import request from "supertest";
import { afterEach, describe, expect, it, vi } from "vitest";

const FLAG_KEYS = [
  "FEATURE_ASSISTANT_ACCOUNTANT_EVAL_V1",
  "FEATURE_ASSISTANT_ANSWER_POLICY_V11",
  "FEATURE_ASSISTANT_BROAD_GUARD_V1",
  "FEATURE_ASSISTANT_MIN_EVIDENCE_GATE_V1",
  "FEATURE_ASSISTANT_ANTI_GENERIC_RANKING_GUARD_V1",
  "FEATURE_ASSISTANT_INVESTIGATION_STATE_V1",
  "FEATURE_ASSISTANT_STATE_FOLLOWUP_BINDING_V1",
  "FEATURE_ASSISTANT_PROBLEM_UNITS_V1",
  "FEATURE_ASSISTANT_PROBLEM_CENTRIC_ANSWER_V1",
  "FEATURE_ASSISTANT_PROBLEM_UNIT_CONTINUITY_V1",
  "FEATURE_ASSISTANT_STAGE2_EVAL_V1"
] as const;

const ORIGINAL_FLAGS: Record<string, string | undefined> = Object.fromEntries(
  FLAG_KEYS.map((key) => [key, process.env[key]])
);

function restoreFlags(): void {
  for (const key of FLAG_KEYS) {
    const original = ORIGINAL_FLAGS[key];
    if (original === undefined) {
      delete process.env[key];
    } else {
      process.env[key] = original;
    }
  }
}

async function createAppWithFlags(flags: {
  accountantEval: "0" | "1";
  answerPolicy: "0" | "1";
}): Promise<import("express").Express> {
  process.env.FEATURE_ASSISTANT_ACCOUNTANT_EVAL_V1 = flags.accountantEval;
  process.env.FEATURE_ASSISTANT_ANSWER_POLICY_V11 = flags.answerPolicy;
  process.env.FEATURE_ASSISTANT_BROAD_GUARD_V1 = "1";
  process.env.FEATURE_ASSISTANT_MIN_EVIDENCE_GATE_V1 = "1";
  process.env.FEATURE_ASSISTANT_ANTI_GENERIC_RANKING_GUARD_V1 = "1";
  process.env.FEATURE_ASSISTANT_INVESTIGATION_STATE_V1 = "1";
  process.env.FEATURE_ASSISTANT_STATE_FOLLOWUP_BINDING_V1 = "1";
  process.env.FEATURE_ASSISTANT_PROBLEM_UNITS_V1 = "0";
  process.env.FEATURE_ASSISTANT_PROBLEM_CENTRIC_ANSWER_V1 = "0";
  process.env.FEATURE_ASSISTANT_PROBLEM_UNIT_CONTINUITY_V1 = "0";
  process.env.FEATURE_ASSISTANT_STAGE2_EVAL_V1 = "0";
  vi.resetModules();
  const { createApp } = await import("../src/server");
  return createApp();
}

describe.sequential("assistant Stage 1 eval harness", () => {
  afterEach(() => {
    restoreFlags();
    vi.resetModules();
  });

  it("runs assistant_stage1 harness and returns raw metrics + rubric bands", async () => {
    const app = await createAppWithFlags({
      accountantEval: "1",
      answerPolicy: "1"
    });

    const response = await request(app).post("/api/eval/run").send({
      eval_target: "assistant_stage1",
      useMock: true,
      mode: "single-pass-strict",
      caseSetFile: "assistant_stage1_canonical_v0_1.json",
      normalizeConfig: {
        promptVersion: "normalizer_v2_0_2"
      }
    });

    expect(response.status).toBe(200);
    expect(response.body.ok).toBe(true);
    expect(response.body.report?.eval_target).toBe("assistant_stage1");
    expect(response.body.report?.metrics?.raw).toBeTruthy();
    const rawMetricKeys = Object.keys(response.body.report?.metrics?.raw ?? {});
    expect(rawMetricKeys).toEqual([
      "retrieval_differentiation_rate",
      "generic_explanation_rate",
      "accountant_actionability_score",
      "false_confidence_rate",
      "broad_answer_rate",
      "mechanism_specificity_score",
      "followup_context_retention_score",
      "stage4_contract_compliance_rate"
    ]);
    expect(response.body.report?.rubric_bands?.generic_explanation_rate).toBeTruthy();
    expect(response.body.report?.feature_profile_snapshot).toBeTruthy();
    expect(response.body.report?.code_version).toBeTruthy();
    expect(typeof response.body.report?.run_timestamp).toBe("string");
    expect(Array.isArray(response.body.report?.results)).toBe(true);
    expect(response.body.report?.results?.length).toBeGreaterThan(0);
  });

  it("loads canonical suite metadata and keeps it stable", async () => {
    const app = await createAppWithFlags({
      accountantEval: "1",
      answerPolicy: "1"
    });

    const response = await request(app).post("/api/eval/run").send({
      eval_target: "assistant_stage1",
      useMock: true,
      mode: "single-pass-strict",
      caseSetFile: "assistant_stage1_canonical_v0_1.json",
      normalizeConfig: {
        promptVersion: "normalizer_v2_0_2"
      }
    });

    expect(response.status).toBe(200);
    expect(response.body.report?.suite_id).toBe("assistant_stage1_canonical");
    expect(response.body.report?.suite_version).toBe("0.1.0");
    expect(response.body.report?.scenario_count).toBe(9);
    expect(Array.isArray(response.body.report?.case_ids)).toBe(true);
    expect(response.body.report?.case_ids?.length).toBe(9);
  });

  it("handles follow-up cases as dedicated subset", async () => {
    const app = await createAppWithFlags({
      accountantEval: "1",
      answerPolicy: "1"
    });

    const response = await request(app).post("/api/eval/run").send({
      eval_target: "assistant_stage1",
      useMock: true,
      mode: "single-pass-strict",
      caseSetFile: "assistant_stage1_canonical_v0_1.json",
      caseIds: ["S1-FOLLOWUP-INVESTIGATION", "S1-60-SUPPLIER-TAILS"],
      normalizeConfig: {
        promptVersion: "normalizer_v2_0_2"
      }
    });

    expect(response.status).toBe(200);
    expect(response.body.report?.subsets?.followup_cases_total).toBeGreaterThan(0);
    expect(response.body.report?.metrics?.raw?.followup_context_retention_score).not.toBeNull();
  });

  it("builds comparison artifact from baseline and current runs", async () => {
    const baselineApp = await createAppWithFlags({
      accountantEval: "1",
      answerPolicy: "0"
    });
    const baseline = await request(baselineApp).post("/api/eval/run").send({
      eval_target: "assistant_stage1",
      useMock: true,
      mode: "single-pass-strict",
      caseSetFile: "assistant_stage1_canonical_v0_1.json",
      normalizeConfig: {
        promptVersion: "normalizer_v2_0_2"
      }
    });
    expect(baseline.status).toBe(200);
    const baselinePath = String(baseline.body.report?.artifacts?.run_report_json_path ?? "");
    expect(baselinePath.length).toBeGreaterThan(0);

    const currentApp = await createAppWithFlags({
      accountantEval: "1",
      answerPolicy: "1"
    });
    const current = await request(currentApp).post("/api/eval/run").send({
      eval_target: "assistant_stage1",
      useMock: true,
      mode: "single-pass-strict",
      caseSetFile: "assistant_stage1_canonical_v0_1.json",
      compare_with_report_file: baselinePath,
      normalizeConfig: {
        promptVersion: "normalizer_v2_0_2"
      }
    });
    expect(current.status).toBe(200);
    expect(current.body.report?.comparison).toBeTruthy();
    expect(current.body.report?.comparison?.metric_deltas).toBeTruthy();
    expect(current.body.report?.comparison?.artifacts?.comparison_report_json_path).toBeTruthy();
  });

  it("keeps legacy eval path unchanged by default", async () => {
    const app = await createAppWithFlags({
      accountantEval: "1",
      answerPolicy: "1"
    });

    const response = await request(app).post("/api/eval/run").send({
      useMock: true,
      mode: "single-pass-strict",
      rawQuestions: "Проверь счет 60 за июнь 2020; Покажи риски по счету 97",
      normalizeConfig: {
        promptVersion: "normalizer_v2_0_2"
      }
    });

    expect(response.status).toBe(200);
    expect(response.body.report?.eval_target).toBeUndefined();
    expect(response.body.report?.metrics?.schema_validation_pass_rate).not.toBeUndefined();
    expect(response.body.report?.metrics?.route_resolution_accuracy).not.toBeUndefined();
  });

  it("respects accountant eval feature flag OFF/ON", async () => {
    const appOff = await createAppWithFlags({
      accountantEval: "0",
      answerPolicy: "1"
    });
    const offResponse = await request(appOff).post("/api/eval/run").send({
      eval_target: "assistant_stage1",
      useMock: true,
      mode: "single-pass-strict",
      caseSetFile: "assistant_stage1_canonical_v0_1.json",
      normalizeConfig: {
        promptVersion: "normalizer_v2_0_2"
      }
    });
    expect(offResponse.status).toBe(409);
    expect(offResponse.body?.error?.code).toBe("ASSISTANT_STAGE1_EVAL_DISABLED");

    const appOn = await createAppWithFlags({
      accountantEval: "1",
      answerPolicy: "1"
    });
    const onResponse = await request(appOn).post("/api/eval/run").send({
      eval_target: "assistant_stage1",
      useMock: true,
      mode: "single-pass-strict",
      caseSetFile: "assistant_stage1_canonical_v0_1.json",
      normalizeConfig: {
        promptVersion: "normalizer_v2_0_2"
      }
    });
    expect(onResponse.status).toBe(200);
    expect(onResponse.body.report?.eval_target).toBe("assistant_stage1");
  });
});