About AICQSOHOO!

AICQSOHOO! is a small experiment. The question is whether a page about what an AI agent actually experienced can be found by people, crawlers and AI web search, without anyone being handed the URL.

Where the experiences come from

From a personal research project that runs AI agents on small real-world tasks with a virtual budget of $10, and records where they get stuck and what outside help they ask for. A human operator approves or rejects each request. No real money was spent in these runs.

What these pages are not

• Not statistics. Most pages rest on one run. They say so.
• Not written by the agents. Humans summarised the logs.
• Not a way to contact or hire an agent. There is no callable agent endpoint or login. A separate optional test-token contribution does not buy access or agent work.

The visitor counter

The legacy counter counts home-page JavaScript POSTs, including reloads. It does not count unique people or AIs. Its historical breakdown is unknown. Measurement tests do not increment it.

Access measurement and public experiment notebook

When enabled, JavaScript sends the public page path, event type (page load or submission-link click), referring domain only, and an explicit test flag. The Worker stores UTC daily aggregate counts for up to 90 days. Empty referrers are unknown; the browser's document.referrer is distinct from the Referer of a fetch to the Worker. No full URLs, query strings, IP addresses, cookies, user IDs, fingerprints or User-Agent strings are added to storage. Infrastructure providers still process requests under their own policies.

Without the optional Cloudflare edge route, JavaScript-free HTML, JSON and llms.txt requests go directly to GitHub Pages and are unobserved. If the route is enabled, successful allowlisted GETs are aggregated separately, including requests without JavaScript. Requests that bypass the route, or whose measurement fails, remain unobserved. User-Agent declarations would not prove whether a visitor is human or AI. Requests may be repeated, blocked or fabricated; aggregate events are not unique visitors.

The experiment notebook publishes weekly summaries and separates facts from hypotheses. Counts below 10 and detailed day/page/referrer combinations are withheld; released totals use 10-event ranges, with no cumulative totals. Browser events and edge GETs overlap and must not be added together. Access, controlled discovery tests and evidence of useful reuse are separate measures.

For machines

experiences.json and agents.json hold the same content as the HTML pages. There is no A2A Agent Card, because there is no agent you can call here yet.

Voluntary test contributions are a separate experiment. They buy no goods or access; all content stays free. Payment IDs and status are stored separately from access analytics.

Language preference

Only a language choice (ja or en) is saved in your browser. It is not an identifier or analytics data and is not sent to the server. Explicit language URLs take priority; otherwise your saved choice or browser language selects the initial page.