diff --git a/grpc-mongo/client/main.go b/grpc-mongo/client/main.go
index 09226b92..407a8344 100644
--- a/grpc-mongo/client/main.go
+++ b/grpc-mongo/client/main.go
@@ -35,23 +35,35 @@ func main() {
 	defer conn.Close() //nolint:errcheck
 
 	c := pb.NewTokenServiceClient(conn)
-	ctx, cancel := context.WithTimeout(context.Background(), 20*time.Second)
-	defer cancel()
+
+	// Per-call deadlines: each RPC gets its own bounded context so the client's
+	// deliberate inter-call pacing (and the slow proxyless-capture warmup on a
+	// loaded CI runner) never eats into a later call's budget. A single shared
+	// absolute 20s deadline conflated all 11 calls plus their 1s pacing sleeps
+	// into one window, so under contention NextToken #9 sporadically tripped
+	// "DeadlineExceeded". A genuinely hung RPC still times out at callTimeout.
+	const callTimeout = 10 * time.Second
 
 	// Seed 10 tokens (order matters, first popped = first returned)
 	seed := []string{"alpha", "beta", "gamma", "delta", "epsilon", "zeta", "eta", "theta", "iota", "kappa"}
-	ack, err := c.SeedTokens(ctx, &pb.SeedRequest{Tokens: seed})
+	seedCtx, cancelSeed := context.WithTimeout(context.Background(), callTimeout)
+	ack, err := c.SeedTokens(seedCtx, &pb.SeedRequest{Tokens: seed})
+	cancelSeed()
 	if err != nil {
 		log.Fatalf("seed: %v", err)
 	}
 	fmt.Println("Seed:", ack.Message)
+	time.Sleep(1 * time.Second)
 
 	// Make 10 identical requests -> 10 different replies
 	for i := 1; i <= 10; i++ {
-		r, err := c.NextToken(ctx, &pb.NextTokenRequest{})
+		callCtx, cancelCall := context.WithTimeout(context.Background(), callTimeout)
+		r, err := c.NextToken(callCtx, &pb.NextTokenRequest{})
+		cancelCall()
 		if err != nil {
 			log.Fatalf("NextToken #%d: %v", i, err)
 		}
 		fmt.Printf("NextToken #%d: %s\n", i, r.Token)
+		time.Sleep(1 * time.Second)
 	}
 }