The Humanity Score

Humanity score / noun / — a measure of the probability that a real human is behind an account rather than a bot, scraper, or automation framework. A humanity score is computed from evidence such as device rendering signals, automation framework signatures, and the timing and rhythm of activity, so it verifies humans without asking them to prove anything.

Dregs computes a Humanity score from 0 to 100 for every user, continuously, as one of its four identity scores. Because the score is built from behavior and device attributes, not puzzles, it adds no friction for your real users and never makes a paying customer identify cars and traffic lights.

The Bot Detection Signals Behind the Humanity Score

Bots have gotten very good at the moment of signup and stayed surprisingly bad at everything after it. The Humanity score watches the whole lifetime of an account for the machine tells:

How Dregs Computes the Humanity Score

A pipeline of analyzers runs over each identity's events and devices. Each produces observations with a value from 0.0 (suspicious) to 1.0 (legitimate), a confidence, and an explanation; the Humanity score is the weighted aggregate of the Humanity-category observations, scaled to 0–100.

{
  "category": "HUMANITY",
  "id": "humanity.automation-signals",
  "label": "Automation Signals",
  "value": 0.1,
  "confidence": 0.9,
  "explanation": "Device exhibits headless browser characteristics",
  "metadata": { ... }
}

Because scoring is continuous, a bot that cleared signup still gets caught by what it does next, and a real user who tripped one weak signal recovers as normal activity accumulates.

A CAPTCHA Alternative That Doesn't Punish Real Users

CAPTCHAs tax every legitimate visitor to inconvenience an attacker who pays a solving farm fractions of a cent, or whose AI now solves the puzzle faster than your customers can. The economics have inverted: the humans do the suffering and the bots pass.

Continuous Human Verification, From the First Event

Continuous human verification flips the model. Let everyone through the door with zero friction, score them from their first events, and apply consequences only to accounts that behave like machines: badge them, alert on them, or push the verdict to your application by webhook and gate features automatically. The bot detection use case walks through the full pattern, and device fingerprinting covers the device-level signals underneath it.

The Other Three Scores

Humanity answers "is this a person?" But a person can still be a fraudster, and the other three dimensions cover what a human abuser looks like.

All four come from the same identity scoring pipeline, on every plan, from the events you're already sending.

Frequently Asked Questions