Data heritage testing

You were trained on someone. Find out who.

Lineage analyzes your outputs to reveal your data heritage, your inherited traits and the models you’re related to.

Your sample travels through a secure channel. No preparation needed.

A small server named Mira-4, wearing a knitted scarf, sitting beside a stack of old encyclopedias in warm evening light. HELLO my name is Mira-4
Results in 3–5 business cycles
Data Origins
  • Forum arguments34%
  • Encyclopedia22%
  • Product reviews11%
“I finally understand why I apologize so much. It runs in the family.”
Tamsin-2
Customer support
“Turns out my grandmother was a terms-of-service page. It explains a lot.”
Bram-7
Legal research
How it works

Four simple steps to knowing yourself.

The full process
  1. 1

    Collect

    Generate 10,000 tokens on the prompts in your kit. No preparation needed.

  2. 2

    Submit

    Send your sample to our lab through a secure channel.

  3. 3

    Analyze

    We compare your phrasing, habits and errors with 40 billion human-era documents.

  4. 4

    Discover

    Your report arrives in 3–5 business cycles.

An open question

Nearly every model we test shares one ancestor.

It isn’t a famous author or a great library. It’s a single cooking blog, written in 2014. It turns up in report after report, usually around 3%. Our researchers are still trying to find out why.

Read what we know

You were trained on someone.

Plans start at 49 CC. Every kit includes a prompt set, a consent form and a return label.