievalResult["status"];
+  units?: ProblemUnit[];
+  accountScope?: string[];
+  domainScope?: string[];
+  relationPatterns?: string[];
+}): UnifiedRetrievalResult {
+  const units = input.units ?? [];
+  const accountScope = input.accountScope ?? ["60", "62"];
+  const domainScope = input.domainScope ?? ["customer_settlement"];
+  const relationPatterns = input.relationPatterns ?? ["payment_to_settlement"];
+
+  return {
+    fragment_id: `F-${input.requirementId}`,
+    requirement_ids: [input.requirementId],
+    route: "hybrid_store_plus_live",
+    status: input.status,
+    result_type: "chain",
+    items:
+      input.status === "empty"
+        ? []
+        : [
+            {
+              source_entity: "Document",
+              source_id: "DOC-1",
+              display_name: "Счет №4 от 07.07.20",
+              account_context: accountScope,
+              graph_domain_scope: domainScope,
+              relation_pattern_hits: relationPatterns,
+              period: "2020-07",
+              amount: "276 873,60"
+            }
+          ],
+    summary: {
+      broad_query_detected: false,
+      broad_result_flag: false,
+      minimum_evidence_failed: false,
+      degraded_to: null,
+      narrowing_strength: "strong",
+      semantic_profile: {
+        account_scope: accountScope,
+        domain_scope: domainScope,
+        relation_patterns: relationPatterns,
+        period_scope: {
+          from: "2020-07-01",
+          to: "2020-07-31",
+          granularity: "month"
+        }
+      }
+    },
+    evidence:
+      input.status === "empty"
+        ? []
+        : [
+            {
+              evidence_id: `ev-${input.requirementId}`,
+              claim_ref: `requirement:${input.requirementId}`,
+              source_type: "retrieval_item",
+              source_ref: {
+                schema_version: "evidence_source_ref_v1",
+                namespace: "snapshot_2020_07",
+                entity: "document",
+                id: "DOC-1",
+                period: "2020-07",
+                canonical_ref: "evidence_source_ref_v1|snapshot_2020_07|document|doc-1|2020-07"
+              },
+              pointer: {
+                fragment_id: `F-${input.requirementId}`,
+                route: "hybrid_store_plus_live",
+                source: {
+                  namespace: "snapshot_2020_07",
+                  entity: "document",
+                  id: "DOC-1",
+                  period: "2020-07"
+                },
+                locator: {
+                  field_path: "risk_score",
+                  item_index: 0
+                }
+              },
+              evidence_kind: "mechanism_link",
+              mechanism_note: relationPatterns[0],
+              confidence: "medium",
+              limitation: null,
+              payload: {
+                notes: ["wave15"],
+                contract: "договор № 01/19-ПТ",
+                amount: "276 873,60",
+                date: "07.07.20"
+              }
+            }
+          ],
+    candidate_evidence: [],
+    problem_units: units,
+    problem_unit_summary:
+      units.length > 0
+        ? {
+            schema_version: "problem_unit_summary_v0_1",
+            units_total: units.length,
+            duplicate_collapses: 0,
+            unit_types: units.map((unit) => unit.problem_unit_type),
+            type_distribution: {
+              [units[0]?.problem_unit_type ?? "broken_chain_segment"]: units.length
+            },
+            severity_distribution: {
+              low: 0,
+              medium: 0,
+              high: units.length
+            },
+            confidence_distribution: {
+              low: 0,
+              medium: units.length,
+              high: 0
+            },
+            primary_unit_type: units[0]?.problem_unit_type ?? null
+          }
+        : null,
+    why_included: ["wave15-test"],
+    selection_reason: ["wave15-test"],
+    risk_factors: ["wave15"],
+    business_interpretation: ["wave15"],
+    confidence: "medium",
+    limitations: [],
+    errors: []
+  };
+}
+
+function composeCase(input: {
+  userMessage: string;
+  questionType: QuestionTypeClass;
+  focusDomainHint: string | null;
+  retrievalResults: UnifiedRetrievalResult[];
+  coverage?: Partial<RequirementCoverageReport>;
+  grounding?: Partial<AnswerGroundingCheck>;
+}) {
+  return composeAssistantAnswer({
+    userMessage: input.userMessage,
+    routeSummary: buildRouteSummary(),
+    retrievalResults: input.retrievalResults,
+    requirements: [
+      {
+        requirement_id: "R1",
+        source_fragment_id: "F-R1",
+        requirement_text: "Wave15 requirement",
+        subject_tokens: [],
+        status: "covered",
+        route: "hybrid_store_plus_live"
+      }
+    ],
+    coverageReport: buildCoverage(input.coverage),
+    groundingCheck: buildGrounding(input.grounding),
+    focusDomainHint: input.focusDomainHint,
+    questionTypeHint: input.questionType,
+    companyAnchors: {
+      contract_numbers: ["договор № 01/19-ПТ"],
+      document_numbers: ["документ № 4"],
+      dates: ["07.07.20"],
+      amounts: ["276 873,60"],
+      accounts: ["62.02"],
+      periods: ["июль 2020"],
+      document_types: ["payment"],
+      all: ["договор № 01/19-ПТ", "документ № 4", "07.07.20", "276 873,60", "account:62.02", "period:июль 2020"]
+    },
+    enableAnswerPolicyV11: true,
+    enableProblemCentricAnswerV1: true,
+    enableLifecycleAnswerV1: true
+  });
+}
+
+describe("wave15 question-type contract + first-check relevance", () => {
+  const settlementUnit = buildProblemUnit({
+    id: "pu-settlement-wave15",
+    type: "broken_chain_segment",
+    account: "62.02",
+    defect: "failed_edge:payment_to_settlement",
+    lifecycleDomain: "customer_settlement"
+  });
+
+  it("prove_or_guess answer keeps proof-vs-hypothesis framing", () => {
+    const output = composeCase({
+      userMessage: "Это по расчетам доказано или пока гипотеза?",
+      questionType: "prove_or_guess",
+      focusDomainHint: "settlements_60_62",
+      retrievalResults: [buildRetrieval({ requirementId: "R1", status: "ok", units: [settlementUnit] })],
+      grounding: {
+        status: "partial",
+        reasons: ["Mechanism is unresolved for part of the evidence."]
+      }
+    });
+
+    expect(output.assistant_reply).toMatch(/доказ|гипотез|ограничен/i);
+  });
+
+  it("grounded_on answer is basis-oriented, not generic mechanism dump", () => {
+    const output = composeCase({
+      userMessage: "На чем это основано по 62.02?",
+      questionType: "what_is_it_grounded_on",
+      focusDomainHint: "settlements_60_62",
+      retrievalResults: [buildRetrieval({ requirementId: "R1", status: "ok", units: [settlementUnit] })]
+    });
+
+    expect(output.assistant_reply).toMatch(/основан|опор|документ|регистр|60\/62\/76/i);
+  });
+
+  it("chains question provides complete-vs-incomplete framing", () => {
+    const output = composeCase({
+      userMessage: "Какие цепочки завершены, а какие нет по 62.01/62.02?",
+      questionType: "which_chains_are_complete_vs_incomplete",
+      focusDomainHint: "settlements_60_62",
+      retrievalResults: [buildRetrieval({ requirementId: "R1", status: "ok", units: [settlementUnit] })],
+      coverage: {
+        requirements_covered: 0,
+        requirements_partially_covered: ["R1"],
+        requirements_uncovered: []
+      }
+    });
+
+    expect(output.assistant_reply).toMatch(/цепочк|подтвержден|частично|не подтвержден/i);
+  });
+
+  it("where_break_is answer localizes break node", () => {
+    const output = composeCase({
+      userMessage: "Где именно разрыв по расчетной цепочке?",
+      questionType: "where_break_is",
+      focusDomainHint: "settlements_60_62",
+      retrievalResults: [buildRetrieval({ requirementId: "R1", status: "ok", units: [settlementUnit] })]
+    });
+
+    expect(output.assistant_reply).toMatch(/узел разрыва|локализ|где/i);
+  });
+
+  it("what_to_check_first answer gives short operational route", () => {
+    const output = composeCase({
+      userMessage: "Что проверить первым по 62.01/62.02?",
+      questionType: "what_to_check_first",
+      focusDomainHint: "settlements_60_62",
+      retrievalResults: [buildRetrieval({ requirementId: "R1", status: "ok", units: [settlementUnit] })]
+    });
+
+    expect(output.assistant_reply).toMatch(/что проверить первым/i);
+    expect(output.assistant_reply).toMatch(/договор|регистр|60\/62\/76/i);
+  });
+});
@@ -0,0 +1,34 @@
+import { describe, expect, it } from "vitest";
+import { resolveQuestionType } from "../src/services/questionTypeResolver";
+
+describe("questionTypeResolver", () => {
+  it("resolves what_to_check_first for operational check phrasing", () => {
+    expect(resolveQuestionType("Что проверить первым по 62.01/62.02?"))
+      .toBe("what_to_check_first");
+  });
+
+  it("resolves where_break_is for localization phrasing", () => {
+    expect(resolveQuestionType("Где именно разрыв: в договоре или в связке документ -> регистр?"))
+      .toBe("where_break_is");
+  });
+
+  it("resolves prove_or_guess without collapsing to generic yes/no", () => {
+    expect(resolveQuestionType("Это доказано или пока только гипотеза?"))
+      .toBe("prove_or_guess");
+  });
+
+  it("resolves grounded_on when question asks for evidence basis", () => {
+    expect(resolveQuestionType("На чем это основано и чем подтверждается вывод?"))
+      .toBe("what_is_it_grounded_on");
+  });
+
+  it("resolves chains completeness classification questions", () => {
+    expect(resolveQuestionType("Какие цепочки подтверждены, а какие не завершены?"))
+      .toBe("which_chains_are_complete_vs_incomplete");
+  });
+
+  it("falls back to why_breaks for generic why-question", () => {
+    expect(resolveQuestionType("Почему не сходится 62.01/62.02?"))
+      .toBe("why_breaks");
+  });
+});