Pepperwind Trading Co.

A fictional company, built as data. Every person, order, invoice and journal entry in it is synthetic, generated by Misata. This repository is the free preview. The full company is available on misata.studio for $30.

Try it before you download: query a slice in your browser with DuckDB, nothing to install, and check your answers against graded questions.

Pepperwind is a spice and tea trader from Kochi that sells in India, Europe and the United States. Over thirty-six months, from October 2023 to September 2026, it runs eleven business systems: a storefront, a subscription club, a warehouse, a general ledger in three currencies, support, CRM, marketing, product analytics, HR and payroll, and company mail and chat. The systems share keys, so a single order can be followed from the customer's session through payment, picking, shipping, the ledger and any support ticket it caused.

Most sample databases are one schema made for one tutorial. Pepperwind is closer to what you meet at work: several systems that each hold part of the story and do not always agree. It comes with an answer key, so you can check your results. Pepperwind is not real data, and nothing in it describes a real company or person.

What people use it for

• Dashboards and BI demos that need more than one fact table
• SQL, analytics engineering and dbt courses with joins across systems
• Warehouse and pipeline projects: staging, deduplication, slowly changing dimensions, currency conversion
• Finance exercises: reconciliation, intercompany, tax and currency
• Churn, fraud and forecasting models on events with a known cause
• Testing AI analysts and text-to-SQL features against questions with known answers

Preview and full pack

	Free preview (this repository)	Full pack
Customers	1% of consumers; B2B accounts and HR in full	all
Tables	223	223
Rows	1,818,718	79,233,785
Months	36	36
Answer key	computed on the sample	complete
KPIs with their values	on the sample	92, every month
Graded questions	129 development questions	420: 138 development, 282 held-out test
Licence	CC BY 4.0	commercial use, no resale
Price	free	$30, one payment

The preview is built the same way as the full pack, with the same tables and columns. Its answers are computed on its own rows, so they differ from the full pack's.

Systems

System	Tables	Rows in the preview	What it holds
core	16	18,244	markets, organisation, people, products, SKUs, price lists, warehouses
commerce	12	89,320	customers, carts, orders and lines, promotions, referrals, returns, reviews
billing	11	75,832	the subscription club: prices, subscriptions, invoices, payments, disputes, refunds
logistics	17	320,183	stock, transfers, packing runs, club boxes, shipments, tracking, RTO
ledger	93	266,688	general ledger by entity (India, EU, US, cross-border), intercompany, tax, duty, FX
crm	13	59,822	B2B accounts, contacts, leads, opportunities, quotes, territories
marketing	8	121,219	campaigns, sends, email events, consent, UTM, daily ad spend
product	10	422,066	web and app sessions, events, identity stitching, experiments, attribution
support	13	33,179	tickets, events, messages, agents and shifts, SLAs, CSAT, knowledge base
hr	13	233,316	employment, contracts, compensation, payroll, attendance, leave, reviews
comms	17	178,849	email, chat, calendar, documents, policies, consent, DSARs, audit log

Quick start

Each table is its own config in the viewer above, named <system>.<table>.

from datasets import load_dataset

orders = load_dataset("misatadata/pepperwind", "commerce.order", split="train")

Or query the Parquet files directly with DuckDB:

import duckdb

duckdb.sql("""
    SELECT channel, currency, count(*) AS orders, round(sum(total)) AS revenue
    FROM 'data/commerce/order.parquet'
    WHERE cancelled_at IS NULL
    GROUP BY ALL
    ORDER BY currency, revenue DESC
""").show()

Files

• data/<system>/<table>.parquet: every table, one Parquet file each
• answer_key/: the answer key on this sample: what was planted where, the incidents, and every KPI's value
• tasks/public_dev.jsonl: the development questions with their gold answers
• docs/: the data dictionary, one page per system, and dictionary.json
• pepperwind_preview.zip: everything above in one download
• SHA256SUMS: a checksum for every data file

Is any of it real?

No. Pepperwind Trading Co. is fictional and every row is synthetic. No real company, brand, person, address or phone number is used. Every email domain ends in .example, and every IP address is in a reserved documentation or benchmarking range. Its patterns come from the business plan it was built to follow, not from any real company's records, so use it to build, teach, test and demonstrate, not to draw conclusions about real markets.

Before buying the full pack, read what it is and is not. Start with this preview: if it does not fit what you are building, the full pack will not either.

Licence and citation

This preview is released under CC BY 4.0. Credit it as "Pepperwind Trading Co. by Misata". The full pack has its own licence, which allows commercial use and does not allow reselling or redistributing the dataset itself.

@misc{pepperwind2026,
  title  = {Pepperwind Trading Co.: a synthetic multi-system company dataset},
  author = {{Misata}},
  year   = {2026},
  url    = {https://huggingface.co/datasets/misatadata/pepperwind},
  note   = {Version 1.0}
}

Need data shaped to your own business? Misata builds datasets from a description or your schema, with the answer key.