pi-harness-runtime 1.1.439 → 1.1.441

package/packages/jev-judge/tests/typesafe-provider.test.ts

@@ -1,640 +0,0 @@
-/**
- * Unit tests for the TypeSafe Jev provider.
- *
- * The TypeSafe API is stubbed at the `fetch` boundary, so no network calls and
- * no credentials are involved.
- *
- * Regression coverage:
- *
- * 1. `buildDecision()` used to evaluate the choice answer as
- *
- *      answer.choice ?? question.type === "choice" ? options[0] : ""
- *
- *    which JS parses as `(answer.choice ?? (question.type === "choice")) …`
- *    - so the model's choice was DISCARDED and the first option always won.
- *
- * 2. The request contract: the TypeSafe API rejects a choice question whose
- *    `criteria` is an ARRAY ("Input should be a valid dictionary") and a
- *    score question without `criteria` ("Field required"). Both used to be
- *    sent that way, so every real auto-continue request failed with 422 and
- *    the judge silently fell back to its "wait" default.
- */
-
-import { afterEach, describe, expect, it } from "bun:test";
-import { TypeSafeJudge } from "../src/typesafe-provider.js";
-import { JevProviderError } from "../src/provider-error.js";
-import type { JevChoiceResponse, JevNoulResponse, JevScoreResponse } from "../src/types.js";
-
-const originalFetch = globalThis.fetch;
-
-interface StubAnswer {
-	type: "noul" | "choice" | "score";
-	noul?: number;
-	choice?: string;
-	score?: number;
-	confidence?: number;
-}
-
-interface CapturedRequest {
-	url: string;
-	body: Record<string, unknown>;
-}
-
-let lastRequest: CapturedRequest | null = null;
-
-function stubTypeSafe(answers: Record<string, StubAnswer>): void {
-	lastRequest = null;
-	globalThis.fetch = (async (url: string, init?: { body?: string }) => {
-		lastRequest = {
-			url,
-			body: init?.body ? (JSON.parse(init.body) as Record<string, unknown>) : {},
-		};
-		return {
-			ok: true,
-			status: 200,
-			json: async () => ({ answers, model: "test-model", latency_ms: 1 }),
-			text: async () => JSON.stringify({ answers }),
-		};
-	}) as unknown as typeof fetch;
-}
-
-function stubFailure(status: number, body = "boom"): void {
-	lastRequest = null;
-	globalThis.fetch = (async (url: string, init?: { body?: string }) => {
-		lastRequest = {
-			url,
-			body: init?.body ? (JSON.parse(init.body) as Record<string, unknown>) : {},
-		};
-		return {
-			ok: false,
-			status,
-			json: async () => ({ detail: body }),
-			text: async () => body,
-		};
-	}) as unknown as typeof fetch;
-}
-
-function stubNetworkError(error?: unknown): void {
-	globalThis.fetch = (async () => {
-		throw error ?? new Error("fetch failed");
-	}) as unknown as typeof fetch;
-}
-
-/**
- * The error shape undici produces for a connection failure: a TypeError whose
- * cause carries the system error code.
- */
-function undiciConnectionError(): Error {
-	return Object.assign(new TypeError("fetch failed"), {
-		cause: Object.assign(new Error("connect ECONNREFUSED 127.0.0.1:443"), {
-			code: "ECONNREFUSED",
-		}),
-	});
-}
-
-function makeJudge(): TypeSafeJudge {
-	return new TypeSafeJudge({
-		apiKey: "test-key",
-		baseUrl: "https://typesafe.invalid",
-		model: "test-model",
-	});
-}
-
-afterEach(() => {
-	globalThis.fetch = originalFetch;
-	lastRequest = null;
-});
-
-describe("TypeSafeJudge - choice answers", () => {
-	it("uses the choice returned by the model", async () => {
-		stubTypeSafe({
-			answer: { type: "choice", choice: "gamma", confidence: 0.8 },
-		});
-
-		const result = await makeJudge().evaluate("state", {
-			answer: {
-				type: "choice",
-				instructions: "Pick one",
-				options: ["alpha", "beta", "gamma"],
-			},
-		});
-
-		const response = result.decisions.answer.response as JevChoiceResponse;
-		expect(response.choice).toBe("gamma");
-		expect(result.decisions.answer.probability).toBe(0.8);
-	});
-
-	it("sends the choice criteria as a dictionary", async () => {
-		stubTypeSafe({
-			answer: { type: "choice", choice: "high", confidence: 0.6 },
-		});
-
-		await makeJudge().evaluate("state", {
-			answer: {
-				type: "choice",
-				instructions: "Pick one",
-				options: ["critical", "high"],
-				optionDescriptions: {
-					critical: "it is critical",
-					high: "it is high",
-				},
-			},
-		});
-
-		const questions = lastRequest?.body.questions as Record<
-			string,
-			{ type: string; criteria: unknown }
-		>;
-		expect(questions.answer.type).toBe("choice");
-		expect(questions.answer.criteria).toEqual({
-			critical: "it is critical",
-			high: "it is high",
-		});
-	});
-
-	it("accepts a criteria dictionary directly", async () => {
-		stubTypeSafe({
-			answer: { type: "choice", choice: "high", confidence: 0.6 },
-		});
-
-		await makeJudge().evaluate("state", {
-			answer: {
-				type: "choice",
-				instructions: "Pick one",
-				criteria: { critical: "it is critical", high: "it is high" },
-			},
-		});
-
-		const questions = lastRequest?.body.questions as Record<
-			string,
-			{ criteria: unknown }
-		>;
-		expect(questions.answer.criteria).toEqual({
-			critical: "it is critical",
-			high: "it is high",
-		});
-	});
-
-	it("matches the option spelling case insensitively", async () => {
-		stubTypeSafe({
-			answer: { type: "choice", choice: "GAMMA", confidence: 0.7 },
-		});
-
-		const result = await makeJudge().evaluate("state", {
-			answer: {
-				type: "choice",
-				instructions: "Pick one",
-				options: ["alpha", "beta", "gamma"],
-			},
-		});
-
-		const response = result.decisions.answer.response as JevChoiceResponse;
-		expect(response.choice).toBe("gamma");
-	});
-
-	it("passes an unknown choice through instead of substituting an option", async () => {
-		stubTypeSafe({
-			answer: { type: "choice", choice: "delta", confidence: 0.7 },
-		});
-
-		const result = await makeJudge().evaluate("state", {
-			answer: {
-				type: "choice",
-				instructions: "Pick one",
-				options: ["alpha", "beta"],
-			},
-		});
-
-		const response = result.decisions.answer.response as JevChoiceResponse;
-		expect(response.choice).toBe("delta");
-	});
-});
-
-describe("TypeSafeJudge - noul and score answers", () => {
-	it("keeps noul answers intact", async () => {
-		stubTypeSafe({ answer: { type: "noul", noul: 0.75 } });
-
-		const result = await makeJudge().evaluate("state", {
-			answer: { type: "noul", instructions: "Is it true?" },
-		});
-
-		const response = result.decisions.answer.response as JevNoulResponse;
-		expect(response.noul).toBe(0.75);
-		expect(result.decisions.answer.probability).toBe(0.75);
-	});
-
-	it("sends a score question with a rubric", async () => {
-		stubTypeSafe({ answer: { type: "score", score: 3, confidence: 0.6 } });
-
-		await makeJudge().evaluate("state", {
-			answer: {
-				type: "score",
-				instructions: "How bad?",
-				min: 1,
-				max: 5,
-				criteria: ["1 - none", "2 - low", "3 - medium", "4 - high", "5 - severe"],
-			},
-		});
-
-		const questions = lastRequest?.body.questions as Record<
-			string,
-			{ type: string; criteria: unknown }
-		>;
-		expect(questions.answer.type).toBe("score");
-		expect(questions.answer.criteria).toEqual([
-			"1 - none",
-			"2 - low",
-			"3 - medium",
-			"4 - high",
-			"5 - severe",
-		]);
-	});
-
-	it("derives a rubric from min and max when none is given", async () => {
-		stubTypeSafe({ answer: { type: "score", score: 3.37, confidence: 0.6 } });
-
-		const result = await makeJudge().evaluate("state", {
-			answer: {
-				type: "score",
-				instructions: "How bad?",
-				min: 1,
-				max: 5,
-			},
-		});
-
-		const questions = lastRequest?.body.questions as Record<
-			string,
-			{ criteria: string[] }
-		>;
-		expect(Array.isArray(questions.answer.criteria)).toBe(true);
-		expect(questions.answer.criteria.length).toBe(5);
-
-		// The API answers with the position in the rubric, the reported value
-		// is mapped back onto the min/max scale.
-		const response = result.decisions.answer.response as JevScoreResponse;
-		expect(response.score).toBeCloseTo(4.37, 5);
-		expect(result.decisions.answer.probability).toBeCloseTo(0.8425, 5);
-	});
-
-	it("maps the rubric position onto the min/max scale", async () => {
-		stubTypeSafe({ answer: { type: "score", score: 4.5, confidence: 0.6 } });
-
-		const result = await makeJudge().evaluate("state", {
-			answer: {
-				type: "score",
-				instructions: "How good?",
-				min: 0,
-				max: 100,
-			},
-		});
-
-		const response = result.decisions.answer.response as JevScoreResponse;
-		// 10 levels were derived (0..100), 4.5 of 9 steps => 50
-		expect(response.score).toBeCloseTo(50, 5);
-		expect(result.decisions.answer.probability).toBeCloseTo(0.5, 5);
-	});
-
-	it("clamps a rubric position outside the rubric", async () => {
-		stubTypeSafe({ answer: { type: "score", score: 99, confidence: 0.6 } });
-
-		const result = await makeJudge().evaluate("state", {
-			answer: {
-				type: "score",
-				instructions: "How good?",
-				min: 1,
-				max: 5,
-			},
-		});
-
-		const response = result.decisions.answer.response as JevScoreResponse;
-		expect(response.score).toBe(5);
-		expect(result.decisions.answer.probability).toBe(1);
-	});
-});
-
-describe("TypeSafeJudge - request validation", () => {
-	it("rejects a choice question without options before sending anything", async () => {
-		let called = false;
-		globalThis.fetch = (async () => {
-			called = true;
-			return { ok: true, status: 200, json: async () => ({}), text: async () => "{}" };
-		}) as unknown as typeof fetch;
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "choice", instructions: "Pick one" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect(error).toBeInstanceOf(JevProviderError);
-		expect((error as JevProviderError).retryable).toBe(false);
-		expect((error as JevProviderError).message).toContain("options");
-		expect(called).toBe(false);
-	});
-
-	it("rejects a score question with too many rubric levels", async () => {
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: {
-					type: "score",
-					instructions: "How good?",
-					criteria: Array.from({ length: 12 }, (_, i) => `level ${i}`),
-				},
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect(error).toBeInstanceOf(JevProviderError);
-		expect((error as JevProviderError).retryable).toBe(false);
-	});
-});
-
-describe("TypeSafeJudge - wire format", () => {
-	it("sends only the fields the API accepts", async () => {
-		stubTypeSafe({ answer: { type: "noul", noul: 0.5 } });
-
-		await makeJudge().evaluate("state", {
-			answer: { type: "noul", instructions: "Is it true?" },
-		});
-
-		// The API rejects unknown fields with 400 "Invalid request" - a local
-		// bookkeeping field that leaks into the body breaks every request.
-		expect(Object.keys(lastRequest?.body ?? {}).sort()).toEqual([
-			"model",
-			"questions",
-			"state",
-		]);
-	});
-
-	it("sends only the fields the API accepts for a question", async () => {
-		stubTypeSafe({ answer: { type: "score", score: 2, confidence: 0.5 } });
-
-		await makeJudge().evaluate("state", {
-			answer: {
-				type: "score",
-				instructions: "How bad?",
-				min: 1,
-				max: 5,
-				criteria: ["1 - none", "2 - low", "3 - medium", "4 - high", "5 - severe"],
-			},
-		});
-
-		const questions = lastRequest?.body.questions as Record<string, Record<string, unknown>>;
-		expect(Object.keys(questions.answer).sort()).toEqual([
-			"criteria",
-			"instructions",
-			"type",
-		]);
-		// The local scale bookkeeping must not reach the wire.
-		expect(questions.answer.min).toBeUndefined();
-		expect(questions.answer.max).toBeUndefined();
-		expect(questions.answer.scoreScales).toBeUndefined();
-	});
-
-	it("sends only the fields the API accepts for a choice question", async () => {
-		stubTypeSafe({ answer: { type: "choice", choice: "yes", confidence: 0.5 } });
-
-		await makeJudge().evaluate("state", {
-			answer: {
-				type: "choice",
-				instructions: "Pick one",
-				options: ["yes", "no"],
-			},
-		});
-
-		const questions = lastRequest?.body.questions as Record<string, Record<string, unknown>>;
-		expect(Object.keys(questions.answer).sort()).toEqual([
-			"criteria",
-			"instructions",
-			"type",
-		]);
-		expect(questions.answer.options).toBeUndefined();
-	});
-
-	it("describes a named candidate by its name when no description is given", async () => {
-		stubTypeSafe({ answer: { type: "choice", choice: "yes", confidence: 0.5 } });
-
-		await makeJudge().evaluate("state", {
-			answer: {
-				type: "choice",
-				instructions: "Pick one",
-				options: ["yes", "no"],
-			},
-		});
-
-		const questions = lastRequest?.body.questions as Record<string, { criteria: unknown }>;
-		expect(questions.answer.criteria).toEqual({ yes: "yes", no: "no" });
-	});
-
-	it("resolves the candidates of a criteria only question", async () => {
-		stubTypeSafe({ answer: { type: "choice", choice: "high", confidence: 0.5 } });
-
-		const result = await makeJudge().evaluate("state", {
-			answer: {
-				type: "choice",
-				instructions: "Pick one",
-				criteria: { critical: "it is critical", high: "it is high" },
-			},
-		});
-
-		const response = result.decisions.answer.response as JevChoiceResponse;
-		expect(response.choice).toBe("high");
-	});
-});
-
-describe("TypeSafeJudge - error classification", () => {
-	it("marks a contract error (422) as not retryable", async () => {
-		stubFailure(422, "Input should be a valid dictionary");
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect(error).toBeInstanceOf(JevProviderError);
-		expect((error as JevProviderError).status).toBe(422);
-		expect((error as JevProviderError).retryable).toBe(false);
-	});
-
-	it("marks a bad request (400) as not retryable", async () => {
-		stubFailure(400, "Invalid request.");
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect((error as JevProviderError).status).toBe(400);
-		expect((error as JevProviderError).retryable).toBe(false);
-	});
-
-	it("marks a rate limit (429) as retryable", async () => {
-		stubFailure(429, "slow down");
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect((error as JevProviderError).status).toBe(429);
-		expect((error as JevProviderError).retryable).toBe(true);
-	});
-
-	it("marks a server error (500) as retryable", async () => {
-		stubFailure(500);
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect((error as JevProviderError).status).toBe(500);
-		expect((error as JevProviderError).retryable).toBe(true);
-	});
-
-	it("marks an auth error (401) as retryable", async () => {
-		stubFailure(401);
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect((error as JevProviderError).status).toBe(401);
-		expect((error as JevProviderError).retryable).toBe(true);
-	});
-
-	it("marks a network failure as retryable", async () => {
-		stubNetworkError(undiciConnectionError());
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect(error).toBeInstanceOf(JevProviderError);
-		expect((error as JevProviderError).retryable).toBe(true);
-	});
-
-	it("marks a timeout as retryable", async () => {
-		stubNetworkError(Object.assign(new Error("The operation timed out"), { name: "TimeoutError" }));
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect((error as JevProviderError).retryable).toBe(true);
-	});
-
-	it("does NOT mark a bug of our own as retryable", async () => {
-		// A malformed URL is not a provider outage: falling back to another
-		// provider would only hide the bug.
-		stubNetworkError(new TypeError("Failed to parse URL"));
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect(error).toBeInstanceOf(JevProviderError);
-		expect((error as JevProviderError).retryable).toBe(false);
-	});
-
-	it("marks the failure of several addresses as retryable", async () => {
-		// undici tries the addresses of a host one after the other and reports
-		// the failures as an AggregateError. The cause here is NOT a transport
-		// failure: the addresses are what makes the attempt a transport
-		// failure.
-		const aggregate = new AggregateError(
-			[
-				Object.assign(new Error("certificate has expired"), { code: "CERT_HAS_EXPIRED" }),
-				Object.assign(new Error("connect EHOSTUNREACH 10.0.0.1:443"), {
-					code: "EHOSTUNREACH",
-				}),
-			],
-			"all failed"
-		);
-		Object.assign(aggregate, { cause: new TypeError("Failed to parse URL") });
-		stubNetworkError(aggregate);
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect((error as JevProviderError).retryable).toBe(true);
-	});
-
-	it("marks an expired provider certificate as retryable", async () => {
-		for (const code of [
-			"CERT_HAS_EXPIRED",
-			"SELF_SIGNED_CERT_IN_CHAIN",
-			"DEPTH_ZERO_SELF_SIGNED_CERT",
-			"UNABLE_TO_VERIFY_LEAF_SIGNATURE",
-		]) {
-			stubNetworkError(Object.assign(new Error(`certificate error ${code}`), { code }));
-
-			const error = (await makeJudge()
-				.evaluate("state", {
-					answer: { type: "noul", instructions: "Is it true?" },
-				})
-				.catch((e: unknown) => e)) as JevProviderError;
-
-			expect(`${code}: ${(error as JevProviderError).retryable}`).toBe(`${code}: true`);
-		}
-	});
-
-	it("does NOT mark a stream bug of our own as retryable", async () => {
-		// Writing to a finished stream is our bug, not a provider outage.
-		stubNetworkError(
-			Object.assign(new Error("write after end"), { code: "ERR_STREAM_WRITE_AFTER_END" })
-		);
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect((error as JevProviderError).retryable).toBe(false);
-	});
-
-	it("marks an unreachable network as retryable", async () => {
-		stubNetworkError(
-			Object.assign(new Error("connect ENETUNREACH 10.0.0.1:443"), { code: "ENETUNREACH" })
-		);
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect((error as JevProviderError).retryable).toBe(true);
-	});
-
-	it("marks a TLS failure as retryable", async () => {
-		stubNetworkError(
-			Object.assign(new Error("certificate altname mismatch"), {
-				code: "ERR_TLS_CERT_ALTNAME_INVALID",
-			})
-		);
-
-		const error = (await makeJudge()
-			.evaluate("state", {
-				answer: { type: "noul", instructions: "Is it true?" },
-			})
-			.catch((e: unknown) => e)) as JevProviderError;
-
-		expect((error as JevProviderError).retryable).toBe(true);
-	});
-});