> For the complete documentation index, see [llms.txt](https://v2.dataos.info/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://v2.dataos.info/consume/understand/understand-and-trust-with-ai.md).

# Understand in natural language

Before you ask a business question, confirm that the Data Product exposes the right governed metrics and dimensions, that the data is fresh and trustworthy, and how the answer is derived. Your [AI assistant](/consume/consume-with-ai/connect-clients.md) can pull all of that without opening the catalog, [Studio](/consume/understand/overview.md), or a [SQL client](/consume/activate/mysql-clients.md).

The prompts and responses below are illustrative examples from a sample catalog. Your environment will return different products, metrics, and results.

![Claude Desktop response showing the Orders360 semantic schema](/files/pB7PoauZJDJjLdXDxPlg)

<details>

<summary>Full response (example)</summary>

Here's a full overview of the **Orders360** data product:

***

**🛒 Orders360**

*Sales analytics data product on Snowflake*

| Field              | Value                                            |
| ------------------ | ------------------------------------------------ |
| **Description**    | All information about the orders360 data product |
| **Domain**         | Sales                                            |
| **Tags**           | snowflake, orders360                             |
| **Glossary terms** | glossary.data\_product                           |
| **Owner**          | dana.lee                                         |
| **Use cases**      | None configured                                  |
| **Limitations**    | None configured                                  |
| **README**         | Not configured                                   |

View data product

***

**📐 Semantic layer (4 tables)**

| # | Name                       | Type     | Description                        |
| - | -------------------------- | -------- | ---------------------------------- |
| 1 | customer                   | Semantic | Customer dimension                 |
| 2 | customer\_orders\_enriched | Semantic | Enriched customer-order fact table |
| 3 | orders                     | Semantic | Orders fact table                  |
| 4 | products                   | Semantic | Products dimension                 |

***

**🏗️ Physical models (8 tables)**

| # | Name                           | Type            | Description                                              |
| - | ------------------------------ | --------------- | -------------------------------------------------------- |
| 1 | CUSTOMER\_ORDERS\_ENRICHED     | Physical        | Enriched customer-order-product fact with all data       |
| 2 | ORDERS                         | Physical        | Orders fact table with incremental loading by order date |
| 3 | PRODUCTS                       | Physical        | Product dimension with full refresh on each run          |
| 4 | CUSTOMERS                      | Physical        | Customer dimension with full refresh                     |
| 5 | RAW\_ORDERS                    | Physical (Seed) | Raw order data loaded from CSV                           |
| 6 | RAW\_CUSTOMERS                 | Physical (Seed) | Raw customer data loaded from CSV                        |
| 7 | RAW\_PRODUCTS                  | Physical (Seed) | Raw product data loaded from CSV                         |
| 8 | RAW CUSTOMER\_ORDERS\_ENRICHED | Physical (Seed) | Enriched customer, order, and product seed data          |

***

**📊 Business metrics (3)**

| # | Metric           | Description                                                 |
| - | ---------------- | ----------------------------------------------------------- |
| 1 | customer\_growth | Customer growth by segment and account status over time     |
| 2 | order\_volume    | Daily order volume by product category and customer segment |
| 3 | revenue\_trend   | Monthly revenue trend by customer and product category      |

***

The product follows a standard medallion architecture: **raw seeds → physical Gold models → semantic layer** on Snowflake. It's wired up to power BI tools (Power BI / Snowflake activation) for sales analytics around orders, customers, and products.

Explore all tables: product-sandbox

</details>

## Check what's queryable

Ask the assistant for the queryable surface of a data product before you ask for actual numbers. You'll see available measures, dimensions, time dimensions, segments, metrics, and semantic joins.

![Claude Desktop response showing available metrics and dimensions](/files/1VF9KU3wJXyVvqnZ9j6l)

### What you can ask

<table><thead><tr><th width="216.06817626953125">Question type</th><th>Example</th></tr></thead><tbody><tr><td>Available metrics</td><td><code>What metrics does orders360 expose?</code></td></tr><tr><td>Dimensions for analysis</td><td><code>What dimensions can I use for revenue analysis?</code></td></tr><tr><td>Fields in a model</td><td><code>What fields does the orders\_enriched model have?</code></td></tr><tr><td>Model relationships</td><td><code>How are the models in orders360 related?</code></td></tr><tr><td>Capability check</td><td><code>Can this data product answer questions by customer segment?</code></td></tr></tbody></table>

### What you get back

A list of [semantic models](/consume/understand/assets/semantics.md) with their measures, dimensions, and relationships. Treat it as the menu, not the order.

## Trace lineage

Ask about lineage when you need to know where the data comes from, or what would break if a source changed.

![Claude Desktop response showing revenue lineage for Orders360](/files/u5EKxJtmi7alHGpFiGEw)

<details>

<summary>Full response (example)</summary>

Let me trace the lineage of the revenue figure.

Here's the full lineage of the **revenue figure** in Orders360:

***

**📐 Metric definition**

The `revenue_trend` business metric (monthly revenue trend by customer and product category) is powered by the **`TOTAL_REVENUE`** measure in the semantic layer:

```
TOTAL_REVENUE = SUM(orders.TOTAL_AMOUNT)
```

***

**🏗️ Where `TOTAL_AMOUNT` comes from**

Tracing back through the physical models:

**`ORDERS` (Gold table)**

> `TOTAL_AMOUNT` - *"Total order amount including tax and shipping"*

A derived column calculated from raw order fields:

| Component     | Column                  | Description                                   |
| ------------- | ----------------------- | --------------------------------------------- |
| Base price    | `UNIT_PRICE × QUANTITY` | Price per unit × quantity ordered             |
| Less discount | `DISCOUNT`              | Discount rate (0.0–1.0) applied to unit price |
| Plus tax      | `TAX`                   | Tax amount charged                            |
| Plus shipping | `SHIPPING_COST`         | Shipping cost for the order                   |

So: **`TOTAL_AMOUNT = (UNIT_PRICE × QUANTITY × (1 - DISCOUNT)) + TAX + SHIPPING_COST`**

***

**🔗 Full lineage chain**

```
RAW_ORDERS (seed CSV)
    └── ORDERS (Gold, Snowflake - incremental by order date)
            └── orders (Semantic model)
                    └── TOTAL_REVENUE = SUM(orders.TOTAL_AMOUNT)
                            └── revenue_trend (BusinessMetric)
```

The `orders` semantic model also joins `customer` (on `CUSTOMER_ID`) and `products` (on `PRODUCT_ID`), which is how `revenue_trend` can be sliced by customer segment and product category.

***

Explore the full schema: Semantics tab | ORDERS table in Workbench

</details>

### What you can ask

<table><thead><tr><th width="201.83966064453125">Question type</th><th>Example</th></tr></thead><tbody><tr><td>Upstream sources</td><td><code>Where does the revenue figure in orders360 come from?</code></td></tr><tr><td>Column lineage</td><td><code>What feeds the total\_revenue column?</code></td></tr><tr><td>Downstream impact</td><td><code>What downstream data products would break if I retire this table?</code></td></tr><tr><td>Sensitive data in lineage</td><td><code>Does this table's lineage touch any PII columns?</code></td></tr></tbody></table>

### What you get back

Enough to judge impact before you trust or change a number: a lineage graph with upstream sources, downstream dependents, transformation hints where available, an impact summary, affected data products, sensitive tables, owners to notify, and cycle warnings.

### Boundaries

* No row values, aggregates, or statistics.
* Needs an exact data product or table name. Search first if you're not sure.

## Check the table profile

Ask for a table profile when you need to know whether a table is usable: size, freshness, null rates, and value distributions.

![Claude Desktop response showing table profile information](/files/TzK7cMHzKejy3djQXcfh)

### What you can ask

<table><thead><tr><th width="173.36529541015625">Question type</th><th>Example</th></tr></thead><tbody><tr><td>Freshness</td><td><code>How fresh is the orders\_enriched table?</code></td></tr><tr><td>Row count</td><td><code>How many rows does this table have?</code></td></tr><tr><td>Null columns</td><td><code>Which columns in orders\_enriched have nulls?</code></td></tr><tr><td>Value distribution</td><td><code>What is the distribution of order\_status?</code></td></tr><tr><td>Range check</td><td><code>What is the min and max of account\_revenue?</code></td></tr></tbody></table>

### What you get back

The signals that say whether a table is fit to use before you query it: row count, column count, size, last profiled timestamp, column-level null rates, distinct counts, min, max, and distributions for queried columns.

### Boundaries

* No row contents.
* Returns the latest profile. It doesn't re-run the profiler.
* No profile-vs-profile comparison across time.

## Check quality

Ask about quality when you need to know whether configured checks are passing and the data product is safe to trust.

![Claude Desktop response showing data product quality checks](/files/VxU2fAIERPyPAlEe5dMz)

### What you can ask

<table><thead><tr><th width="211.35931396484375">Question type</th><th>Example</th></tr></thead><tbody><tr><td>Overall quality</td><td><code>Are the quality checks on orders360 passing?</code></td></tr><tr><td>Failing checks</td><td><code>What is failing on the orders data product right now?</code></td></tr><tr><td>Check detail</td><td><code>Why is the invalid\_emails check failing?</code></td></tr><tr><td>Quality by dimension</td><td><code>What is the completeness score for orders360?</code></td></tr></tbody></table>

### What you get back

A clear read on whether the data is safe to act on: a dimension-by-dimension summary (completeness, validity, uniqueness, timeliness), last evaluation time, a list of failing checks with model and column, and diagnostics for any specific check.

### Boundaries

* Doesn't modify, retrigger, or fix checks.
* Returns the latest state. Historical time travel isn't the primary path.
* Doesn't auto-include owner contact. Ask separately when you need the owner.

## Check run history

Ask about run history when you need to know whether the pipeline ran successfully and when data was last refreshed.

![Claude Desktop response showing data product run history](/files/Z4yQKH1SZzAgkthCz0hf)

### What you can ask

<table><thead><tr><th width="171.1326904296875">Question type</th><th>Example</th></tr></thead><tbody><tr><td>Latest run status</td><td><code>When was orders360 last refreshed?</code></td></tr><tr><td>Run success</td><td><code>Did yesterday's run succeed?</code></td></tr><tr><td>Run errors</td><td><code>Why did the last run fail?</code></td></tr><tr><td>Schedule</td><td><code>When is the next scheduled run for orders360?</code></td></tr><tr><td>Recent history</td><td><code>Show me the last five runs for this data product.</code></td></tr></tbody></table>

### What you get back

Enough to confirm the data is current before you rely on it: latest run status, timestamp, rows processed, models affected, environment, duration, run-level errors, and the next scheduled run when available.

### Boundaries

* Doesn't retrigger, cancel, or schedule runs.
* Doesn't return semantic query results.
* Schedule fields depend on the metadata configured for the data product.

## Typical trust flow

1. **Schema** to see what measures and dimensions are on offer.
2. **Quality** to check whether business checks are passing.
3. **Runs** to confirm data was recently refreshed.
4. **Lineage** when you need to understand where a number comes from.
5. **Table profile** when you want to check nulls or distributions before querying.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://v2.dataos.info/consume/understand/understand-and-trust-with-ai.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.