The independent accuracy test for AI tariff classification

Every classification tool publishes an accuracy figure, and none of them is independent. Xzar tests your classifier on real classification decisions issued in 2026, graded exactly against the code that was actually issued. The answer keys never leave our servers, so the test can't be tuned to.

44%exact 10-digit code, leading general-purpose AI model

60%exact 8-digit code, same model

27%rule of thumb: copy the most similar earlier decision

1 in 5answers changed when the same case was asked twice

200 real decisions issued in 2026. Where does your classifier land?

No code? Click the button anyway. Work email; your account gets the full set of fresh cases, and your first 50 cases are on us. Results are private unless you choose to publish.

What you get

Scores at 2, 4, 6, 8 and 10 digits, plus a section-by-section breakdown of where you miss.
A comparison with the frontier-model baseline and the rule of thumb on the same cases.
A receipt for every case.
Full scorecard report ($5,000): 200 fresh cases, a consistency check (every case run twice), a written report and method note, and one rerun after your next model update. Email hello@xzar.si.

Run your classifier (about 10 lines)

Your classifier reads a product description and returns a code, exactly as it does for your customers.

import requests

API, KEY = "https://api.xzar.si", "YOUR_API_KEY"
H = {"Authorization": f"Bearer {KEY}"}

for _ in range(50):
    ep = requests.post(f"{API}/v1/episodes", headers=H, json={"taskset": "tariffs-classification"}).json()
    code = my_classifier(ep["instruction"])          # product description in, tariff code out
    requests.post(f"{API}/v1/episodes/{ep['episode_id']}/submit", headers=H, json={"answer": {"code": code}})

print(requests.get(f"{API}/v1/scorecard", headers=H).json()["tariffs"])

Each submit returns the score and a receipt. More in the docs. Questions: hello@xzar.si · Terms · Privacy