Writing proof from your typing record

Why not rely on detection

Core records which parts you typed and which you pasted. This is the provenance. Rather than guessing, it shows from the record exactly which parts you wrote. The backbone of the certificate is a record — live input on this device plus a signature — not an inference.

What the certificate contains

The certificate ties these four to the text and signs them together:

• Content hash: a code computed from the text. Change the text and the value changes, so verification fails.
• Provenance: the breakdown of which parts are typed, pasted, or from AI.
• Typing rhythm: how you typed while writing.
• Signature: signed with Core’s private key (Ed25519) to prevent tampering.

Seeing the difference from copy-typing

Even when provenance shows text was typed, someone could fake it by retyping AI-generated text by hand. Still, composing your own text and copying out existing text leave structurally different traces of revision and rhythm — a finding from writing-process research (Fuchs & Krivokapić 2016; Crossley et al. 2024). Core uses this tendency to corroborate provenance.

• Rhythm variation: how much the keystroke intervals swing. Composing while thinking swings; copy-typing tends toward a steady pace.
• Revision: how often you go back to fix text you already wrote. Copy-typing shows little.
• Deletion rate: how much you write and delete. Copy-typing shows little.

This is not an AI detector — it shows traces that rarely arise when copying. We confirmed the tendency per writer on the public KLiCKe corpus (~5,000 people), but it was checked on short copy tasks in English, and long retyping or deliberate faking has limits.

Reading the result

When typing is the main source and the rhythm looks human, a wax-seal check mark appears with the verdict “Likely written by a person.”