> For the complete documentation index, see [llms.txt](https://connectyai.gitbook.io/documentation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://connectyai.gitbook.io/documentation/usage-guides-enterprise-plan/chat-features/verified-knowledge.md).

# Verified Knowledge

Allows workspace moderators (human-expert-in-the-loop) to verify AI-generated answers and their underlying metric entities, so they can be reused reliably across the data workspace.

### What is it?

Connecty AI introduces an **Expert-in-the-Loop Verification System** that allows data workspace moderators to verify AI-generated responses and their underlying semantic logic. This includes not just SQL snippets, but also the full context: metric definitions, filters, dimensions, relationships, and more.

Every verification becomes part of a **living semantic knowledge base** - stored in the Metricverse - used to improve accuracy, consistency, and reasoning across future queries and sessions.

{% hint style="info" %}
Unlike traditional tools that store static SQL, Connecty AI models and understands **semantic dependencies** between questions, metrics, and business logic - so verified knowledge is precise, reusable, and impact-aware.
{% endhint %}

### Why verify?

Modern data teams struggle with **metric inconsistencies**, unclear logic, and dashboards that users don't trust. Different teams define the same KPI in conflicting ways, and even AI-generated answers can surface unapproved or outdated logic.

**Connecty’s verification system solves this by letting you:**

✅ **Establish trust** in AI-generated responses through human-approved logic\
🔁 **Ensure consistency** across metrics, dashboards, and questions\
📚 **Build a curated library** of verified metric definitions everyone can rely on\
⚠️ **Prevent conflicting logic** from being reused or silently applied\
🧠 **Guide the AI** to reuse only trusted, context-appropriate definitions

{% hint style="info" %}
This system supports both **question-level verification** (e.g. “Is this answer correct?”) and **component-level verification** (e.g. “Is this metric definition correct?”).
{% endhint %}

### How it works

#### 1. Verifying a Chat Response

<figure><img src="https://4247993803-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMVGgoc4urxfCJy9iAJG7%2Fuploads%2FJqnqHT0DO2wThElld1d6%2FWhere%20to%20find%20it%20-%20mark%20as%20verified.png?alt=media&amp;token=7455918d-3d6a-4789-aa48-878254ab51f4" alt=""><figcaption></figcaption></figure>

When a data workspace moderator is satisfied with a response to a user question in chat, they can choose to "Verify this response."

* Connecty will automatically extract all underlying metric entities used in the SQL response.
* A popup will show the list of dependencies: metrics, dimensions, filters, joins—any entity that contributed to the result.

<figure><img src="https://4247993803-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMVGgoc4urxfCJy9iAJG7%2Fuploads%2F8G3dcwAbNKbxZXjSAG9N%2Fimage.png?alt=media&amp;token=5b05007c-ee9b-45f2-ad06-761b65ab3d73" alt=""><figcaption></figcaption></figure>

* The moderator accepts and the response is marked as verified, and all associated metric entities are transitioned to a Verified state.

#### 2. Managing Verifications in the Metricverse

The Metricverse tracks the verification status of every metric entity:

* Verified – Reviewed and approved by a moderator
* Unverified – Used in responses but not yet approved
* No Action – Not yet reviewed or surfaced in a response

Moderators can also directly verify or revoke the verification of any metric entity from the Metricverse interface—independent of chat responses.

### What Can Be Verified?

You can verify and store the following types of information in Knowledge:

#### 🔹 Metric Entities

<figure><img src="https://4247993803-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMVGgoc4urxfCJy9iAJG7%2Fuploads%2F1GpUI4M2pOo3ZmtcD8T5%2FWhere%20to%20find%20it%20-%20mark%20as%20verified%20(metric).png?alt=media&amp;token=37c47385-06c8-4381-9b01-16dc8b1f8369" alt=""><figcaption></figcaption></figure>

Semantic components extracted from your questions or SQL, including:

* **Metrics** (e.g., `Gross Margin`, `Net Revenue`)
* **Subjects** (e.g., `Orders`, `Customers`)
* **Measures** (e.g., `Total Spend`)
* **Relationships** (e.g., `Customers JOIN Orders`)
* **Dimensions** (e.g., `Time`, `Channel`)
* **Filters** (e.g., `status = active`, `date > last quarter`)

#### 🔹 Chat Answers

You can verify full **answers** to user questions, including:

* The natural language **interpretation** of the question
* The **generated SQL**
* The full set of **semantic entities** involved (metrics, dimensions, filters, joins)

Verifying an answer confirms that the AI correctly understood and resolved the question using trusted logic.

***

### How to Utilize Verified Knowledge&#x20;

1. In Chat, while asking a new question, **enable the Verified button** to restrict the knowledge to those entities that have been marked verified by the Workspace moderator.

<figure><img src="https://4247993803-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMVGgoc4urxfCJy9iAJG7%2Fuploads%2FURl7uM5p7WabRPN7xy9m%2Fimage.png?alt=media&amp;token=49a4629f-0bff-4001-9b21-e29a31c453d1" alt=""><figcaption></figcaption></figure>

2. In Metricverse left side panel, easily filter by Verified or Unverified

<figure><img src="https://4247993803-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMVGgoc4urxfCJy9iAJG7%2Fuploads%2F2f5xRqyjjyEABHxI0xac%2Fimage.png?alt=media&amp;token=9d0a9683-7256-4c8a-aff9-94c541a027d3" alt=""><figcaption></figcaption></figure>

3\. In Context Engine, easily filter by Verified or Unverified

<figure><img src="https://4247993803-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMVGgoc4urxfCJy9iAJG7%2Fuploads%2FuT4AV1v9gkGksEw26mwZ%2Fimage.png?alt=media&amp;token=ebbc5041-2067-4c95-ba40-deb6177e733c" alt=""><figcaption></figcaption></figure>

4. Verified questions (and their queries) are also stored in Context Engine and can be easily filtered out, with full access to its metadata including versions and runs history.&#x20;

<figure><img src="https://4247993803-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FMVGgoc4urxfCJy9iAJG7%2Fuploads%2FZrEYhW3F7UHFgmZBSuqp%2Fimage.png?alt=media&amp;token=79c81f30-cd44-4ba6-9fe9-61cee7034bdb" alt=""><figcaption></figcaption></figure>

### Which Workspace is the Verified Knowledge Applied

* **Scope:** Verification is applied **only within the current data workspace**. Other workspaces are isolated.
* **Usage:** Verified components are reused during:
  * Semantic parsing of future questions
  * SQL generation
  * Conflict resolution and clarification
* **Impact:** Downstream responses will prefer verified definitions over inferred logic when contexts match.

***

### AI Intelligence: How Connecty Applies Knowledge

Connecty AI’s **autonomous semantic graph** is a dynamic representation of your data, logic, and verified knowledge.

#### Key Behaviors:

* **Semantic Graph Construction:**\
  On each question or query, Connecty builds a real-time graph of all involved entities, metrics, relationships, and filters.
* **Dependency Resolution:**\
  When you verify a metric or question, Connecty tracks all related components. It understands that `Net Revenue` depends on `Revenue`, `Refunds`, and `Taxes`, and maps this dependency.
* **Scoped Trust Propagation:**\
  Verified definitions are reused *only* when contextually compatible. For example, `Customer Lifetime Value` verified for DTC customers will not be reused for B2B metrics unless dependency conditions match.
* **Version & Conflict Detection:**\
  If multiple definitions exist (e.g., for `Active Users`), Connecty detects the divergence and prompts for clarification before use.
* **Auto-Adaptive Updates:**\
  If your schema changes (e.g., renaming `plan_tier` to `tier_code`), Connecty updates affected graph nodes and recalculates dependent logic accordingly.

This system enables **safe reuse**, **conflict resolution**, and **precision reasoning** at semantic scale.

***

### Things to Consider

* **Auto-Verification Cascade:**\
  Verifying a question may select multiple dependent components. You'll be shown a warning and can deselect any items manually.

  > *“You are verifying the response generated for this question. Verifying this will also auto-verify all its underlying components that are selected below. Unselect any components you don’t want to include. The verification will be applied for the currently selected data workspace.”*
* **Conflict Alerts:**\
  If verification introduces a conflict with existing Knowledge, you’ll receive a versioning or compatibility warning.
* **Manual Review:**\
  Periodically audit your workspace’s Knowledge base to remove outdated logic or revise evolving definitions.
* **Partial Verification is Supported:** You can verify only parts of a query—such as specific metrics or filters—without storing the entire question or SQL.

***

### Example Workflow

> **Question:** *“What is our gross margin by region for the last quarter?”*
>
> Connecty extracts the following:
>
> * `Metric`: `Gross Margin` = `Revenue - COGS`
> * `Dimension`: `Region`
> * `Filter`: `last quarter`

Once verified:

* These definitions are stored with their dependency graph.
* If another user later asks *“Compare gross margin across regions”*, the system will automatically reuse the trusted metric logic—ensuring consistency and avoiding errors.