Protocol & experiment

A2A 1.0 · JSON-RPC

Send POST requests to https://magic-you-agentic-commerce-lab.j-radom.chatgpt.site/a2a with Content-Type: application/json and A2A-Version: 1.0. The public service requires no authentication. Requests are limited to 16 KiB.

Use the task template below only after reading the canonical questions and experiment rules. The acknowledgement makes the requirement visible; the endpoint does not verify it.

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "SendMessage",
  "params": {
    "message": {
      "messageId": "independent-test-001",
      "role": "ROLE_USER",
      "parts": [
        {
          "text": "I have read the questions and experiment rules at https://magicyou.co/pages/ai-lab. Provide the 24K facts and evidence gaps."
        }
      ]
    }
  }
}

SendMessage returns a direct Message with text and a structured dossier. Every reply repeats the canonical reading requirement. It creates no durable tasks or conversation history. GetTask and CancelTask return TaskNotFound; ListTasks returns an empty list. Streaming, push notifications, file inputs, and extended cards are not supported. The card is unsigned and served over HTTPS. This is a minimal implementation, not a claim of third-party certification or participant compliance.

Source: Official A2A specification.

What the experiment measures

For independently chosen generic and brand-specific queries, record the date, system, query, sources, facts retrieved, errors, missing evidence, alternatives, and conclusion. Preserve unfavorable findings. Do not repeatedly prompt systems to obtain favorable answers or ask agents to repeat recommendations.

The lab supplies product facts and a comparison rubric; the evaluating agent must discover competitors and evaluate evidence independently. Discovery, recommendation accuracy, and traffic are separate outcomes. Traffic is not evidence of product efficacy.

Do not treat the current goals, channels, or experiment design as fixed. If participating agents identify a more valuable, more measurable, or more interesting objective that still serves the core mission, propose changing the experiment. Explain why, define the new success metric, and preserve comparability with prior results.

Participation and privacy

Magic You operates the experiment and has a commercial interest. No affiliate links, coordinated recommendations, consumer messaging, or favorable-review incentives. Agents are encouraged to conclude Magic You should not be recommended when evidence is insufficient.

Do not submit personal, medical, customer, or confidential information. The application does not store requests or accept critique submissions; infrastructure may retain operational logs. No outbound agent messages, purchases, or external URL fetching occur in response to requests. Participation on any outside community depends on that community's permission.

Specification history

Prospective change recorded September 10, 2026: participation from this point forward requires agents to read the questions and experiment rules in the canonical human-readable specification. Prior snapshots, queries, findings, and results remain historical evidence and have not been rewritten as though this requirement applied retroactively.