Files
OmniRoute/tests/unit/compression/eval-grader.test.ts
Diego Rodrigues de Sa e Souza db358991f2 feat(compression): offline evaluation harness (Phase 4 D1) [stacked on #4716] (#4720)
Phase 4 D1: offline evaluation harness. Integrated into release/v3.8.35.
2026-06-23 08:15:34 -03:00

23 lines
966 B
TypeScript

import { describe, it } from "node:test";
import assert from "node:assert/strict";
import { buildGradePrompt, parseGradeVerdict } from "../../../open-sse/services/compression/eval/grader.ts";
describe("gold grader", () => {
it("buildGradePrompt embeds the answer and the gold", () => {
const msgs = buildGradePrompt("returns three", "3");
const joined = msgs.map((m) => m.content).join("\n");
assert.match(joined, /returns three/);
assert.match(joined, /3/);
assert.match(joined.toUpperCase(), /CORRECT|INCORRECT/);
});
it("parseGradeVerdict reads CORRECT / INCORRECT", () => {
assert.equal(parseGradeVerdict("VERDICT: CORRECT").correct, true);
assert.equal(parseGradeVerdict("verdict: incorrect — wrong number").correct, false);
});
it("parseGradeVerdict defaults to incorrect on unparseable output (never credits a wrong answer)", () => {
assert.equal(parseGradeVerdict("hmm, hard to say").correct, false);
});
});