v2.0.2

What's Changed:

This release improves API Key filtering and request details, fixes token accounting and quota state updates, and makes quota history faster and more accurate.

• API Key filters now show keys with recorded usage for the active page and time range across Overview, Realtime, Analysis, and Request Events. Valid selections are retained when switching ranges, and background refreshes keep existing options visible. (#675)
• Request details now record, display, and export the effective service tier reported by CPA. Speed Mode prefers this tier and supports Ultrafast labels. Request endpoint aliases are also displayed consistently across request lists and credential details, while preserving raw endpoint values in tooltips and exports. (#669) (#672)
• Improved Codex and Claude quota-history loading and interaction performance when many cycles and requests are present. Fixed request attribution around daylight-saving transitions and fractional-second boundaries, and added UTC offsets where timestamps would otherwise be ambiguous. (#673)
• Fixed Codex quota state remaining marked as limit reached after response headers explicitly report recovery. Main quota windows now receive consistent limit-state updates, and older pending updates cannot overwrite newer refresh results. (#676)
• Fixed token accounting for plugin adapters and other unrecognized executors when CPA supplies a complete, valid token breakdown. Cache reads, cache creation, and reasoning tokens are normalized before cost calculation and aggregation, addressing cases where cache-read totals exceeded input totals. (#671)
• Protected provider configuration saves against concurrent changes when supported by CPA. Conflicting saves now report an error, and older background metadata snapshots no longer overwrite successful credential changes. (#670)
• Streamlined Top Models by removing its summary toolbar. Analysis model-speed legends now support highlighting through hover, keyboard focus, and click or tap, with fixes for stale selections and tooltips after query changes. (#668)
• Updated CI and release builds to follow the latest Go 1.26 and Node.js 24 patch releases, refreshed build runners and workflow dependencies, and enabled pulling updated Docker base images during builds. (#664)

Historical requests are not backfilled with effective service tiers; requests without this information retain the existing request/response-tier fallback. The token-accounting fix applies to newly ingested requests and does not recalculate historical records.

Full Changelog: v2.0.1...v2.0.2

---

更新内容：

本次更新完善 API Key 筛选与请求详情，修复 Token 统计和配额状态更新，并提升配额历史的加载性能与统计准确性。

• Overview、Realtime、Analysis 和 Request Events 的 API Key 筛选现在按当前页面与时间范围，仅展示有用量记录的候选项。切换范围时保留有效的已选项，后台刷新期间继续显示已有候选列表。(#675)
• 请求详情新增 CPA 上报的实际服务层级记录、展示与导出，Speed Mode 优先使用实际层级，并支持 Ultrafast 标签。同时统一请求列表与凭证详情中的端点别名展示，悬浮提示和导出仍保留原始端点值。(#669) (#672)
• 优化 Codex 与 Claude 在大量周期和请求下的配额历史加载与交互性能。修复夏令时切换及小数秒边界附近的请求归属，并在时间显示存在歧义时补充 UTC 偏移。(#673)
• 修复 Codex 响应头已明确报告恢复后，配额仍显示达到限额的问题。主配额各窗口会同步更新限额状态，较旧的待处理更新也不会覆盖较新的刷新结果。(#676)
• 修复插件适配器及其他未识别执行器的 Token 统计。当 CPA 提供完整、有效的 Token 明细时，在计价与聚合前统一处理缓存读取、缓存创建和推理 Token，解决缓存读取量大于输入总量等问题。(#671)
• 在 CPA 支持的情况下，防止提供商配置保存覆盖其他并发修改，发生冲突时明确提示错误；同时避免旧的后台元数据快照覆盖已成功完成的凭证变更。(#670)
• 移除 Top Models 汇总工具栏，精简图表布局。Analysis 模型速度图例支持悬停、键盘聚焦和点击／触摸高亮，并修复查询变化后残留的选中与悬浮提示状态。(#668)
• CI 与发布构建改为跟随 Go 1.26 和 Node.js 24 的最新补丁版本，更新构建运行环境与工作流依赖，并在 Docker 构建时拉取更新的基础镜像。(#664)

使用提示：历史请求不会回填实际服务层级；缺少该信息时，继续沿用请求／响应层级的展示逻辑。Token 统计修复仅影响新接收的请求，不会重新计算历史记录。