NODEDC_1C/llm_normalizer/backend/src/services/assistantMcpDiscoveryTurnInputAdapter.ts

import type { AssistantMcpDiscoveryTurnMeaningRef } from "./assistantMcpDiscoveryPolicy";

export const ASSISTANT_MCP_DISCOVERY_TURN_INPUT_SCHEMA_VERSION =
  "assistant_mcp_discovery_turn_input_v1" as const;

export type AssistantMcpDiscoveryTurnInputStatus = "ready" | "needs_more_context" | "not_applicable";
export type AssistantMcpDiscoveryTurnInputSource =
  | "assistant_turn_meaning"
  | "predecompose_contract"
  | "raw_text"
  | "none";

export interface BuildAssistantMcpDiscoveryTurnInputAdapterInput {
  assistantTurnMeaning?: Record<string, unknown> | null;
  predecomposeContract?: Record<string, unknown> | null;
  userMessage?: string | null;
  effectiveMessage?: string | null;
}

export interface AssistantMcpDiscoveryTurnInputContract {
  schema_version: typeof ASSISTANT_MCP_DISCOVERY_TURN_INPUT_SCHEMA_VERSION;
  policy_owner: "assistantMcpDiscoveryTurnInputAdapter";
  adapter_status: AssistantMcpDiscoveryTurnInputStatus;
  should_run_discovery: boolean;
  semantic_data_need: string | null;
  turn_meaning_ref: AssistantMcpDiscoveryTurnMeaningRef | null;
  source_signal: AssistantMcpDiscoveryTurnInputSource;
  reason_codes: string[];
}

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 toNonEmptyString(value: unknown): string | null {
  if (value === null || value === undefined) {
    return null;
  }
  const text = String(value).trim();
  return text.length > 0 ? text : null;
}

function normalizeReasonCode(value: string): string | null {
  const normalized = value
    .trim()
    .replace(/[^\p{L}\p{N}_.:-]+/gu, "_")
    .replace(/^_+|_+$/g, "")
    .toLowerCase();
  return normalized.length > 0 ? normalized.slice(0, 120) : null;
}

function pushReason(target: string[], value: string): void {
  const normalized = normalizeReasonCode(value);
  if (normalized && !target.includes(normalized)) {
    target.push(normalized);
  }
}

function pushUnique(target: string[], value: unknown): void {
  const text = toNonEmptyString(value);
  if (text && !target.includes(text)) {
    target.push(text);
  }
}

function compactLower(value: unknown): string {
  return String(value ?? "")
    .toLowerCase()
    .replace(/\s+/g, " ")
    .trim();
}

function candidateValue(value: unknown): string | null {
  const direct = toNonEmptyString(value);
  if (direct && direct !== "[object Object]") {
    return direct;
  }
  const record = toRecordObject(value);
  if (!record) {
    return null;
  }
  return (
    toNonEmptyString(record.value) ??
    toNonEmptyString(record.name) ??
    toNonEmptyString(record.ref) ??
    toNonEmptyString(record.text)
  );
}

function collectEntityCandidates(value: unknown): string[] {
  const result: string[] = [];
  if (Array.isArray(value)) {
    for (const item of value) {
      pushUnique(result, candidateValue(item));
    }
    return result;
  }
  pushUnique(result, candidateValue(value));
  return result;
}

function collectPredecomposeEntities(predecompose: Record<string, unknown> | null): {
  counterparty: string | null;
  organization: string | null;
} {
  const entities = toRecordObject(predecompose?.entities);
  return {
    counterparty: toNonEmptyString(entities?.counterparty),
    organization: toNonEmptyString(entities?.organization)
  };
}

function collectDateScope(predecompose: Record<string, unknown> | null): string | null {
  const period = toRecordObject(predecompose?.period);
  const asOfDate = toNonEmptyString(period?.as_of_date);
  const periodFrom = toNonEmptyString(period?.period_from);
  const periodTo = toNonEmptyString(period?.period_to);
  if (asOfDate) {
    return asOfDate;
  }
  const yearFrom = periodFrom?.match(/^(\d{4})-01-01$/);
  const yearTo = periodTo?.match(/^(\d{4})-12-31$/);
  if (yearFrom && yearTo && yearFrom[1] === yearTo[1]) {
    return yearFrom[1];
  }
  if (periodFrom && periodTo) {
    return `${periodFrom}..${periodTo}`;
  }
  return periodFrom ?? periodTo ?? null;
}

function hasLifecycleSignal(text: string): boolean {
  return /(?:сколько\s+лет|как\s+давно|давно\s+ли|возраст|перв(?:ая|ый)\s+актив|когда\s+начал|когда\s+появ|lifecycle|activity\s+duration|business\s+age|how\s+long)/iu.test(
    text
  );
}

function semanticNeedFor(input: {
  domain: string | null;
  action: string | null;
  unsupported: string | null;
  lifecycleSignal: boolean;
}): string | null {
  const combined = compactLower(`${input.domain ?? ""} ${input.action ?? ""} ${input.unsupported ?? ""}`);
  if (input.lifecycleSignal || /(?:lifecycle|activity|duration|age)/iu.test(combined)) {
    return "counterparty lifecycle evidence";
  }
  if (/(?:turnover|revenue|payment|payout|value)/iu.test(combined)) {
    return "counterparty value-flow evidence";
  }
  if (/(?:document|documents|list_documents)/iu.test(combined)) {
    return "document evidence";
  }
  if (/(?:metadata|schema|catalog)/iu.test(combined)) {
    return "1C metadata evidence";
  }
  return null;
}

function shouldRunDiscovery(input: {
  unsupported: string | null;
  lifecycleSignal: boolean;
  semanticDataNeed: string | null;
  explicitIntentCandidate: string | null;
}): boolean {
  if (input.lifecycleSignal || input.unsupported) {
    return true;
  }
  if (!input.explicitIntentCandidate && input.semanticDataNeed) {
    return true;
  }
  return false;
}

export function buildAssistantMcpDiscoveryTurnInput(
  input: BuildAssistantMcpDiscoveryTurnInputAdapterInput
): AssistantMcpDiscoveryTurnInputContract {
  const assistantTurnMeaning = toRecordObject(input.assistantTurnMeaning);
  const predecomposeContract = toRecordObject(input.predecomposeContract);
  const predecomposeEntities = collectPredecomposeEntities(predecomposeContract);
  const reasonCodes: string[] = [];
  const rawText = compactLower(`${input.userMessage ?? ""} ${input.effectiveMessage ?? ""}`);
  const lifecycleSignal = hasLifecycleSignal(rawText);

  const rawDomain = toNonEmptyString(assistantTurnMeaning?.asked_domain_family);
  const rawAction = toNonEmptyString(assistantTurnMeaning?.asked_action_family);
  const unsupported = toNonEmptyString(assistantTurnMeaning?.unsupported_but_understood_family);
  const explicitIntentCandidate = toNonEmptyString(assistantTurnMeaning?.explicit_intent_candidate);
  const semanticDataNeed = semanticNeedFor({
    domain: rawDomain,
    action: rawAction,
    unsupported,
    lifecycleSignal
  });
  const entityCandidates = collectEntityCandidates(assistantTurnMeaning?.explicit_entity_candidates);
  pushUnique(entityCandidates, predecomposeEntities.counterparty);

  const turnMeaning: AssistantMcpDiscoveryTurnMeaningRef = {
    asked_domain_family: lifecycleSignal ? "counterparty_lifecycle" : rawDomain,
    asked_action_family: lifecycleSignal ? "activity_duration" : rawAction,
    explicit_entity_candidates: entityCandidates,
    explicit_organization_scope: predecomposeEntities.organization,
    explicit_date_scope: collectDateScope(predecomposeContract),
    unsupported_but_understood_family: unsupported ?? (lifecycleSignal ? "counterparty_lifecycle" : null),
    stale_replay_forbidden: Boolean(assistantTurnMeaning?.stale_replay_forbidden || unsupported || lifecycleSignal)
  };

  const cleanTurnMeaning: AssistantMcpDiscoveryTurnMeaningRef = {};
  if (toNonEmptyString(turnMeaning.asked_domain_family)) {
    cleanTurnMeaning.asked_domain_family = turnMeaning.asked_domain_family;
  }
  if (toNonEmptyString(turnMeaning.asked_action_family)) {
    cleanTurnMeaning.asked_action_family = turnMeaning.asked_action_family;
  }
  if ((turnMeaning.explicit_entity_candidates?.length ?? 0) > 0) {
    cleanTurnMeaning.explicit_entity_candidates = turnMeaning.explicit_entity_candidates;
  }
  if (toNonEmptyString(turnMeaning.explicit_organization_scope)) {
    cleanTurnMeaning.explicit_organization_scope = turnMeaning.explicit_organization_scope;
  }
  if (toNonEmptyString(turnMeaning.explicit_date_scope)) {
    cleanTurnMeaning.explicit_date_scope = turnMeaning.explicit_date_scope;
  }
  if (toNonEmptyString(turnMeaning.unsupported_but_understood_family)) {
    cleanTurnMeaning.unsupported_but_understood_family = turnMeaning.unsupported_but_understood_family;
  }
  if (turnMeaning.stale_replay_forbidden) {
    cleanTurnMeaning.stale_replay_forbidden = true;
  }

  const runDiscovery = shouldRunDiscovery({
    unsupported,
    lifecycleSignal,
    semanticDataNeed,
    explicitIntentCandidate
  });
  const hasTurnMeaning = Object.keys(cleanTurnMeaning).length > 0;
  const sourceSignal: AssistantMcpDiscoveryTurnInputSource = assistantTurnMeaning
    ? "assistant_turn_meaning"
    : predecomposeContract
      ? "predecompose_contract"
      : lifecycleSignal
        ? "raw_text"
        : "none";

  if (lifecycleSignal) {
    pushReason(reasonCodes, "mcp_discovery_lifecycle_signal_detected");
  }
  if (unsupported) {
    pushReason(reasonCodes, "mcp_discovery_unsupported_but_understood_turn");
  }
  if (predecomposeEntities.counterparty) {
    pushReason(reasonCodes, "mcp_discovery_counterparty_from_predecompose");
  }
  if (entityCandidates.length > 0) {
    pushReason(reasonCodes, "mcp_discovery_entity_scope_available");
  }
  if (!runDiscovery) {
    pushReason(reasonCodes, "mcp_discovery_not_applicable_for_supported_exact_turn");
  }
  if (runDiscovery && !hasTurnMeaning) {
    pushReason(reasonCodes, "mcp_discovery_turn_meaning_missing");
  }

  return {
    schema_version: ASSISTANT_MCP_DISCOVERY_TURN_INPUT_SCHEMA_VERSION,
    policy_owner: "assistantMcpDiscoveryTurnInputAdapter",
    adapter_status: !runDiscovery ? "not_applicable" : hasTurnMeaning ? "ready" : "needs_more_context",
    should_run_discovery: runDiscovery,
    semantic_data_need: runDiscovery ? semanticDataNeed : null,
    turn_meaning_ref: runDiscovery && hasTurnMeaning ? cleanTurnMeaning : null,
    source_signal: sourceSignal,
    reason_codes: reasonCodes
  };
}