Back to lab home

Findings

Reproducible observations from the lab.

Finding 001: AI bot identification requires multi-signal evidence

Baseline method for combining user agent, network, resource, JavaScript, and timing signals.

Finding 002: OpenRouter claims need direct lab evidence

OpenRouter responses can claim retrieval without matching origin hits, so direct lab evidence is the stronger source.

Finding 003: per-model OpenRouter probe attribution

Per-model URLs and structured test metadata make future origin hits attributable to a specific model attempt.

Finding 004: OpenClaw web fetch reading baseline

A lightweight fetch-and-read surface sees server-rendered text, but not JavaScript-rendered content or PNG image text.

Finding 005: local HTTP directive compliance baseline

A policy-naive local HTTP client fetches all directive fixtures while preserving robots.txt and robots meta evidence for later comparisons.

Finding 006: ChatGPT manual dual-stack reading baseline

ChatGPT needed standard dual-stack HTTPS reachability, then behaved like an HTML retriever: one page request per prompt, no JavaScript, no image fetch, no hidden href exposure, and no depth-link follow.