feat: record cache hit/miss per request in audit trail and WebUI

- Req: add CacheHit and CacheMiss fields (carrying upstream cache
  accounting from either prompt_tokens_details.cached_tokens or legacy
  prompt_cache_hit_tokens)
- recordChatUsage (non-streaming): copy cache fields from resp.TokenUsage
- pumpStream (streaming): write lastUsage cache fields back onto rec at
  stream end, so streaming requests carry cache data too
- CSV export: add first_byte_ms, cache_hit_tokens, cache_miss_tokens
  columns alongside the existing latency/prompt/completion
- WebUI request-records table: add a Cache column showing hit% per row
  (green/amber tag with tooltip hit/miss breakdown; em-dash when the
  upstream reported no cache data)

internal/gateway/api.go

@@ -171,7 +171,7 @@ func (g *Gateway) handleStatsAPI(w http.ResponseWriter, r *http.Request) {
 		for _, k := range g.core.ListKeys() {
 			names[keyID(k.Key)] = k.Name
 		}
-		_ = cw.Write([]string{"time", "key", "key_name", "type", "model", "source", "status", "ok", "prompt_tokens", "completion_tokens", "latency_ms", "error"})
+		_ = cw.Write([]string{"time", "key", "key_name", "type", "model", "source", "status", "ok", "prompt_tokens", "completion_tokens", "latency_ms", "first_byte_ms", "cache_hit_tokens", "cache_miss_tokens", "error"})
 		for _, rec := range g.stats.AuditRecords(from, to, key) {
 			_ = cw.Write([]string{
 				time.UnixMilli(rec.Time).Format(time.RFC3339),
@@ -185,6 +185,9 @@ func (g *Gateway) handleStatsAPI(w http.ResponseWriter, r *http.Request) {
 				strconv.FormatInt(rec.Prompt, 10),
 				strconv.FormatInt(rec.Compl, 10),
 				strconv.FormatInt(rec.LatMs, 10),
+				strconv.FormatInt(rec.FirstByteMs, 10),
+				strconv.FormatInt(rec.CacheHit, 10),
+				strconv.FormatInt(rec.CacheMiss, 10),
 				rec.Err,
 			})
 		}

internal/gateway/chat.go

@@ -570,6 +570,17 @@ func recordChatUsage(rec *Req, req *types.ChatRequest, resp *types.UnifiedRespon
 	if rec.Compl == 0 {
 		rec.Compl = estimateTextTokens(resp.Content, resp.ReasoningContent, resp.ToolCalls)
 	}
+	// Cache accounting: whichever source format the adapter normalized into
+	// (prompt_tokens_details.cached_tokens or legacy prompt_cache_hit_tokens),
+	// read it back so the request record carries the hit/miss split.
+	if d := resp.TokenUsage.PromptTokensDetails; d != nil && d.CachedTokens > 0 {
+		rec.CacheHit = int64(d.CachedTokens)
+	} else if resp.TokenUsage.PromptCacheHit > 0 {
+		rec.CacheHit = int64(resp.TokenUsage.PromptCacheHit)
+	}
+	if resp.TokenUsage.PromptCacheMiss > 0 {
+		rec.CacheMiss = int64(resp.TokenUsage.PromptCacheMiss)
+	}
 }
 
 // writeChatCompletion renders a unified response as an OpenAI
@@ -748,6 +759,16 @@ func (g *Gateway) pumpStream(w http.ResponseWriter, rec *Req, chunks <-chan type
 	var tut *types.TokenUsage
 	if lastUsage != nil {
 		tut = lastUsage
+		// Write the upstream's cache accounting back onto the request record
+		// so the audit trail carries the hit/miss split for streaming too.
+		if d := tut.PromptTokensDetails; d != nil && d.CachedTokens > 0 {
+			rec.CacheHit = int64(d.CachedTokens)
+		} else if tut.PromptCacheHit > 0 {
+			rec.CacheHit = int64(tut.PromptCacheHit)
+		}
+		if tut.PromptCacheMiss > 0 {
+			rec.CacheMiss = int64(tut.PromptCacheMiss)
+		}
 	} else if rec.Prompt+rec.Compl > 0 {
 		u := types.TokenUsage{
 			Prompt:     int(rec.Prompt),

internal/gateway/stats.go

@@ -30,6 +30,11 @@ type Req struct {
 	// to the first SSE chunk sent to the client); for non-streaming it
 	// equals LatMs. 0 when unmeasured (legacy records).
 	FirstByteMs int64 `json:"first_byte_ms,omitempty"`
+	// CacheHit / CacheMiss carry the upstream prompt-cache accounting
+	// (DeepSeek-style hit/miss tokens) when the upstream reports it.
+	// Both 0 = upstream gave no cache data.
+	CacheHit  int64 `json:"cache_hit_tokens,omitempty"`
+	CacheMiss int64 `json:"cache_miss_tokens,omitempty"`
 	OK    bool  `json:"ok"`
 	// Status http status code
 	Status int `json:"status"`

internal/gateway/ui/index.html

@@ -1971,18 +1971,30 @@
         const slice = records.slice().reverse().slice(0, 300);
         el.innerHTML =
           `<table><tr><th>${t("thTime")}</th><th class="num">${t("thStatus")}</th><th>${t("thKey")}</th><th>${t("thType")}</th><th>${t("thModel")}</th><th>${t("thSrc")}</th>
-    <th class="num">${t("thPrompt")}</th><th class="num">${t("thCompl")}</th><th class="num">${t("thLatMs")}</th></tr>` +
+    <th class="num">${t("thPrompt")}</th><th class="num">${t("thCompl")}</th><th class="num">${t("thCache") || "缓存"}</th><th class="num">${t("thLatMs")}</th></tr>` +
           slice
             .map(
               (r) => `<tr>
       <td class="t-tag">${fmtTime(r.time)}</td>
       <td class="num">${r.ok ? `<span class="tag tag-green">${r.status || 200}</span>` : `<span class="tag tag-red" title="${esc(r.error || "")}">${r.status || 500}</span>`}</td>
       <td>${esc(keyNames[r.key] ? keyNames[r.key] + " · " + r.key : r.key)}</td><td class="t-tag">${esc(r.type)}</td><td>${esc(r.model)}</td><td>${esc(r.source || "")}</td>
-      <td class="num">${fmtTok(r.prompt_tokens)}</td><td class="num">${fmtTok(r.completion_tokens)}</td><td class="num">${fmtMs(r.latency_ms)}</td></tr>`,
+      <td class="num">${fmtTok(r.prompt_tokens)}</td><td class="num">${fmtTok(r.completion_tokens)}</td>
+      <td class="num">${cacheCell(r)}</td>
+      <td class="num">${fmtMs(r.latency_ms)}</td></tr>`,
             )
             .join("") +
           "</table>";
       }
+
+      // cacheCell renders the per-request cache-hit column: a percentage when
+      // the upstream reported cache accounting, otherwise an em dash.
+      function cacheCell(r) {
+        const hit = r.cache_hit_tokens || 0;
+        if (!hit && !(r.cache_miss_tokens > 0)) return '<span class="muted">—</span>';
+        const prompt = r.prompt_tokens || 0;
+        const pct = prompt > 0 ? Math.round((hit * 100) / prompt) : 0;
+        return `<span class="tag ${pct >= 50 ? "tag-green" : pct > 0 ? "tag-amber" : ""}" title="命中 ${hit} / 未命中 ${r.cache_miss_tokens || 0}">${pct}%</span>`;
+      }
       function showModelConfig(srcName, model) {
         const el = $("#conncfg");
         if (!el) return;