Check brand adherence

You shipped a page that is supposed to look like the brand. Does it? Brand adherence turns that question into a number and a work list: one score from 0 to 1, prose recommendations, and concrete fixes with exact target values. This guide walks the workflow end to end. For the underlying model, see Brand adherence; for what runs inside it, see How the engine works.

When to use it

A rebuild against the original. You re-implemented a site and want proof that nothing drifted.
A generated page against the brand it was prompted with. An agent or a template produced the page; check it before it ships.
A redesign against the current brand. Measure how far the new direction moved, and whether the moves were the ones you intended.

In every case the roles are the same: reference_url is the brand standard, source_url is the page being judged. Both URLs must be reachable by the engine.

Run a check

Create the job

Call POST /judge/brand-adherence with the two URLs. Both sides are extracted automatically, so you do not create submissions first.

Poll the job

Call GET /judge/brand-adherence/{job_id} until status is completed. Stop on failed and read error.

Read the verdict

Call GET /judge/brand-adherence/{job_id}/result for the score, the recommendations, and the fixes.

Full example

import httpx
import time

API_KEY = "your-api-key"
BASE = "https://api.tastelabs.com"
HEADERS = {"X-API-Key": API_KEY}

def check_adherence(reference_url: str, source_url: str) -> dict:
    created = httpx.post(
        f"{BASE}/judge/brand-adherence",
        headers=HEADERS,
        json={"reference_url": reference_url, "source_url": source_url},
    )
    created.raise_for_status()
    job_id = created.json()["job_id"]
    print(f"Job created: {job_id}")

    while True:
        job = httpx.get(f"{BASE}/judge/brand-adherence/{job_id}", headers=HEADERS)
        job.raise_for_status()
        body = job.json()
        if body["status"] == "completed":
            break
        if body["status"] == "failed":
            raise RuntimeError(body.get("error") or "Adherence job failed")
        time.sleep(5)

    verdict = httpx.get(
        f"{BASE}/judge/brand-adherence/{job_id}/result", headers=HEADERS
    )
    verdict.raise_for_status()
    return verdict.json()

verdict = check_adherence("https://stripe.com", "https://staging.example.com")
print(f"Score: {verdict['score']:.2f}")
for item in verdict["recommendations"]:
    print(f"- {item}")

const API_KEY = "your-api-key";
const BASE = "https://api.tastelabs.com";
const HEADERS = { "X-API-Key": API_KEY, "Content-Type": "application/json" };

async function checkAdherence(referenceUrl: string, sourceUrl: string) {
  const created = await fetch(`${BASE}/judge/brand-adherence`, {
    method: "POST",
    headers: HEADERS,
    body: JSON.stringify({ reference_url: referenceUrl, source_url: sourceUrl }),
  });
  const { job_id } = await created.json();
  console.log(`Job created: ${job_id}`);

  while (true) {
    const job = await fetch(`${BASE}/judge/brand-adherence/${job_id}`, {
      headers: HEADERS,
    });
    const body = await job.json();
    if (body.status === "completed") break;
    if (body.status === "failed") throw new Error(body.error ?? "Failed");
    await new Promise((res) => setTimeout(res, 5000));
  }

  const verdict = await fetch(`${BASE}/judge/brand-adherence/${job_id}/result`, {
    headers: HEADERS,
  });
  return verdict.json();
}

const verdict = await checkAdherence(
  "https://stripe.com",
  "https://staging.example.com"
);
console.log(`Score: ${verdict.score}`);
for (const item of verdict.recommendations) console.log(`- ${item}`);

Act on the verdict

The three parts of the verdict answer three different questions:

score answers “how close is it?”. One number from 0 to 1.
recommendations answer “what should change?”. Prose strings, worst-first, with exact target values where they exist.
fixes answer “what can be changed mechanically?”. Structured objects, worst-first. Each carries an action discriminator (for example snap_to_token or add_color_token) plus the fields that action needs, so a script or an agent can apply them without interpretation.

Both lists are worst-first, so the first items always move the result most. Work from the top. Jobs are private by default. To share a verdict with someone who has no API key, make the job public with PATCH /judge/brand-adherence/{job_id}/visibility and {"is_public": true}. The result endpoint then serves the verdict without authentication, so the link works for anyone.

Getting Started

Use cases

Brand extractor

Brand search

Brand adherence verifier

Integrate Agents

Reference

When to use it

Run a check

Full example

Act on the verdict

Next steps

Verification in an agent loop

Brand adherence

​When to use it

​Run a check

​Full example

​Act on the verdict

​Share the result

​Next steps

Verification in an agent loop

Brand adherence

When to use it

Run a check

Full example

Act on the verdict

Share the result

Next steps