29 lines
1.2 KiB
TypeScript
29 lines
1.2 KiB
TypeScript
import request from "supertest";
|
|
import { describe, expect, it } from "vitest";
|
|
import { createApp } from "../src/server";
|
|
|
|
describe("POST /api/eval/run", () => {
|
|
it("runs v2 eval using inline rawQuestions batch", async () => {
|
|
const app = createApp();
|
|
const response = await request(app).post("/api/eval/run").send({
|
|
normalizeConfig: {
|
|
promptVersion: "normalizer_v2_0_2",
|
|
useMock: true
|
|
},
|
|
useMock: true,
|
|
mode: "single-pass-strict",
|
|
rawQuestions:
|
|
"Проверь хвосты по поставщикам и разложи цепочку; Как вообще по ФСБУ; Покажи топ рисков за июнь 2020"
|
|
});
|
|
|
|
expect(response.status).toBe(200);
|
|
expect(response.body.ok).toBe(true);
|
|
expect(response.body.report?.schema_version).toBe("v2_0_2");
|
|
expect(response.body.report?.cases_total).toBe(3);
|
|
expect(typeof response.body.report?.metrics?.schema_validation_pass_rate).toBe("number");
|
|
expect(response.body.report?.metrics?.route_resolution_accuracy).not.toBeUndefined();
|
|
expect(response.body.report?.metrics?.execution_state_consistency_rate).not.toBeUndefined();
|
|
expect(Array.isArray(response.body.report?.results)).toBe(true);
|
|
});
|
|
});
|