Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
Python calibrationclassificationdecision-modelhuggingfacejevmodernbertmultilingualnlppythonpytorchroutingtyped-decisionszero-shot
// A question's `instructions` is the text the model is asked, so a null or empty one is not a
// weak question, it is a question with nothing to answer. Python's `Agent._check_question`
// rejects null, a blank string, and an empty list or dict; the TypeScript port only checked that
// the key was present, and `toInternal` then serialised whatever it found, so `null` reached the
// model as the literal prompt "null" and `{}` as "{}".
import { describe, expect, it } from "vitest";
import { Agent, checkQuestion } from "../src/agent.js";
const fakeProvider = () => ({
async runEncoder(b: any) {
const n = b?.inputIds?.length ?? 1;
return { lastHidden: Array.from({ length: n }, () => [[1, 0], [0, 1]]) };
},
async runHead(_h: any, batch?: any) {
const n = batch?.inputIds?.length ?? 1;
return {
logits: Array.from({ length: n }, () => [2, 0, 1, 0]),
act: Array.from({ length: n }, () => [3, 0]),
};
},
});
const CHOICE = { type: "choice", criteria: { x: "yes", y: "no" } };
// Python: agent.py _check_question, the three rejections in a row.
const REJECTED: Array<[string, unknown]> = [
["null", null],
["undefined", undefined],
["an empty string", ""],
["a blank string", " "],
["an empty list", []],
["an empty dict", {}],
];
describe("checkQuestion instructions", () => {
it.each(REJECTED)("rejects %s", (_label, ins) => {
expect(() => checkQuestion("q", { ...CHOICE, instructions: ins })).toThrow(
/'instructions' must not be (None|empty)/,
);
});
it("names the question and what to fix", () => {
expect(() => checkQuestion("q", { ...CHOICE, instructions: null })).toThrow(
'question "q": \'instructions\' must not be None',
);
});
it("accepts instructions that can be answered", () => {
for (const ins of ["Is it urgent?", " padded ", 42, { k: "v" }, ["a", "b"]]) {
expect(() => checkQuestion("q", { ...CHOICE, instructions: ins })).not.toThrow();
}
});
it("still reports a missing key as missing", () => {
expect(() => checkQuestion("q", { ...CHOICE })).toThrow("no 'instructions'");
});
});
describe("checkQuestion id", () => {
it("rejects a blank question id, as Python does", () => {
for (const qid of ["", " "]) {
expect(() => checkQuestion(qid, { ...CHOICE, instructions: "q?" })).toThrow(
"question id must be a non-empty string",
);
}
});
});
describe("Agent", () => {
it("rejects an unanswerable question instead of prompting with it", async () => {
const a = new Agent({ provider: fakeProvider() } as any);
for (const method of ["predict", "systemOne"] as const) {
for (const [_label, ins] of REJECTED) {
await expect(
a[method]("hello", { q: { ...CHOICE, instructions: ins } } as any),
).rejects.toThrow('question "q": \'instructions\' must not be');
}
}
});
it("still answers a question whose instructions are real", async () => {
const a = new Agent({ provider: fakeProvider() } as any);
const r: any = await a.systemOne("hi", {
q: { ...CHOICE, instructions: "Is it urgent?" },
});
expect(r.answers.q.choice).toBe("x");
});
});