diff --git a/README.md b/README.md
index 6c79510..6f33c85 100644
--- a/README.md
+++ b/README.md
@@ -258,9 +258,12 @@ go run ./cmd/proxy-loadgen `
 ```
 
 命令输出 JSON 报告，包含成功/失败分类、`TunnelsEstablished`、
-`ExtractResponsesValidated`、`ExtractReturned`、`ExtractValidationFailures`、固定内存的
+`ExtractResponsesValidated`、`ExtractReturned`、`ExtractValidationFailures`、`RateStartsGenerated`、
+`RateStartsDropped`、固定内存的
 连接握手延迟分位上界、吞吐和 Go 运行时内存/GC 快照。它不会输出提取响应中的地址或
-凭据，也不构成 100,000 QPS 证明。
+凭据，也不构成 100,000 QPS 证明。限速场景以 `Requests` 表示实际发起数；当
+`RateStartsDropped` 非零时，目标速率受压测端并发容量或时间窗限制，报告吞吐不得
+标注为已达到配置的 `-rate`。
 
 ## 关键配置与入口
 
diff --git a/cmd/proxy-loadgen/main_test.go b/cmd/proxy-loadgen/main_test.go
index 4dfc688..c6e8b41 100644
--- a/cmd/proxy-loadgen/main_test.go
+++ b/cmd/proxy-loadgen/main_test.go
@@ -18,7 +18,10 @@ func TestExecutePassesBoundedWorkloadAndWritesJSON(t *testing.T) {
 		"-requests", "3", "-concurrency", "2", "-timeout", "2s", "-header", "X-Run: fixed", "-body", `{"count":1}`,
 	}, func(_ context.Context, options loadgen.Options) (loadgen.Report, error) {
 		received = options
-		return loadgen.Report{Requests: 3, Completed: 3, Succeeded: 3, Duration: time.Second}, nil
+		return loadgen.Report{
+			Requests: 3, Completed: 3, Succeeded: 3, Duration: time.Second,
+			RateStartsGenerated: 3, RateStartsDropped: 1,
+		}, nil
 	}, &stdout, &stderr)
 	if code != 0 || received.TargetURL != "https://target.example/path" || received.ProxyURL != "http://gateway.example:8080" ||
 		received.Method != "post" || received.Requests != 3 || received.Duration != 0 || received.Concurrency != 2 ||
@@ -26,7 +29,8 @@ func TestExecutePassesBoundedWorkloadAndWritesJSON(t *testing.T) {
 		t.Fatalf("execute() = %d; options=%+v stderr=%q", code, received, stderr.String())
 	}
 	var report loadgen.Report
-	if err := json.Unmarshal(stdout.Bytes(), &report); err != nil || report.Succeeded != 3 {
+	if err := json.Unmarshal(stdout.Bytes(), &report); err != nil || report.Succeeded != 3 ||
+		report.RateStartsGenerated != 3 || report.RateStartsDropped != 1 {
 		t.Fatalf("JSON report = (%+v, %v)", report, err)
 	}
 }
diff --git a/docs/testing/test-strategy.md b/docs/testing/test-strategy.md
index f034030..fe88805 100644
--- a/docs/testing/test-strategy.md
+++ b/docs/testing/test-strategy.md
@@ -182,6 +182,7 @@ Admin 应用层测试覆盖 typed-nil 依赖、Actor/SourceIP 映射、Routing C
 配置：Snapshot 规模、Proxy 容量、路由、重试、日志级别
 场景：协议、连接复用、响应体、持续时间、升压曲线、故障注入
 结果：QPS、建连速率、active、p50/p95/p99、错误、CPU、RSS、GC、FD、网络
+限速：配置 rate、实际 Requests、RateStartsGenerated、RateStartsDropped
 不变量：capacity、ownership、extraction、idempotency、admin audit、reserve 检查结果
 结论：通过/失败，以及适用边界
 ```
diff --git a/internal/loadgen/http.go b/internal/loadgen/http.go
index 76be38a..ae57b88 100644
--- a/internal/loadgen/http.go
+++ b/internal/loadgen/http.go
@@ -75,6 +75,8 @@ type Report struct {
 	ExtractReturned           uint64
 	ExtractValidationFailures uint64
 	Throughput                float64
+	RateStartsGenerated       uint64
+	RateStartsDropped         uint64
 	Latency                   LatencyReport
 	Runtime                   RuntimeReport
 }
@@ -110,6 +112,8 @@ type counters struct {
 	extractResponsesValidated atomic.Uint64
 	extractReturned           atomic.Uint64
 	extractValidationFailures atomic.Uint64
+	rateStartsGenerated       atomic.Uint64
+	rateStartsDropped         atomic.Uint64
 	latency                   latencyHistogram
 }
 
@@ -154,7 +158,7 @@ func Run(ctx context.Context, options Options) (Report, error) {
 			return report(started, normalized.Scenario, stats), err
 		}
 	} else {
-		runForDuration(ctx, normalized, execute)
+		runForDuration(ctx, normalized, execute, stats)
 	}
 	return report(started, normalized.Scenario, stats), nil
 }
@@ -285,7 +289,7 @@ func runFixed(ctx context.Context, options Options, execute func(context.Context
 	workers.Wait()
 }
 
-func runForDuration(ctx context.Context, options Options, execute func(context.Context)) {
+func runForDuration(ctx context.Context, options Options, execute func(context.Context), stats *counters) {
 	workloadContext, cancel := context.WithTimeout(ctx, options.Duration)
 	defer cancel()
 	if options.Rate == 0 {
@@ -334,14 +338,20 @@ func runForDuration(ctx context.Context, options Options, execute func(context.C
 			last = now
 			tokens := int(credit)
 			credit -= float64(tokens)
+			if tokens <= 0 {
+				continue
+			}
+			stats.rateStartsGenerated.Add(uint64(tokens))
 			if tokens > options.Concurrency {
+				stats.rateStartsDropped.Add(uint64(tokens - options.Concurrency))
 				tokens = options.Concurrency
 				credit = 0
 			}
-			for range tokens {
+			for index := 0; index < tokens; index++ {
 				select {
 				case jobs <- struct{}{}:
 				case <-workloadContext.Done():
+					stats.rateStartsDropped.Add(uint64(tokens - index))
 					workers.Wait()
 					return
 				}
@@ -655,7 +665,9 @@ func report(started time.Time, scenario Scenario, stats *counters) Report {
 		Status5xx: stats.status5xx.Load(), TimeoutErrors: stats.timeoutErrors.Load(), RequestErrors: stats.requestErrors.Load(),
 		TunnelsEstablished:        stats.tunnelsEstablished.Load(),
 		ExtractResponsesValidated: stats.extractResponsesValidated.Load(), ExtractReturned: stats.extractReturned.Load(),
-		ExtractValidationFailures: stats.extractValidationFailures.Load(), Latency: stats.latency.Report(completed),
+		ExtractValidationFailures: stats.extractValidationFailures.Load(),
+		RateStartsGenerated:       stats.rateStartsGenerated.Load(), RateStartsDropped: stats.rateStartsDropped.Load(),
+		Latency: stats.latency.Report(completed),
 		Runtime: RuntimeReport{NumCPU: runtime.NumCPU(), Goroutines: runtime.NumGoroutine(), AllocBytes: memory.Alloc, SysBytes: memory.Sys, GCCount: memory.NumGC},
 	}
 	if duration > 0 {
diff --git a/internal/loadgen/http_test.go b/internal/loadgen/http_test.go
index eb5a0ca..d6d88a4 100644
--- a/internal/loadgen/http_test.go
+++ b/internal/loadgen/http_test.go
@@ -71,6 +71,32 @@ func TestRunTimeBoxedRateIsBounded(t *testing.T) {
 	}
 }
 
+func TestRunReportsRateStartsDroppedByConcurrency(t *testing.T) {
+	started := make(chan struct{}, 1)
+	server := httptest.NewServer(http.HandlerFunc(func(response http.ResponseWriter, request *http.Request) {
+		select {
+		case started <- struct{}{}:
+		default:
+		}
+		<-request.Context().Done()
+		response.WriteHeader(http.StatusNoContent)
+	}))
+	defer server.Close()
+
+	report, err := Run(context.Background(), Options{
+		TargetURL: server.URL, Duration: 80 * time.Millisecond, Rate: 100_000, Concurrency: 1, RequestTimeout: time.Second,
+	})
+	select {
+	case <-started:
+	default:
+		t.Fatal("rate-limited workload did not start a request")
+	}
+	if err != nil || report.RateStartsGenerated <= report.Requests || report.RateStartsDropped == 0 ||
+		report.RateStartsDropped > report.RateStartsGenerated {
+		t.Fatalf("Run() = (%+v, %v)", report, err)
+	}
+}
+
 func TestRunExecutesBoundedCONNECTTunnelWorkload(t *testing.T) {
 	listener, err := net.Listen("tcp", "127.0.0.1:0")
 	if err != nil {