Misata Logo
Misata Studio
Studio by invitation

Ground truth
for AI analytics.

Misata builds company data to the numbers you set, with an answer key that is right by construction: the SQL and the value of every metric, every month. Test your AI analyst against it and see why each wrong answer is wrong, or fill an empty dashboard with a story that holds.

See a query checked
  • No real data needed
  • Same seed, same data
  • SQL, Parquet, dbt and more
New Premium datasets with the answer key Use it from Claude, Cursor or VS Code
A stack of ledger sheets with a certificate seal

A dashboard’s churn query, checked against the key

Stated, and held in the data
MRR
$50.0k in Nov 2024 to $200.0k in Oct 2026, exact in all 24 months
Churn spike
5.0% of accounts over Q3 2025
A dashboard wrote
SELECT month, leavers * 1.0 / (SELECT count(*) FROM accounts) AS churn FROM months
Does not match the key Fix proved by re-running it

Over every one of the accounts, not those still in

The query divides each month’s leavers by all the accounts in the table: those begun after the month and those gone before it included. The KPI divides by those in at the month’s start.

The demo dataset: B2B SaaS, seed 11, data to 17 Oct 2026. The verdict is the checker’s own, word for word.

How AI analytics is tested today

Most answer keys are written by hand. Many are wrong.

Gold answers written by hand

An expert audit found errors in 52.8% of the gold answers in BIRD Mini-Dev, a standard text-to-SQL benchmark.

Jin et al., 2026

A test company built by hand

Tens of thousands of lines of generators and models for one world, then something to grade the answers against it.

A model as the judge

A second model’s opinion of the first one’s answer, with a rubric someone has to keep right.

Misata’s answers are not labelled. The data is built to them.

Each number is set before a row exists, then read back from the delivered tables in SQL and shown on the certificate. There is nothing to annotate and nothing to judge.

How we verify
The checker

It doesn’t just say wrong. It says why.

Paste the SQL a dashboard tool or an AI analyst wrote. The checker runs it, compares it with the key, and when the numbers differ it names the cause and proves the fix by running the corrected query. It reads a monthly series or a single tile, and Postgres’s to_char and MySQL’s DATE_FORMAT where they mean the same.

The query

Revenue by month of issue, voided invoices included

The checker

A filter is missing; and the months come from another date

Two changes. The KPI leaves out the 64 invoices whose status is void; the query counts them. The query reads its months from invoices.issued_at; the KPI’s months are invoices.period_start.

The query

Win rate over the deals created in the month

The checker

Over every deal opened in the month, open ones included

The query divides the wins by every deal created in the month, those still open included. The KPI divides by the deals decided in the month, won or lost.

The query

Churn shown as a percentage

The checker

Another scale

The query gives the rate as a percentage; read as a fraction, every value is the key’s exactly.

The query

Churn with its start bound written <=

The checker

A row on the boundary is counted on the other side

The query compares accounts.first_contract_date with <=; the KPI with <: a row dated exactly on the boundary falls on the other side of it.

Join the waitlist Each verdict is the checker’s output on the demo dataset.
Seed data for demos and empty states

An empty dashboard sells nothing.

Describe the business and the story you want to show. Misata builds the accounts, subscriptions, invoices and events behind it, every foreign key valid and every number you stated held in the data. Change a number on the plan and build again; the same seed gives the same data, so the demo never drifts.

Join the waitlist Demo data
  • MRR exact in all 24 stated months
  • Churn of 5.0% over Q3 2025, as stated
  • Same seed, same rows, every build

From the demo dataset above.

How it works

From a sentence to a dataset you can check.

  1. A sentence, your own schema, or SQL DDL. Misata reads the numbers you state and the tables they belong in, and tells you what it assumed.

A B2B SaaS whose MRR grows from $50k to $200k over two years, with 2% monthly churn spiking to 5% in Q3 2025.
DescribeYour schemaPaste SQL DDLDraw tables
With every dataset

The proof travels with the data.

A certificate, not a promise

Every dataset lists each number you stated, the value the data holds and the SQL that reads it. Share it as a link: it lists each table’s digest, so anyone can confirm they hold the same data.

The answer key leaves with the data

answer_key.json and a long CSV: each metric’s SQL, its value in every month, what was stated and the tolerance. Load it into your eval harness or your CI.

Draw the outcome you want

Drag a curve and the data follows it month by month. Each curve names the metric in the answer key that reads it.

Your agent can use it too

Misata’s MCP server puts the engine in Claude, Cursor or any MCP client: plan a dataset, write or validate a design, generate it, query it in SQL, and read its certificate and answer key.

Connect your agent

Ships in your stack

SQL for your database, Parquet, DuckDB, dbt, Great Expectations, Prisma, DBML, TypeScript, and the recipe that rebuilds the same data any time.

Start with data that is already built

Free datasets where the tables actually join.

12 verified multi-table datasets, 377,657 rows in all. Free, no signup, public domain. Download one, or join the waitlist to make your own in Studio.

See all datasets
Premium datasets · from $15 · free previews

Datasets that come with the answer key.

Real data never tells you why things happened. These do: the cause of every claim denial, the shift behind every control-chart signal, the remaining life of every machine, the demand an empty shelf lost. Grade an analysis, a model or a class against the truth.

Compare them All four for $39
Four dataset sheets fanned out, with a key and a verified seal: the answer key comes with the data

New · pharmaceutical manufacturing

Batch Release World

$49

One oral-solids plant, five systems, two years

Two years of a pharmaceutical plant's records, from process order to QP release, and the answer key to what went wrong there.

  • ERP

    AUFK · AFRU · MATDOC · QALS · QAVE · QMEL

  • LIMS

    samples · tests · injections · audit trail

  • MES

    batch records · in-process checks · rooms

  • eQMS

    deviations · OOS · CAPA · QP register

Event logs

OCEL 2.0 · XES · one case per finished batch

Rows
661,497
Tables
65
Joins
25
Checks
133 of 133
Buy for $49Free preview and profile
Where a finished batch's time goes, stage by stagemedian90th pct
  • Order to start12.8 d
  • Manufacture2.5 d
  • Bulk testing to decision4.9 d
  • Wait to pack9.3 d
  • Packaging0.3 d
  • Packaging record review1.5 d
  • To QP certification0.2 d

In the answer key

  • Assay testing went from 1.3 to 3.5 days while two HPLCs were down

  • 83% of batches made with coarse API went past dissolution stage 1, against 12% with fine

  • One analyst reintegrated 18.2 chromatograms per 100 tests; the next highest 3.7

  • Weekly sales, at regular price and on promotion

    Retail Data Warehouse

    $15

    Grocery chain, 10 stores, fiscal 2024 and 2025

    Two fiscal years of a grocery chain, from the shelf to the receipt

    Answer key

    • baseline_units
    • price_effect
    • promo_lift
    • +7 more
    Rows
    2,790,852
    Tables
    14
    Checks
    117 of 117
    Free previewProfile and preview

  • An X-bar chart with the truth underneath: flatness, characteristic 139

    SPC for Precision Machining

    $15

    3 plants, 36 CNC machines, six months

    Six months of control charts with the ground truth underneath

    Answer key

    • active_cause_id
    • true_cause_shift_sigma
    • true_tool_wear_sigma
    • +6 more
    Rows
    515,442
    Tables
    15
    Checks
    101 of 101
    Free previewProfile and preview

  • Vibration through a life that ends in failure

    Predictive Maintenance Fleet

    $15

    240 machines at 4 plants, two years

    Two years of sensor readings, failures, repairs and costs for 240 machines

    Answer key

    • failure_mode
    • true_life_days
    • true_failure_at
    • +6 more
    Rows
    179,038
    Tables
    6
    Checks
    67 of 67
    Free previewProfile and preview

  • First-pass denial rate, by month of service

    Healthcare Claims and Denials

    $15

    10 payers, 8 facilities, calendar 2025

    A year of revenue cycle with the reason behind every denial

    Answer key

    • cause_coverage_inactive
    • denial_cause
    • preventable
    • +6 more
    Rows
    356,609
    Tables
    8
    Checks
    95 of 95
    Free previewProfile and preview
Questions, answered plainly

What people ask first.

Misata builds connected company data to the numbers you set, such as MRR from $50k to $200k with a churn spike in Q3, and gives you the answer key: the SQL and the value of every metric in every month. Teams use it to test AI analysts, text-to-SQL agents and dashboards against known answers, and to fill demos and empty states with data that tells a planned story.

State the numbers.
Get the answer key.

Connected tables built to your numbers, a certificate that each one holds, and the key to check every query against.

Browse free datasets