Glossary
The terms reports use
Reports describe what one controlled visit observed, in the plainest language the evidence allows. These are the technical terms that still appear, each defined once here so every report can link to it.
Brave Shields
The ad and tracker blocker built into the Brave browser. Reports run every observed request through Brave's own open-source ad-block engine (adblock-rust) with Brave's default filter lists. Only network requests are matched: cosmetic filtering, which hides page elements, is not simulated.
“Matched Shields filter lists”
The number of a page's requests that matched Brave's default filter lists while the page loaded normally (nothing was blocked). It is one of three distinct Shields numbers a report can carry, and they are deliberately never blended: matched requests (what the engine would target on this visit's traffic), requests the engine actually aborted in a Blocker-comparison visit with blocking active, and the total drop in third-party requests between the paired visits. The total drop is usually the largest: it can include follow-on requests that never started once their sources were blocked, plus ordinary run-to-run variance between the two visits.
Global Privacy Control (GPC)
A signal the browser sends with every request asking the site not to sell or share your data. It carries legal weight in several US states, including California. The GPC diff mode visits the page with and without the signal to show what differed between the two visits; request counts cannot show whether data sales stopped.
Third party
Any domain other than the site you scanned and its subdomains. A third-party request is not automatically tracking: content delivery networks and embeds are third parties too. That is why reports separately label known services.
Known service
A third-party domain that matched a curated catalog of recognizable advertising, analytics, social, and session-replay services. The catalog is a lower bound: a third party the catalog does not name can still track.
Fingerprint-like calls
Calls to browser APIs (canvas, WebGL, audio, WebRTC) that can help distinguish your device from others. Many uses are legitimate, such as charts and media, so reports count the calls, flag matched behavior patterns, and treat them as review prompts rather than proof of tracking.
Provenance
The causal chain behind a request: which script started it, and which script injected that script. It answers why the page contacted a domain instead of only recording that it happened.
Keystroke capture test
The scan types a synthetic sentinel value into the page's form fields (never submitting) and watches outgoing traffic for that value, including base64, hex, and hashed encodings. It covers fields on the loaded page, not flows behind logins or extra steps.
Advertising pixels
Tracking pixels from platforms like Meta, TikTok, and X. Reports decode which events each pixel fired (PageView, Purchase, and so on; a site-defined event name is generalized to "custom event" rather than stored) and whether personal-identifier fields were attached. Detection checks that a known identifier parameter carries a non-empty value; the value is inspected only transiently and never persisted, exposed, or semantically interpreted, so the platforms' documented hashing is not verified.
Consent comparison
Two paired visits: one clicking "Accept all" on the cookie/consent banner and one clicking "Reject all" (recognized banner controls or an exact accept/reject label, first layer only). The diff shows what differed between the visits. Legacy v1 reports record only that a click was dispatched; r2 reports also record bounded consent-platform readbacks and say whether registration was verified or contradicted, only a weak banner-transition signal was seen, or verification was unavailable or failed. Even a verified r2 visit records requests from before and after its click, so traffic can still be pre-choice, strictly necessary, or processing claimed under legitimate interest. A visit where no control was found stays pre-consent and the report says so.