Drive any consumer-facing site with an LLM, capture trackers + screenshots
at sensitive moments, produce a sealed evidence chain.
Run
Full URL including scheme. The agent enters here and navigates from there.
Drives a checkout flow with a test persona, stops before payment submit.
The agent flags verbatim language matching the selected buckets at every page it captures. Conservative — puffery is rejected. Skip if you don't need this.
Listens to every page in the agent's browser context and records hits against the
known-trackers catalog (Google Analytics, Meta Pixel, LiveRamp, Tapad, etc.) with
URL, method, and frame attribution. Powers the pixel-leak / CIPA case-line evidence
in the litigation PDF. Uncheck if you only need representations or the consent-UX audit.
Compares the verbatim representation against Wayback Machine snapshots. Fast mode answers "still up?" and "always been there?"; Full mode also bounds the date the rep was first added — useful for class-period definition. Requires at least one bucket above.
Defaults to $AGENT_MODEL_WEB or DeepSeek v3.
Each run produces hashed + RFC 3161 timestamped evidence at
/evidence/<crawl_id>.