> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zhentan.me/llms.txt
> Use this file to discover all available pages before exploring further.

# Screening

> How AI transaction screening works from the user's perspective.

Every transaction you initiate is scored 0–100 against your behavioral profile before anything executes. The result is one of three verdicts.

## The Three Verdicts

<CardGroup cols="3">
  <Card title="APPROVE" icon="circle-check" iconType="duotone">
    **Score \< 40**

    Your transaction matches your normal patterns. The agent co-signs and executes immediately — no action needed from you.
  </Card>

  <Card title="REVIEW" icon="circle-question" iconType="duotone">
    **Score 40–70**

    Something looks unusual. You'll receive a Telegram message with the full risk analysis and \[Approve] \[Reject] buttons.
  </Card>

  <Card title="BLOCK" icon="circle-xmark" iconType="duotone">
    **Score > 70**

    The transaction is blocked outright. You'll receive a Telegram alert explaining why.
  </Card>
</CardGroup>

## What Gets Scored

Your profile is built from your past transactions. Each new transaction is checked against:

* **Recipient** — is this address in your known contacts?
* **Amount** — is this within your typical range for this recipient?
* **Daily volume** — would this push you over your daily limit?
* **Time of day** — is this happening during your normal activity window?

## Deep Analysis on REVIEW

For every REVIEW-tier transaction, the agent automatically runs external security checks before sending you the Telegram notification:

* **GoPlus Security** — recipient address reputation, sanctions, phishing, and known scam detection
* **Honeypot.is** — token security: honeypot detection, mintable supply, buy/sell tax rates

You receive the full report so you can make an informed decision.

## Responding via Telegram

When a transaction is in REVIEW, you'll get a Telegram message with:

* Transaction details (recipient, amount, token)
* Risk score and reasons
* Deep analysis results
* `[Approve]` and `[Reject]` buttons

Tap **Approve** to execute, or **Reject** to block it.

## Overriding the Agent

The agent advises — it doesn't gatekeep. In a **Protected** wallet, every proposal already sits in [app.safe.global](https://app.safe.global) at 1 of 2 signatures. If the agent flags or blocks something you're sure about, confirm it there with your backup key and execute — Zhentan syncs the result back automatically.

## Turning Screening Off

Screening is a choice in Protected wallets:

1. Toggle it off in **Settings → Zhentan Guard** (or via Telegram).
2. Your transactions now need **both of your keys** — the agent never signs what it didn't screen. Each send queues at 1 of 2 and completes when you co-sign with your backup wallet, either in-app (from transaction history) or in the Safe app.

Guarded wallets can't turn screening off — with one user key against a threshold of two, no second signature could exist. Add a backup key first. [Legacy v1 accounts](/technology/legacy) are the exception: their agent co-signs without analysis until they upgrade.

## Learning Over Time

After every confirmed transaction, Zhentan records the pattern. Recipients you transact with regularly become "known," amounts normalize, and your screening becomes more accurate over time — with fewer false REVIEW alerts.

## Telegram Commands

| Command                          | Action                                          |
| -------------------------------- | ----------------------------------------------- |
| `approve <tx-id>`                | Manually approve a pending transaction          |
| `reject <tx-id>`                 | Manually reject a pending transaction           |
| `get-status`                     | See current screening mode and recent decisions |
| `screening on` / `screening off` | Toggle screening                                |