How do I get labelled training data with the failures I need?
Describe the system and the rate of the event you care about, for example 60 wind turbines with gearbox failures at 2%, and Misata Studio generates the time-stamped data with those events injected at exactly that rate and labelled, so you can train and honestly evaluate a model. It scales to millions of rows and exports to Parquet, Arrow Feather, JSON Lines or CSV.
Best for: ML and data science teams who need labelled examples of rare events that real data lacks or cannot be shared. Updated 2026-09-21.
Sensor data for 60 wind turbines with gearbox failures injected at a 2% rate over three years, labelled.
- Exact event rates
- Ground-truth labels
- Time-series that behave like the machine
- Parquet, Feather, JSON Lines or CSV
Why it holds up
Known ground truth
You know which rows are failures and why, so recall and precision are measured against the truth.
Rare events on demand
Ask for the case you have never seen, at the rate you choose.
Scale
Generation shards to millions of rows with totals still exact across shards.
Questions
- Can I control the failure or fraud rate exactly?
- Yes. A rate you state is applied exactly and listed in the certificate.
- Is the data labelled?
- Yes. Injected events carry their labels, which act as ground truth.
- How large can the dataset be?
- Runs have been generated at about 10 million rows in under a minute, in sharded Parquet.