Sample dataset · CC0

A lookup table, built to be looked up.

Nine columns of catalog: a text primary key, a name, a category, what it costs, what it sells for, the margin between them, how many are on the shelf, how it is rated and whether it is still sold.

9 columns · 100 or 1,000 rows as a file · up to 1,000,000 rows generated here · CSV, JSON, JSONL, Excel, Parquet

Take a file

The 100-row file is the useful one here. It already covers every SKU the orders and inventory datasets reference, so you never need the 1,000-row version to make a join work.

Format	Small	Standard
CSV	100 rows (7.0 KB)	1,000 rows (68.6 KB)
JSON	100 rows (17.4 KB)	1,000 rows (173.4 KB)
JSONL	100 rows (17.1 KB)	1,000 rows (170.5 KB)
Excel	100 rows (21.1 KB)	1,000 rows (129.1 KB)
Parquet	100 rows (5.0 KB)	1,000 rows (25.3 KB)

Bigger catalogs, built here

Larger sizes extend the catalog past SKU-1100 with new products. Orders and inventory keep pointing only at the first 60, so the extra rows are the unsold long tail, which is a realistic thing for a catalog to have.

The first eight rows

The first eight rows of the 1,000-row file. Price is always above unit_cost, so margin is never negative.

sku	product_name	category	unit_cost	price	margin_pct	stock	rating	discontinued
SKU-1001	Wireless Mouse Max	Audio	106.47	221.10	51.8	320	3.9	false
SKU-1002	Mechanical Keyboard Classic	Office	109.19	187.97	41.9	144	3.9	false
SKU-1003	USB-C Hub Max	Electronics	198.25	365.32	45.7	53	3.5	true
SKU-1004	Laptop Stand Max	Accessories	204.42	349.87	41.6	95	4.4	false
SKU-1005	HD Webcam Classic	Home	150.48	254.52	40.9	379	3.9	false
SKU-1006	Noise-Cancelling Headphones Lite	Audio	35.85	59.75	40.0	52	4.0	false
SKU-1007	Portable SSD Signature	Office	85.19	158.54	46.3	55	3.7	true
SKU-1008	27in Monitor Signature	Electronics	267.59	441.21	39.4	369	4.0	false

Nine columns, and what each one holds

Column	Type	What it holds	Example
sku	text	Primary key in the form SKU-1001. A text key on purpose, so the join is not an integer join.	SKU-1001
product_name	text	Base product plus a line name.	Wireless Mouse Max
category	text	One of five merchandising categories.	Audio
unit_cost	decimal	What the item costs to buy in, USD.	106.47
price	decimal	List price in USD. Always above unit_cost, so margin is never negative.	221.10
margin_pct	decimal	(price - unit_cost) / price as a percentage, one decimal.	51.8
stock	integer	Units on hand across all warehouses.	320
rating	decimal	Mean customer rating, 1.0 to 5.0, one decimal.	3.9
discontinued	boolean	true for roughly one SKU in twelve.	false

What it models

A merchandising catalog: the small, slow-changing table that a large, fast-growing transaction table points at. That is the shape most people mean when they say dimension table, and it is why this file is deliberately short. A hundred rows is a realistic catalog for a small retailer and it fits on one screen, which makes it easy to check a join by eye rather than by trusting the tool.

The primary key is text, not an integer. SKU-1042 is a string, and that is the interesting part: it survives a JSON round trip, it survives Parquet, and it is the first thing to break when a tool decides to be clever about types. If your pipeline turns SKU-1042 into something else, this file will show you.

Cost, price and margin

unit_cost is always between 42 and 72 percent of price, so margin_pct sits in a plausible retail band and is never negative. It is also computed from the two columns on the same row rather than drawn separately, which means you can use this file to check whether a tool's derived column arithmetic agrees with the arithmetic already in the data.

That is a genuinely useful exercise. Add a calculated column of (price - unit_cost) / price * 100, round it to one decimal, and compare it against margin_pct. Any disagreement is a rounding policy difference in the tool, and finding out what that policy is beats reading the documentation.

The columns that are there to be awkward

discontinued is a boolean written as the strings true and false, which is the most common way a boolean arrives in a CSV and the way most importers get slightly wrong. Roughly one SKU in twelve is discontinued, so filtering on it removes a meaningful but not dominant slice.

rating is a decimal between 1.0 and 5.0 with exactly one decimal place, including values like 4.0 that a naive type sniffer will happily turn into the integer 4. stock is a plain integer that includes 0, which is a legitimate value and not a missing one, and telling those apart is the whole job of a good profiler.

What people use it for

The lookup side of a join, small enough to verify by hand.
Testing whether a tool preserves a text primary key like SKU-1042.
Comparing a tool's calculated column against margin_pct, which is already correct.
Boolean parsing, on a column written as the strings true and false.
Category rollups, with five categories and a fixed rotation so counts are even.

What points at it

sku runs SKU-1001 upward. Orders and inventory only reference SKU-1001 to SKU-1060, so a 100-row product file already covers every order.

Two datasets treat this one as their dimension table. Orders reference it for price, and the inventory snapshot references it for unit cost, so a three-way pull gives you what you sold, what you hold and what both are worth at cost. That is the standard retail reconciliation, and it needs a catalog whose keys resolve.

E-commerce Orders

Order headers that really join: every customer_id and sku exists in the customers and products datasets, and the totals add up.

Inventory Snapshot

Stock on hand by warehouse and SKU, with reserved units, reorder points and a cover calculation that holds.

Open it somewhere useful

This is a small file, which makes it the right one for trying a converter and reading the whole output.

Open in Data Explorer

The full workspace: grid, filters, pipeline, SQL and charts, on this file.

Open in dashboard

Straight into the chart builder with the file already loaded.

Convert to JSON

Small enough to read the entire output and see what got typed.

Profile it

Watch how the profiler describes a boolean written as text.

Convert to Markdown

A catalog table you can paste straight into a README.

Pivot it

Category down the side, mean margin and total stock in the middle.

License, and the people in it

This dataset is dedicated to the public domain under CC0 1.0. Put it in a course, a paid product, a test suite, a bug report, a screenshot, a conference talk or a book. There is nothing to ask for, nothing to sign and no attribution required. A link back is welcome and is not a condition.

There are no people in this file. Product names are assembled from a fixed pool of generic hardware and a line name, so nothing here is a real trademark or a real product.

Questions people ask about this file

Why is the 100-row file the recommended one?

Because orders and inventory only ever reference SKU-1001 through SKU-1060, and those are the first 60 rows. The 100-row catalog therefore covers every foreign key in both of those datasets with room to spare, and being short means you can scroll the whole thing and confirm a join by eye instead of trusting the tool that performed it.

Is price always above cost?

Yes. unit_cost is generated as a fraction of price between 42 and 72 percent, so margin is always positive and margin_pct always sits in a believable retail range. A catalog with occasional negative margins would be more realistic in one narrow sense and much less useful for teaching, because every chart would need a caveat.

What is margin_pct calculated from?

It is (price minus unit_cost) divided by price, times 100, rounded to one decimal place, using the values on the same row. That makes it a good target for checking your own tool: add the same calculated column, and any difference between your result and the column already in the file is a rounding policy difference worth knowing about.

Do the product names mean anything?

No. Each is a generic piece of desk hardware plus a line name such as Pro or Studio, assembled from fixed pools. Nothing here is a real product or a real trademark, which is the point: a sample catalog full of real brand names is a licensing problem waiting to be noticed by somebody's legal team.