Realistic UK data, with the dirt left in

Synthetic datasets with typos, gaps and duplicates planted on purpose, and an answer sheet of every one. No real people.

The data you need is the data you can't have

Real data is dirty. But the data people test and learn on rarely is.

Locked

Real data can't be shared

Customer and patient records are protected, so they can't go to a test team, a supplier or a classroom.

Too clean

Test data is tidied up

Sample and teaching datasets arrive already cleaned, so the hardest part of real data work never gets practised.

Rare

Fraud is hard to find

Fraud is a small share of real records, so there are never enough labelled cases to test against.

So models, pipelines and students meet dirty data for the first time when it matters most.

Everything real data has, except the risk

Data that behaves like the real thing, that anyone can use.

customers.csv0 real people
Eleanor Vance · BelfastinventedMarcus Bell · LeedsinventedPriya Nair · Cardiffinvented

Share it with anyone

Every record is invented, so datasets can go to test teams, suppliers and classrooms.

How dirtyyou choose
LightModerateHeavy
Eleanor Vnace12/03/2026

Dirty on purpose

Typos, gaps, duplicates and broken formats, as light or heavy as you need.

is_fraudthe rate you set
Labelled on every row

Fraud when you need it

Choose how much of the data is fraud, so there are always enough cases to test against.

Answer sheet9 patterns
Format drift337 rows · 3.4%
Typos170 rows · 1.7%
Duplicate records62 rows · 0.6%

Know exactly what's wrong

Every dataset comes with an answer sheet, so you can see what was caught and what was missed.

Regenerateidentical
Run 1 · seed 42same rows
Run 2 · seed 42same rows
Run 3 · seed 7fresh data

The same data, every time

Keep the settings to rerun a test on identical data, or change them for a fresh set.

UK formatsbuilt in
40-28-19BS1 4DJ943 476 5919£452.0007700 900123

Built for UK data

Sort codes, postcodes, NHS numbers and pounds, in the formats UK systems really use.

Two workspaces, one generator

Answer sheet9 patterns
Format drift337 rows · 3.4%
Typos170 rows · 1.7%
Duplicate records62 rows · 0.6%

For data and fraud teams

Test fraud models, pipelines and cleaning tools on data that behaves like production, without touching customer records.

Open the industry workspace
ClassPublished
Cleaning insurance claims
Wed 10:00 · 2 hrs · 1 dataset · 1 notebook
18 of 24 enrolled

For lecturers and students

Teach data cleaning with a dataset per class, the answer key, a notebook sized to the session and a model to beat.

Open the universities workspace

Frequently asked questions

The short answers. For anything else, get in touch.

It makes realistic synthetic UK datasets with real-world data problems planted on purpose, plus an answer sheet of what was broken. Teams use it to test models and data pipelines; lecturers use it to teach data cleaning.

Make your first dirty dataset

Realistic synthetic UK data with the dirt left in, and the answer sheet to check against. For data teams and universities.

Part of

© 2026 Dirty Data Generator, a FinTech West productSynthetic data only. No real customer, patient or student records.