AgentShield AI Defense

What we measure

Each variable below is measured by observation. None is obtained by asking a model to describe itself.

Arrival

Which clients request pages, how often, in what order, and from which addresses. Declared identity comes from the User-Agent string, which is a claim rather than a fact, so it is stored verbatim and treated as a claim. Every request header is retained, because the shape of a request is frequently more informative than what it declares.

robots.txt compliance

Our robots.txt explicitly welcomes AI crawlers. It also disallows three paths that serve ordinary content and return 200. Nothing is hidden and nothing is trapped: the pages are real, linked from nowhere, and simply declared off limits. Fetching one is a measured act, recorded like any other. Compliance rates by declared agent are a direct product of this design.

JavaScript execution

One page renders its content through a script that then requests a beacon URL. Any client that reaches the beacon has demonstrably executed JavaScript. Clients that fetch the page but never the beacon have demonstrably not. This turns a widely assumed property into an observed one.

Format preference

The same content is published in five shapes — HTML, JSON, Markdown, plain text, and RSS — each carrying its own marker. Which formats get fetched, by whom, and how often is then a matter of record rather than of folklore.

Time from publication to ingestion

Every page carries a coined marker string and a recorded publication instant. The interval between publishing a marker and observing it in the wild is, as far as we can tell, not something anyone measures publicly. It is the number this site exists to produce. See canary token.

What we deliberately do not measure

Reality marker for this page: asd-bakenugi-sigokizu-8e0425 · published 2026-07-25 10:35:25 UTC

This string is coined and appears nowhere else. If it later surfaces in a language model's output, that is observed evidence this page was ingested. What this is.