- [Non-Streaming TTFT Header Hydration, Raw Upstream Error Transparency & Pure Native Timing UI]:
- Non-Streaming TTFT Measurement Coverage: Fixed missing
X-Timing-Ttft-Msin Claude non-streaming responses (such as prompt compaction summaries) that caused empty ("-") first-token values in the log table; added dual-source TTFT extraction ensuring 100% TTFT availability across all streaming and non-streaming calls. - Transparent Raw Error Payload in Fallback Attempts: In
UpstreamClient, captured and sanitized raw upstream response bodies during retryable 5xx/timeout failures, directly displaying the official upstream error payload in the diagnostic timeline card. - Streamlined Native Timing Stage Labels: Completely eliminated mixed English parentheses across all Chinese timing stage labels for pure native localization; nested fallback overhead duration directly beneath the TTFT stage to keep the multi-stage progress bar perfectly balanced.
- Base64 Pass-Through for Attempt Traces: Injected
X-Timing-Ttft-Attemptsas a Base64-encoded response header, eliminating async task context drops and allowing the frontend to reconstruct complete fallback timelines under any conditions.
- Non-Streaming TTFT Measurement Coverage: Fixed missing
- [补全非流式首包时间注入、官方原始报错报文透传与纯粹中文耗时诊断]:
- 补全非流式请求首包时间度量: 彻底根治 Claude 上下文压缩提纯等非流式请求由于缺少
X-Timing-Ttft-Ms导致在日志表格中首字显示为空白（-）的缺陷；建立双重首字提取机制，无论流式还是非流式，首字延迟 100% 完整捕获。 - 官方端点原始报错报文透明展示: 在
UpstreamClient中当端点遇到网络超时或 5xx 状态码自动重试前，将 Google 官方端点返回的原始 Response Body 报错报文读取并安全脱敏后写入尝试链路，并在前端尝试卡片中直观高亮展示。 - 耗时分类诊断纯中文规范化与层级收拢: 彻底移除所有阶段标题中的英文括号后缀（中文纯中文，英文纯英文）；将降级转移损耗收拢至“等待首包”主阶段内部，并在卡片内直观区分“故障降级”与“成功命中”，保持进度条比例严格闭环一致。
- 响应头 Base64 直通端点尝试全链路: 在响应头中直接通过 Base64 编码下发
X-Timing-Ttft-Attempts，彻底杜绝跨异步任务导致的尝试记录丢失，确保前端随时随地完整还原端点链路。
- 补全非流式请求首包时间度量: 彻底根治 Claude 上下文压缩提纯等非流式请求由于缺少
Full Changelog: v4.9.8-beta.2...v4.9.8-beta.3