You're about to sign a piece of text, but what does signing actually do, and how do we verify it is human. Well, like most technical pieces of writing, it depends. If you're on the Chrome or desktop apps all the typing you're doing is being logged, but not necessarily sent to our database. We log all evidence locally, and then only transmit the evidence required for signing. If you never sign a piece of text, data is never sent to our servers. When you do sign a piece of text we only send the minimum amount of evidence necessary for signing. We don't want to spend money storing unnecessary evidence. The exception is the website writing page. We do send that across so we can support drafts, and other helpful functionality. Once you have finished writing and click sign that triggers three things. First we send the logged data across to the backend. Then we run our Cadence algorithm. The algorithm looks for characteristics that don't line up with conventional human typing. These can be signals like excessively fast typing, or no key holding variance. Things that a human naturally does, but a fake typer does not. Finally we take the text and evidence at the time of signing and create a cryptographic signature. This signature is used to lock the text, author, and Cadence result such that it cannot be modified after the fact. We use a public and private key pair that allows us to sign text, but anyone can verify. This is very important to stop users from writing text, verifying it through Turrii, then modifying before posting. That is not legitimate and we have the signature to prove it was altered. All the signing information is in the stats section at the bottom of any verified text if you ever want to verify yourself. We also recommend pasting in the entire post with the signature to see if anything changed. Once that's done we add the signed text result to our database and publish the verify page. Then you can share the text anywhere you please! We aren't breaking any laws of physical, and we certainly aren't doing anything fancy. It's just a generic algorithm that we're constantly working to improve. Clear distinctions between AI and Human writing shouldn't have to be a black box. If you're interested in learning more please reach out. We're always happy to chat.
Ready to get started?
Download the free plugin, verify your own writing and help rebuild a feed where real thought wins.
02 / How it works
Levels of evidence
Higher levels of app integration increase our technology's confidence that a human wrote the text.
Keystroke timing was captured by the Chrome extension as the author typed, then analyzed against human capability thresholds to verify a natural writing rhythm.
- Chrome extension keystroke capture
- In-platform composition verified
- Human-like typing analysis
Why this score
- The number
- 65 out of 100. Each level has its own band: this text is at the chrome extension level, which runs 63 to 79.
- Where the typing came from
- Captured in the browser, keystroke by keystroke, as the text was typed.
- What held it here
- Typing was recorded in the browser. Browser-only evidence tops out at 79; higher scores need the desktop app's hardware keystroke check.
- What would raise it
- Running the desktop app while typing adds a hardware keystroke check and can lift a text to the desktop app level (80 to 89).
Confidence
How confident is this verification?
Typing evidence powers the confidence score.
65/ 100
Notes
- Typing rhythm is consistent with human input.
The “Typing evidence” view on the letter above marks each part of the text by how it was written, and the play button replays the typing as it happened. Here is the same thing as totals.
- Typed 100% Typed key by key with a human rhythm.
The typing engine scores the timing of every recorded keystroke. It looks at the signals below together; no single one decides the verdict. The exact measurements stay private so the check is hard to game.
- Typing speed
- How fast the text was typed, overall and in short bursts. People speed up and slow down; a script tends not to.
- Gaps between keys
- The pauses from one key to the next. Fingers have a floor; replayed or injected text often does not.
- Rhythm
- How uneven and unpredictable the rhythm is. Real typing is bumpy; machine typing is suspiciously even.
- Key presses
- How long keys stay pressed and how the fingers travel between them, when the recorder sees both key down and key up.
- Thinking pauses
- The longer pauses people take to think between phrases.
- Edits and fixes
- Deleting and rewriting. People revise as they write; pasted or generated text usually arrives finished.
Taken together, these signals read as a person typing at a keyboard.
The server signed a hash of this exact text together with its tier and author. Anyone can check the signature against the public key below. If a single character of the text changed, the content hash would not match.
Record ID
ba414ac2-1a2b-4eff-82e5-5c306d5e28b6
Author ID
204b7374-2d81-4563-a037-9dbd8fe6fd1b
Content hash
sDMROwPoTvSLjyFaEXj_p9iwbpolnlpl1wsZHesY-hY
Signature
tj9hjzjCb-fX1PYQ5KCtD5p4o1ik3m2XEPrwCQOCci2exLkFMvoIPqfStKFhxcK0hyf9dnsmNeGVQQN3IWwfBQ
Key ID
prod-ed25519-001
Algorithm
ed25519
Public key
jJa5XbW6SEMK-N98a2ikV_gEuV0xrJJUoSgw8e-h6ls
Version
3
Signed at
Nov 1, 2026, 5:27 PM UTC
Canonical payload
{"version":3,"id":"ba414ac2-1a2b-4eff-82e5-5c306d5e28b6","author_id":"204b7374-2d81-4563-a037-9dbd8fe6fd1b","tier":1,"verify_code":"3MGE5I1X","content_hash":"sDMROwPoTvSLjyFaEXj_p9iwbpolnlpl1wsZHesY-hY","key_id":"prod-ed25519-001","algorithm":"ed25519","created_at":"2026-11-01T17:27:37.388503Z","tier_code":"V1","author_identity_verified":true,"author_identity_verified_at":"2026-06-01T00:05:46.686927Z"}Start your own
Ready to get started?
Download the free plugin, verify your own writing and help rebuild a feed where real thought wins.
Start now