先检测体验，先检测体验，再核对来源。再核对来源。

1保存一份最小测试请求Save a minimal test request

将模型占位符替换为分组中实际可用的模型，保存为 UTF-8 的 request.json。Replace the model placeholder with an available model and save request.json as UTF-8.

{
  "model": "YOUR_MODEL_ID",
  "max_tokens": 64,
  "stream": true,
  "messages": [
    {
      "role": "user",
      "content": "Reply with: connection-ok"
    }
  ]
}

2原样保存完整响应Save the complete response unchanged

先在当前终端设置 PORTAL_API_KEY。命令适用于 Bash / Zsh；Windows 可在 WSL / Git Bash 执行。使用 --output 保留字节，不要用编辑器重新排版响应。测试调用按实际规则计费。Set PORTAL_API_KEY in the current terminal. Use Bash / Zsh, or WSL / Git Bash on Windows. Save with --output and do not reformat the response in an editor. Test calls use your normal billing rules.

curl --no-buffer --silent --show-error "https://newapi.world/v1/messages" \
  -H "Authorization: Bearer $PORTAL_API_KEY" \
  -H "Content-Type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  --data-binary @request.json \
  --output response.sse

预期 SSE 末尾包含 event: tee.proof。若网关返回 tee.proof_unavailable 或没有证明，结果应为缺少证明，不能判定模型真假。此通用示例未传入客户端 nonce；需要验证本次请求新鲜性时，先按实际网关支持的方式传入随机挑战。Expect event: tee.proof at the end of the SSE. Missing evidence or tee.proof_unavailable means unverified, not a counterfeit model. This generic example does not send a client nonce; use your gateway’s supported challenge mechanism to establish freshness for your request.

3交由第三方逐项核对Check each item with an external verifier

打开第三方验证器，手动载入完整响应，并核对自己保留的请求、挑战与独立 PCR0 基准。不要仅凭页面总状态或响应里的 model 名称判断。Open the verifier and manually load the complete response. Compare against your retained request, challenge, and independent PCR0 reference. Do not rely only on an overall status or the response’s model name.