fix(usage): synchronize Keeper metrics and improve detail refresh

- Align TPS and token normalization with Keeper
- Add Redis usage queue batch pulling with HTTP fallback
- Prevent cache and invalid token double counting
- Refresh usage details immediately after new records arrive

src/i18n/ja.ts

@@ -414,7 +414,7 @@ export const jaOverrides = {
   'usage.stat.averageLatency': '平均レイテンシ',
   'usage.stat.tps': 'TPS',
   'usage.stat.performanceMeta': '有効サンプル {samples} 件 · RPM {rpm}',
-  'usage.stat.performanceMetaTitle': 'TPS {tps}、有効サンプル {samples} 件、RPM {rpm}、平均レイテンシ {latency} ms',
+  'usage.stat.performanceMetaTitle': 'TPS {tps}（総レイテンシで加重）、有効サンプル {samples} 件、RPM {rpm}、平均レイテンシ {latency} ms',
   'usage.stat.cacheHitRate': 'キャッシュヒット率',
   'usage.stat.cacheHitMeta': 'キャッシュ読取 {hit} · 入力 {input}',
   'usage.stat.cacheHitMetaTitle': 'キャッシュヒット率 {rate}%、読取 {hit} Token、入力 {input} Token',
@@ -498,7 +498,7 @@ export const jaOverrides = {
   'usage.column.cache': 'キャッシュ',
   'usage.column.cacheRate': 'キャッシュ率',
   'usage.column.total': '合計',
-  'usage.column.speed': '生成速度',
+  'usage.column.speed': '出力速度',
   'usage.key.noRemark': 'メモなし',
   'usage.previous': '前へ',
   'usage.next': '次へ',

src/i18n/locales/en.ts

@@ -401,7 +401,7 @@ export const en: Record<MessageKey, string> = {
   'usage.stat.averageLatency': 'Average Latency',
   'usage.stat.tps': 'TPS',
   'usage.stat.performanceMeta': '{samples} samples · RPM {rpm}',
-  'usage.stat.performanceMetaTitle': 'TPS {tps}, {samples} valid samples, RPM {rpm}, average latency {latency} ms',
+  'usage.stat.performanceMetaTitle': 'TPS {tps} (weighted by total latency), {samples} valid samples, RPM {rpm}, average latency {latency} ms',
   'usage.stat.cacheHitRate': 'Cache Hit Rate',
   'usage.stat.cacheHitMeta': 'Cache read {hit} · Input {input}',
   'usage.stat.cacheHitMetaTitle': 'Cache hit rate {rate}%, cache read {hit} tokens, input {input} tokens',
@@ -486,7 +486,7 @@ export const en: Record<MessageKey, string> = {
   'usage.column.cache': 'Cache',
   'usage.column.cacheRate': 'Cache Rate',
   'usage.column.total': 'Total',
-  'usage.column.speed': 'Generation Speed',
+  'usage.column.speed': 'Output Speed',
   'usage.key.noRemark': 'No remark',
   'usage.previous': 'Previous',
   'usage.next': 'Next',

src/i18n/locales/zh-CN.ts

@@ -400,7 +400,7 @@ export const zhCN = {
   'usage.stat.averageLatency': '平均延迟',
   'usage.stat.tps': 'TPS',
   'usage.stat.performanceMeta': '{samples} 条有效样本 · RPM {rpm}',
-  'usage.stat.performanceMetaTitle': 'TPS {tps}，{samples} 条有效样本，RPM {rpm}，平均延迟 {latency} ms',
+  'usage.stat.performanceMetaTitle': 'TPS {tps}（按总延迟加权），{samples} 条有效样本，RPM {rpm}，平均延迟 {latency} ms',
   'usage.stat.cacheHitRate': '缓存命中率',
   'usage.stat.cacheHitMeta': '缓存读取 {hit} · 输入 {input}',
   'usage.stat.cacheHitMetaTitle': '缓存命中率 {rate}%，缓存读取 {hit} Token，输入 {input} Token',
@@ -485,7 +485,7 @@ export const zhCN = {
   'usage.column.cache': '缓存',
   'usage.column.cacheRate': '缓存率',
   'usage.column.total': '总计',
-  'usage.column.speed': '生成速度',
+  'usage.column.speed': '输出速度',
   'usage.key.noRemark': '未备注',
   'usage.previous': '上一页',
   'usage.next': '下一页',

src/pages/UsageRecordsPage.tsx

@@ -400,9 +400,9 @@ export function UsageRecordsPage() {
   );
 
   const loadData = useCallback(
-    (quiet = false) => {
+    (quiet = false, immediate = !quiet) => {
       if (!quiet) setLoading(true);
-      return schedulerRef.current!.schedule(() => executeLoadData(quiet), !quiet);
+      return schedulerRef.current!.schedule(() => executeLoadData(quiet), immediate);
     },
     [executeLoadData],
   );
@@ -419,7 +419,7 @@ export function UsageRecordsPage() {
     let disposed = false;
     let unlisten: (() => void) | null = null;
     const refresh = () => {
-      if (!disposed && !document.hidden) void loadData(true);
+      if (!disposed && !document.hidden) void loadData(true, true);
     };
     listen('usage-records-updated', refresh)
       .then((stop) => {
@@ -1440,7 +1440,6 @@ function UsageEventCell({
       const value = formatGenerationSpeed({
         outputTokens: record.tokens.output_tokens,
         latencyMs: record.latency_ms,
-        ttftMs: record.ttft_ms,
       });
       return <td className="usage-td-speed align-center" title={value === '—' ? undefined : value}>{value}</td>;
     }

src/services/usageMetrics.ts

@@ -1,7 +1,6 @@
 type GenerationSpeedInput = {
   outputTokens: number;
   latencyMs: number;
-  ttftMs: number | null;
 };
 
 type CacheReadRateInput = {
@@ -12,21 +11,17 @@ type CacheReadRateInput = {
 export const calculateGenerationSpeed = ({
   outputTokens,
   latencyMs,
-  ttftMs,
 }: GenerationSpeedInput): number | null => {
   if (
     !Number.isFinite(outputTokens) ||
     !Number.isFinite(latencyMs) ||
-    ttftMs === null ||
-    !Number.isFinite(ttftMs) ||
     outputTokens <= 0 ||
-    ttftMs <= 0 ||
-    latencyMs <= ttftMs
+    latencyMs <= 0
   ) {
     return null;
   }
 
-  const speed = outputTokens / ((latencyMs - ttftMs) / 1_000);
+  const speed = outputTokens / (latencyMs / 1_000);
   return Number.isFinite(speed) && speed > 0 ? speed : null;
 };