diff --git a/packages/devkit/.config/nextest.toml b/packages/devkit/.config/nextest.toml
new file mode 100644
index 0000000..08c1e16
--- /dev/null
+++ b/packages/devkit/.config/nextest.toml
@@ -0,0 +1,11 @@
+[profile.default]
+retries = 0
+test-threads = "num-cpus"
+status-level = "all"
+final-status-level = "flaky"
+
+[profile.ci]
+retries = 2
+test-threads = 4
+status-level = "fail"
+final-status-level = "flaky"
diff --git a/packages/devkit/Cargo.toml b/packages/devkit/Cargo.toml
index 7f08e50..1654590 100644
--- a/packages/devkit/Cargo.toml
+++ b/packages/devkit/Cargo.toml
@@ -27,3 +27,15 @@ harness = false
 [[bench]]
 name = "percentile_bench"
 harness = false
+
+[[bench]]
+name = "main"
+harness = false
+
+[[bench]]
+name = "spike_bench"
+harness = false
+
+[[bench]]
+name = "congestion_bench"
+harness = false
diff --git a/packages/devkit/benches/congestion_bench.rs b/packages/devkit/benches/congestion_bench.rs
new file mode 100644
index 0000000..a93a356
--- /dev/null
+++ b/packages/devkit/benches/congestion_bench.rs
@@ -0,0 +1,34 @@
+use criterion::{criterion_group, criterion_main, Criterion};
+use stellar_devkit::simulation::congestion_predictor::{
+    congestion_score, CongestionInput, CongestionPredictor,
+};
+
+fn bench_predict(c: &mut Criterion) {
+    let mut group = c.benchmark_group("congestion_predictor");
+
+    group.bench_function("predict_1000_calls", |b| {
+        b.iter(|| {
+            for i in 0..1_000u64 {
+                let _ = CongestionPredictor::predict(i % 1000, 100 + i * 10);
+            }
+        })
+    });
+
+    group.bench_function("congestion_score_1000_calls", |b| {
+        b.iter(|| {
+            for i in 0..1_000u64 {
+                let input = CongestionInput {
+                    recent_fee_window: 100.0 + i as f64,
+                    capacity_usage: (i % 100) as f64 / 100.0,
+                    spike_count: (i % 10) as u32,
+                };
+                let _ = congestion_score(&input);
+            }
+        })
+    });
+
+    group.finish();
+}
+
+criterion_group!(benches, bench_predict);
+criterion_main!(benches);
diff --git a/packages/devkit/benches/main.rs b/packages/devkit/benches/main.rs
new file mode 100644
index 0000000..41a348f
--- /dev/null
+++ b/packages/devkit/benches/main.rs
@@ -0,0 +1,3 @@
+// Criterion benchmark suite entry point.
+// Individual benchmarks are in their own files (spike_bench, congestion_bench, etc.).
+fn main() {}
diff --git a/packages/devkit/benches/spike_bench.rs b/packages/devkit/benches/spike_bench.rs
new file mode 100644
index 0000000..c52ddd7
--- /dev/null
+++ b/packages/devkit/benches/spike_bench.rs
@@ -0,0 +1,16 @@
+use criterion::{criterion_group, criterion_main, BenchmarkId, Criterion};
+use stellar_devkit::analysis::spike_classifier::SpikeClassifier;
+
+fn bench_iqr_outliers(c: &mut Criterion) {
+    let mut group = c.benchmark_group("spike_classifier");
+    for size in [1_000u64, 10_000, 100_000, 1_000_000] {
+        let fees: Vec<u64> = (0..size).map(|i| 100 + (i % 50) * 10).collect();
+        group.bench_with_input(BenchmarkId::new("iqr_outliers", size), &fees, |b, fees| {
+            b.iter(|| SpikeClassifier::iqr_outliers(fees))
+        });
+    }
+    group.finish();
+}
+
+criterion_group!(benches, bench_iqr_outliers);
+criterion_main!(benches);