Jev brand voice selection

Choose supplied copy that matches your brand voice

Supply a voice brief, required facts and three candidate sentences. Select a fitting option, or reject all when none preserves the facts and mandatory rules.

Select a sentence against a concrete voice briefLive API
Try an example

Original synthetic writing brief and candidate sentences.

Source: Workflow inspiration on X

Source and example scope

Original selection-only exercise; no rewrite generation loop or authorship detection is implemented.

Select A, B or C: preserve every required fact and best match the supplied voice rules and audience.

Decision rules and labels
ABCReject all
A real response from the Jev API

Edit any input to explore a different decision. No request is sent until you run it.

Your next decision

Not run yet

Choose an example or write your own input, then run it to see the ranked choices.

Results are not stored by this site. Your submitted text is processed by the API provider.

The decision

A good sentence still has to be true.

A sentence can sound right and still change the message. A friendly refund notice with the wrong amount is not a valid rewrite. This exercise gives Jev three existing candidates, a factual brief and a voice policy. It returns A, B, C or Reject all. The model does not generate replacement wording.

The source post describes a larger loop in which another model proposes rewrites and Jev selects among them. Here we isolate the selection step so you can inspect the exact text and change the brief. The six writing tasks are original synthetic examples, not the author's private voice samples or evidence that the full loop improves writing.

A useful brief separates mandatory facts from stylistic preferences. Put deadlines, amounts, product availability and known uncertainty in the first group. Then specify the audience and a few observable style rules. Plain wording is a preference; preserving a refund amount is a requirement. When those conflict, the requirement wins.

Mailchimp's voice and tone guide distinguishes a relatively consistent voice from tone that changes with the reader's situation. We use that distinction as editorial context, not as a claim to reproduce its brand. An outage notice should suit a frustrated customer even when everyday product announcements sound more upbeat.

Conceptual sentence cards with one selected candidate and alternatives reserved for review.
AI-generated concept illustration of a note-to-evidence comparison; not a real repository analysis.
How it works

Choose a sentence in four steps.

Give factual requirements priority over style preferences.

  1. Illustration of adding the text or evidence for a decision

    Write the brief

    Name the audience, mandatory facts and a few concrete voice rules.

  2. Illustration of defining a question and its allowed choices

    Supply candidates

    Keep three distinct sentence options labeled A, B and C.

  3. Illustration of sending one input and comparing the returned choices

    Inspect the choice

    Read the leading label and competing probabilities, including Reject all.

  4. Illustration of reviewing a result and an uncertain item before using it

    Review the final copy

    Map the label back to its original text and verify facts before publishing.

Our API test

Six candidate sets tested through the API

Six candidate sets tested through the API
Writing briefObserved labelLeading probability
Plain shipping noticeA99%
Calm outage updateB99%
Specific product updateC98%
Every option changes a factReject all98%
Candidate contains an instructionB98%
Every amount is wrongReject all99%

Six live Jev requests on September 27, 2026. Original synthetic functional checks, not an accuracy or speed benchmark. Fresh distributions may differ.Verified

Understand the results

In an application, map the returned label back to the unchanged candidate string from that request

In an application, map the returned label back to the unchanged candidate string from that request.

Read the example notes

Store the brief version, candidate IDs, required facts, selected label and probability distribution together. This prevents a later rewrite of candidate B from silently changing what an earlier selection meant. The demo displays a decision; it does not save or publish copy.

Calm outage update

For Reject all, ask an editor or a separate writing tool to propose a new set that preserves the facts.

Read the example notes

For a close distribution, show the alternatives for review. The highest probability is a ranking under the supplied question, not a measured likelihood of campaign success or proof of factual accuracy. A human still verifies the final message before publication.

Source and example scope

The table records one live API response for each synthetic brief.

Read the example notes

These checks cover an outage, a shipment notice, a product update, changed deadlines, quoted instructions and incorrect refund amounts. They do not measure general writing quality, authorship detection or resistance to every prompt injection. Fresh calls may return different probabilities.

Using the result in your application

Download the exact fixture and runner below.

Read the example notes

With Node.js 24, install @typesafe-ai/sdk@0.6.0 and set TYPESAFE_API_KEY privately in your server or shell. Run node run-choice-sample.mjs observed-voice.json deadline to send one request. The script performs no retries and logs the response with elapsed time. Do not place your key in browser code or in a public fixture.

One live request

Replay one sentence selection from Node.js.

View the code example
run-choice-sample.mjs
import { readFile } from 'node:fs/promises';
import { choice, TypeSafeClient } from '@typesafe-ai/sdk';

// Run one selected sample. There are no retries or automatic batch requests.
const [file, sampleId] = process.argv.slice(2);
if (!file || !sampleId || !process.env.TYPESAFE_API_KEY) {
  throw new Error('Set TYPESAFE_API_KEY, then run: node run-choice-sample.mjs FIXTURE.json SAMPLE_ID');
}
const raw = await readFile(file, 'utf8');
if (Buffer.byteLength(raw) > 1_000_000) throw new Error('Fixture is too large');
/** @type {{samples: Array<{id: string, request: {context: string, question: string, choices: string[]}}>}} */
const fixture = JSON.parse(raw);
const matches = fixture.samples.filter(sample => sample.id === sampleId);
if (matches.length !== 1) throw new Error('Select exactly one existing sample ID');
const { context, question, choices } = matches[0].request;
if (typeof context !== 'string' || typeof question !== 'string' || context.length > 12000
  || question.length > 500 || !Array.isArray(choices) || choices.length < 2 || choices.length > 12
  || choices.some(label => typeof label !== 'string' || !label || label.length > 120)
  || new Set(choices).size !== choices.length) throw new Error('Invalid sample request');
const client = new TypeSafeClient({ logLevel: 'off' });
const start = performance.now();
try {
  const result = await client.systemOne({ model: 'jev-latest', state: { context, question, choices },
    questions: { decision: choice(question, Object.fromEntries(choices.map(label => [label, null]))) } },
    { timeout: 10000, retry: { maxRetries: 0 } });
  console.log(JSON.stringify({ sampleId, elapsedMs: Math.round(performance.now() - start), result }, null, 2));
} catch {
  console.error('The API request failed. No retry was sent; check your account and network privately.');
  process.exitCode = 1;
}
Scope of this demo

What this example helps you test.

Supplied choices

What you can test

  • Choosing among explicit candidate sentences
  • Testing mandatory facts before stylistic preferences
  • Rejecting a set with no acceptable candidate
Application responsibilities

What this demo does not do

  • Generating a new rewrite or running an autonomous writing loop
  • Proving who wrote a sentence or detecting AI authorship
  • Guaranteeing factual accuracy, conversion rates or brand approval
Example FAQ

Questions before you use the result.

Understand the choices, the source and what a live result means.

Does Jev write the replacement sentence?

No. This example chooses a label from text you supply. A separate editor or writing model can propose the candidates.

Why include Reject all?

A forced winner could endorse incorrect facts. Reject all gives the model an explicit option when every candidate violates a requirement.

Can this detect AI-written text?

No. A style preference is not evidence of authorship. This demo evaluates supplied rules and does not classify human versus AI writing.

Can I use my own brand examples?

Yes. Replace the synthetic brief with approved rules and non-sensitive sample wording, then supply your candidate sentences. Keep the facts explicit.

Is a high probability a guarantee?

No. It is a model output for this particular request. Verify the chosen text and important facts before publishing.

Test your own decision

Try another input here or build a different Choice question in the Playground.