<!-- Source: https://www.misata.studio/docs/studio/use-cases -->
<!-- Docs index: https://www.misata.studio/llms.txt -->

# What you can do with Misata

Find what you want to do. Each section says which tool does it, and how, step by step.

| I want to | Use | Section |
|---|---|---|
| Fill a dashboard, an app or a demo with data that tells a story | Studio | [1](#1-sample-data-that-tells-a-story) |
| Test whether an AI analyst, a text-to-SQL model or a dashboard gets the numbers right | The answer key and checker | [2](#2-test-whether-something-gets-the-numbers-right) |
| Show a story coming out of a real data pipeline | Backlot | [3](#3-a-demo-on-a-real-pipeline) |
| Have an AI assistant make and check data in a conversation | MCP server | [4](#4-data-from-an-ai-assistant) |
| Build any of this into my own product | Platform API | [5](#5-misata-inside-your-own-product) |
| Practise or teach analytics on data with known answers | Studio and the answer key | [6](#6-practise-or-teach-on-data-with-known-answers) |
| Just get a dataset now | Free datasets | [7](#7-a-dataset-right-now) |

## 1. Sample data that tells a story

**When:** an empty dashboard, a new app with nothing in it, a demo, onboarding, a screenshot for a launch.

1. In [Studio](/app), choose **New dataset** and describe the business with the three or four numbers your story turns
   on: *"An online store selling home goods in 2025. About 8,500 orders, revenue from $40k to $95k a month, a holiday
   peak in November and December, 4% of orders refunded."*
2. Read the plan: the tables, their sizes, every number to be held. Fix anything there.
3. Build, then open the **Story** tab to see each number beside what the data holds.
4. **Export** to where your dashboard or app reads data: Postgres, MySQL, SQL Server, Snowflake, BigQuery, ClickHouse,
   DuckDB, SQLite, Parquet, CSV, Excel or dbt seeds.
5. Need it different? On the Overview, write what is wrong (*"far fewer customers"*), draw a curve on the **Curve** tab,
   or turn a control on **Adjust**.

Have your own tables already? Choose **Import a structure** on New dataset (SQL DDL, dbt, Prisma, Django, Rails,
OpenAPI, DBML, CSV headers and more), then state the numbers on your own columns. See
[Getting realistic results](/docs/studio/realistic-results).

## 2. Test whether something gets the numbers right

**When:** you build or evaluate anything that answers questions about data, an AI analyst, a text-to-SQL model, a
dashboard generator, a semantic layer, or a team's own dashboards, and you need to know how often it is right and why
it is wrong when it is not.

1. Build a dataset with a story that covers the questions you care about (growth, churn with a spike, revenue by
   segment). A built-in business (SaaS, e-commerce, banking and payments, health systems) gives the fullest
   [answer key](/docs/studio/answer-key).
2. Load it where your system reads data, or query it in place through the API.
3. Take questions from the key: the Story tab lists them, and the API and MCP serve them.
4. Ask your system each question. Collect the SQL it wrote, the numbers it showed, or both.
5. Paste each into the **checker** on the Story tab (or send it to the API's check). Each comes back right, or wrong
   with the mistake named: churn over every account rather than those at risk, revenue by issue date with voids counted,
   win rate over every deal opened.
6. Change the story or the seed and run it again. Every dataset has its own key, so nothing is ever graded on data it
   has seen before.

## 3. A demo on a real pipeline

**When:** the thing you are showing is a pipeline (a dbt project, a semantic layer, a BI tool on modelled tables), and
the numbers must come out of its models, not be pasted into a chart.

1. Open [Backlot](/app/backlot). Start from a template (subscription revenue, a storefront, payments and disputes, a
   support desk) or bring your own dbt project or Postgres schema.
2. Write the scenes: a number, a window and a shape (*"collected revenue from 42k to 96k over the year, a 25% drop in
   month seven"*).
3. **Run the whole story**: Backlot builds the source data, runs the pipeline, measures each scene on its output and
   corrects the source until every scene holds within 2%.
4. Share the **room**: a link with the story, what the pipeline measured, and a read-only SQL console. Reset it before
   each demo.

See the [Backlot guide](/docs/studio/backlot).

## 4. Data from an AI assistant

**When:** you work in Claude, Cursor or VS Code and want the assistant to make data, query it and check answers without
leaving the conversation.

1. Connect the [MCP server](/mcp) (one line for Claude Code, a URL for Claude, a small file for Cursor or VS Code).
2. Ask for what you need with the numbers that matter. The assistant designs the tables; Misata holds the numbers and
   returns what it checked.
3. Ask the assistant to show the verification, query the data, export it, or grade answers with `check_answers`.

See the [MCP guide](/docs/studio/mcp).

## 5. Misata inside your own product

**When:** your product should create datasets for its own users: sample data in empty states, a sandbox per user, data
that changes when a user asks, or a check of numbers your product computes.

1. In Studio, open **API** and create a key.
2. `POST /v1/datasets` with a prompt, your own tables with their numbers, or a built-in business. Follow it by webhook,
   by polling, or by streaming its steps.
3. Take the data: files, a direct load into Postgres or ClickHouse, or read-only SQL on the dataset where it is.
4. Let users change it in their own words: send what they said as a **revision**. A new version is built and checked;
   the first is never touched.
5. Check what your product computes with `POST /v1/datasets/{id}/check`.
6. Set **instructions** once (*"company names fictional, amounts in euros"*) and every dataset on the account keeps to
   them.

To make it feel instant, keep a few ready datasets (built ahead with a fixed seed) for the businesses your users ask
for most, and show one while a user's own builds. The same request and seed always make the same data.

See the [platform API guide](/docs/guides/platform-api).

## 6. Practise or teach on data with known answers

**When:** a course, a workshop, an interview exercise or a team learning SQL and metrics.

1. Build a dataset with a story worth investigating (a churn spike, a margin slip in one region).
2. Hand out the tables and the questions from the answer key, without the answers.
3. Check each answer with the checker. A wrong one is named for the mistake it is, which is the lesson.

## 7. A dataset right now

**When:** you need realistic multi-table data this minute and nothing specific about it.

Browse the [free datasets](/datasets): every column profiled, rows previewed before download.