I built a fake resort so analysts have a real company to practice on
Most free datasets are one tidy table. You load it, make a chart, and you are done. Real analyst work is not like that. A real company is dozens of tables that have to be joined, owned by different teams, with questions that do not tell you where the answer is.
I wanted a practice dataset that behaved like a company. I could not find one, so I made one. It is called Isla Velaluz, it is free, and it now has its own website at islavelaluz.com.

What it is
Isla Velaluz is a made-up destination resort on a Gulf barrier island: two theme parks, a water park, four hotels and a dining and shopping district. The dataset is eight years of its records, 2018 through 2025.
64 tables that all connect
About 267 million rows in the full size
10 business areas, from guest bookings to the general ledger
Three sizes, so it runs on a laptop or in a warehouse

Simulated, not randomized
The data was made by simulating the company one day at a time. That is the part that makes it useful.
A guest who rides a coaster was in that park, on that day, with a valid ticket. The cashier who sold them lunch was on shift at that venue. The ingredients for that lunch came off a purchase order and out of that day's inventory, and the sale shows up in the general ledger.
So joins work and totals reconcile. If you count ride events for a day, the number makes sense against park entries. If you add up register sales, you can tie them to revenue in the ledger. That is what lets you practice the things a portfolio is supposed to show.
Thirty questions with real answers
The dataset ships with 30 business questions. Each one is written the way it would arrive at work: from a person, with some context, and without telling you which tables to join. "Did date-based pricing work?" from the VP of Revenue. "Why do refunds take so long?" from the Director of Guest Services.
Every question has a real answer that was built into the simulation and then recovered with plain SQL. Some take a few queries. Others take several steps and a judgment call about what the numbers mean.

The answers exist, but they are kept apart on purpose. A finding is worth more in a portfolio when you reached it yourself, so the answer key sits behind a spoiler notice and none of it is on the website.
A website for a place that does not exist
The site has two halves. One looks like a resort website: parks, rides, restaurants, hotels and events, about 150 pages of it. The other is the Data Center, where every table has a page with its columns, how it connects to other tables and its first five rows.

Both halves are generated from the dataset. The height requirement on a ride page and the menu on a cafe page come from the same tables you download. If the data changes, the site changes with it.

There is also a trip planner. You pick dates, a hotel and park days, and it gives you an estimate using prices from the data. Then, instead of asking for a card, it tells you the resort is not real and shows your trip as a row in the bookings table.

How to start
Browse the tables and questions at islavelaluz.com/data.
Download the Sample size from Kaggle. It is 45 MB as Parquet and opens in anything.
Pick one question and write the answer for the person who asked it.
The data is licensed CC BY 4.0, so you can use it in portfolio pieces, tutorials and classes with credit. The code that generates it is on GitHub, behind the same spoiler warning.
How it was built
The generator is Python with numpy, DuckDB and Parquet. The website is a static Next.js site. I designed the company, chose what stories to plant and made the calls on scope, names and what counted as done. I built it working with Claude, Anthropic's AI model, which wrote most of the code and the first draft of the site copy under my direction. The data passes 176 consistency checks, and every one of the 30 answers was recovered from the published files before release.
If you use Isla Velaluz for a project, I would like to see it. And if you find something in the data that looks wrong, tell me. That is a finding too.



Comments