diff --git a/CustomStages/MCADFetchDelayStage.cpp b/CustomStages/MCADFetchDelayStage.cpp
index 6171e44..e36b57e 100644
--- a/CustomStages/MCADFetchDelayStage.cpp
+++ b/CustomStages/MCADFetchDelayStage.cpp
@@ -67,11 +67,13 @@ llvm::Error MCADFetchDelayStage::execute(llvm::mca::InstRef &IR) {
                 (fellThrough ? AbstractBranchPredictorUnit::NOT_TAKEN 
                             : AbstractBranchPredictorUnit::TAKEN);
             BPU->recordTakenBranch(*previousInstrAddr, actualBranchDirection);
-
+            
+            stats.numBranches.inc();
             if(actualBranchDirection != predictedBranchDirection) {
                 // Previous prediction was wrong; this instruction will have extra
                 // latency due to misprediction.
                 delayCyclesLeft += BPU->getMispredictionPenalty();
+                stats.numMispredictions.inc();
                 LLVM_DEBUG(dbgs() << "[MCAD FetchDelayStage] Previous branch at "); 
                 LLVM_DEBUG(dbgs().write_hex(instrAddr->addr));
                 LLVM_DEBUG(dbgs() << " mispredicted, delaying next instruction by " 
diff --git a/CustomStages/MCADFetchDelayStage.h b/CustomStages/MCADFetchDelayStage.h
index c9b1196..5704f7a 100644
--- a/CustomStages/MCADFetchDelayStage.h
+++ b/CustomStages/MCADFetchDelayStage.h
@@ -27,6 +27,7 @@ class MCADFetchDelayStage : public llvm::mca::Stage {
         llvm::mca::InstRef IR;
     };
 
+
     const llvm::MCInstrInfo &MCII;
     std::deque<DelayedInstr> instrQueue = {};
 
@@ -48,6 +49,29 @@ class MCADFetchDelayStage : public llvm::mca::Stage {
     std::optional<MDInstrAddr> previousInstrAddr = std::nullopt;
     std::optional<unsigned> previousInstrSize = std::nullopt;
 
+public: 
+    // Stats
+    // TODO: Move these elsewhere, as they are useful outside of just branch
+    // prediction or the FetchDelayStage; we could also make use of the event
+    // infrastructure that already exists (grep for STALL event)
+    struct OverflowableCount {
+        unsigned long long count;
+        bool overflowed;
+        void inc() {
+            if(count + 1 < count) {
+                overflowed = true;
+            }
+            count++;
+        }
+    };
+
+    struct Statistics {
+        OverflowableCount numBranches = {};
+        OverflowableCount numMispredictions = {};
+    };
+
+    Statistics stats = {};
+
 public:
     MCADFetchDelayStage(const llvm::MCInstrInfo &MCII, MetadataRegistry &MD,
                         AbstractBranchPredictorUnit *BPU,
diff --git a/MCAWorker.cpp b/MCAWorker.cpp
index 422be62..d9e78c7 100644
--- a/MCAWorker.cpp
+++ b/MCAWorker.cpp
@@ -211,6 +211,8 @@ std::unique_ptr<AbstractBranchPredictorUnit> buildBranchPredictor() {
 
 } // anonymous namespace
 
+MCADFetchDelayStage::Statistics *FetchDelayStats = nullptr; // TODO: ugly; move this elsewhere using hardware events
+
 void BrokerFacade::setBroker(std::unique_ptr<Broker> &&B) {
   Worker.TheBroker = std::move(B);
 }
@@ -295,6 +297,7 @@ std::unique_ptr<mca::Pipeline> MCAWorker::createDefaultPipeline() {
   // Create the pipeline stages.
   auto Fetch = std::make_unique<EntryStage>(SrcMgr);
   auto FetchDelay = std::make_unique<MCADFetchDelayStage>(MCII, MDRegistry, BPU.get(), L1I);
+  FetchDelayStats = &FetchDelay->stats; // TODO: ugly; move this elsewhere using hardware events
   auto Dispatch = std::make_unique<DispatchStage>(STI, MRI, MCAPO.DispatchWidth,
                                                   *RCU, *PRF);
   auto Execute =
@@ -610,6 +613,10 @@ void MCAWorker::printMCA(StringRef RegionDescription) {
        << RegionDescription << " ===\n";
 
   MCAPipelinePrinter->printReport(OS);
+  if(FetchDelayStats) {
+    OS << "Branch Instructions: " << FetchDelayStats->numBranches.count << (FetchDelayStats->numBranches.overflowed ? " (overflowed)" : "") << "\n";
+    OS << "Branch Mispredictions: " << FetchDelayStats->numMispredictions.count << (FetchDelayStats->numMispredictions.overflowed ? " (overflowed)" : "") << "\n";
+  }
 }
 
 MCAWorker::~MCAWorker() {