diff --git a/v1/src/sensing/feature_extractor.py b/v1/src/sensing/feature_extractor.py
index 445899c..f5be71c 100644
--- a/v1/src/sensing/feature_extractor.py
+++ b/v1/src/sensing/feature_extractor.py
@@ -103,6 +103,8 @@ class RssiFeatureExtractor:
 
         # Trim to window
         samples = self._trim_to_window(samples)
+        if len(samples) < 4:
+            return RssiFeatures(n_samples=len(samples))
         rssi = np.array([s.rssi_dbm for s in samples], dtype=np.float64)
         timestamps = np.array([s.timestamp for s in samples], dtype=np.float64)
 
@@ -158,6 +160,17 @@ class RssiFeatureExtractor:
 
         return features
 
+    # -- window trimming -----------------------------------------------------
+
+    def _trim_to_window(self, samples: List[WifiSample]) -> List[WifiSample]:
+        """Keep only samples within the most recent ``window_seconds``."""
+        if not samples:
+            return samples
+        latest_ts = samples[-1].timestamp
+        cutoff = latest_ts - self._window_seconds
+        trimmed = [s for s in samples if s.timestamp >= cutoff]
+        return trimmed
+
     # -- time-domain ---------------------------------------------------------
 
     @staticmethod
@@ -165,10 +178,16 @@ class RssiFeatureExtractor:
         features.mean = float(np.mean(rssi))
         features.variance = float(np.var(rssi, ddof=1)) if len(rssi) > 1 else 0.0
         features.std = float(np.std(rssi, ddof=1)) if len(rssi) > 1 else 0.0
-        features.skewness = float(scipy_stats.skew(rssi, bias=False)) if len(rssi) > 2 else 0.0
-        features.kurtosis = float(scipy_stats.kurtosis(rssi, bias=False)) if len(rssi) > 3 else 0.0
         features.range = float(np.ptp(rssi))
 
+        # Guard against constant signals where higher moments are undefined
+        if features.std < 1e-12:
+            features.skewness = 0.0
+            features.kurtosis = 0.0
+        else:
+            features.skewness = float(scipy_stats.skew(rssi, bias=False)) if len(rssi) > 2 else 0.0
+            features.kurtosis = float(scipy_stats.kurtosis(rssi, bias=False)) if len(rssi) > 3 else 0.0
+
         q75, q25 = np.percentile(rssi, [75, 25])
         features.iqr = float(q75 - q25)
 
diff --git a/v1/tests/unit/test_sensing.py b/v1/tests/unit/test_sensing.py
new file mode 100644
index 0000000..bce1eeb
--- /dev/null
+++ b/v1/tests/unit/test_sensing.py
@@ -0,0 +1,704 @@
+"""
+Unit tests for the commodity sensing module (ADR-013).
+
+Tests cover:
+    - Feature extraction from known sinusoidal RSSI input
+    - Classifier producing correct presence/motion from known features
+    - SimulatedCollector determinism (same seed = same output)
+    - CUSUM change-point detection catching step changes
+    - Band power extraction isolating correct frequencies
+    - Backend capabilities and pipeline integration
+"""
+
+from __future__ import annotations
+
+import math
+
+import numpy as np
+import pytest
+from numpy.typing import NDArray
+
+from v1.src.sensing.rssi_collector import (
+    RingBuffer,
+    SimulatedCollector,
+    WifiSample,
+)
+from v1.src.sensing.feature_extractor import (
+    RssiFeatureExtractor,
+    RssiFeatures,
+    cusum_detect,
+    _band_power,
+)
+from v1.src.sensing.classifier import (
+    MotionLevel,
+    PresenceClassifier,
+    SensingResult,
+)
+from v1.src.sensing.backend import (
+    Capability,
+    CommodityBackend,
+    SensingBackend,
+)
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def make_sinusoidal_rssi(
+    freq_hz: float,
+    amplitude: float,
+    baseline: float,
+    duration_s: float,
+    sample_rate: float,
+) -> NDArray[np.float64]:
+    """Generate a clean sinusoidal RSSI signal (no noise)."""
+    n = int(duration_s * sample_rate)
+    t = np.arange(n) / sample_rate
+    return baseline + amplitude * np.sin(2 * np.pi * freq_hz * t)
+
+
+def make_step_signal(
+    baseline: float,
+    step_value: float,
+    step_at_sample: int,
+    n_samples: int,
+) -> NDArray[np.float64]:
+    """Generate a signal with a step change at a specific sample."""
+    signal = np.full(n_samples, baseline, dtype=np.float64)
+    signal[step_at_sample:] = step_value
+    return signal
+
+
+# ===========================================================================
+# RingBuffer tests
+# ===========================================================================
+
+class TestRingBuffer:
+    def test_append_and_get_all(self):
+        buf = RingBuffer(max_size=5)
+        for i in range(3):
+            buf.append(WifiSample(
+                timestamp=float(i), rssi_dbm=-50.0 + i, noise_dbm=-95.0,
+                link_quality=0.8, tx_bytes=0, rx_bytes=0, retry_count=0,
+                interface="test0",
+            ))
+        assert len(buf) == 3
+        samples = buf.get_all()
+        assert len(samples) == 3
+        assert samples[0].rssi_dbm == -50.0
+        assert samples[2].rssi_dbm == -48.0
+
+    def test_ring_buffer_overflow(self):
+        buf = RingBuffer(max_size=3)
+        for i in range(5):
+            buf.append(WifiSample(
+                timestamp=float(i), rssi_dbm=float(i), noise_dbm=-95.0,
+                link_quality=0.8, tx_bytes=0, rx_bytes=0, retry_count=0,
+                interface="test0",
+            ))
+        assert len(buf) == 3
+        samples = buf.get_all()
+        # Oldest two should have been evicted; remaining: 2, 3, 4
+        assert samples[0].rssi_dbm == 2.0
+        assert samples[2].rssi_dbm == 4.0
+
+    def test_get_last_n(self):
+        buf = RingBuffer(max_size=10)
+        for i in range(7):
+            buf.append(WifiSample(
+                timestamp=float(i), rssi_dbm=float(i), noise_dbm=-95.0,
+                link_quality=0.8, tx_bytes=0, rx_bytes=0, retry_count=0,
+                interface="test0",
+            ))
+        last_3 = buf.get_last_n(3)
+        assert len(last_3) == 3
+        assert last_3[0].rssi_dbm == 4.0
+        assert last_3[2].rssi_dbm == 6.0
+
+    def test_clear(self):
+        buf = RingBuffer(max_size=10)
+        buf.append(WifiSample(
+            timestamp=0.0, rssi_dbm=-50.0, noise_dbm=-95.0,
+            link_quality=0.8, tx_bytes=0, rx_bytes=0, retry_count=0,
+            interface="test0",
+        ))
+        buf.clear()
+        assert len(buf) == 0
+
+
+# ===========================================================================
+# SimulatedCollector tests
+# ===========================================================================
+
+class TestSimulatedCollector:
+    def test_deterministic_output_same_seed(self):
+        """Same seed must produce identical samples."""
+        c1 = SimulatedCollector(seed=123, sample_rate_hz=10.0)
+        c2 = SimulatedCollector(seed=123, sample_rate_hz=10.0)
+
+        s1 = c1.generate_samples(5.0)
+        s2 = c2.generate_samples(5.0)
+
+        assert len(s1) == len(s2) == 50
+        for a, b in zip(s1, s2):
+            assert a.rssi_dbm == b.rssi_dbm, (
+                f"RSSI mismatch at same seed: {a.rssi_dbm} != {b.rssi_dbm}"
+            )
+            assert a.noise_dbm == b.noise_dbm
+            assert a.link_quality == b.link_quality
+
+    def test_different_seeds_differ(self):
+        """Different seeds must produce different samples."""
+        c1 = SimulatedCollector(seed=1, sample_rate_hz=10.0)
+        c2 = SimulatedCollector(seed=999, sample_rate_hz=10.0)
+
+        s1 = c1.generate_samples(2.0)
+        s2 = c2.generate_samples(2.0)
+
+        rssi1 = [s.rssi_dbm for s in s1]
+        rssi2 = [s.rssi_dbm for s in s2]
+        # Not all values should match
+        assert rssi1 != rssi2
+
+    def test_sinusoidal_component(self):
+        """With zero noise, should see a clean sinusoid."""
+        c = SimulatedCollector(
+            seed=0,
+            sample_rate_hz=100.0,
+            baseline_dbm=-50.0,
+            sine_freq_hz=1.0,
+            sine_amplitude_dbm=5.0,
+            noise_std_dbm=0.0,  # no noise
+        )
+        samples = c.generate_samples(2.0)
+        rssi = np.array([s.rssi_dbm for s in samples])
+
+        # Mean should be very close to baseline
+        assert abs(np.mean(rssi) - (-50.0)) < 0.5
+
+        # Amplitude should be close to 5 dBm (peak-to-peak ~10)
+        assert np.ptp(rssi) > 9.0
+        assert np.ptp(rssi) < 11.0
+
+    def test_step_change_injection(self):
+        """Step change should shift the signal at the specified time."""
+        c = SimulatedCollector(
+            seed=42,
+            sample_rate_hz=10.0,
+            baseline_dbm=-50.0,
+            sine_amplitude_dbm=0.0,
+            noise_std_dbm=0.0,
+            step_change_at=2.0,
+            step_change_dbm=-10.0,
+        )
+        samples = c.generate_samples(4.0)
+        rssi = np.array([s.rssi_dbm for s in samples])
+
+        # Before step (first 20 samples at 10 Hz = 2 seconds)
+        mean_before = np.mean(rssi[:20])
+        # After step (samples 20-39)
+        mean_after = np.mean(rssi[20:])
+
+        assert abs(mean_before - (-50.0)) < 0.1
+        assert abs(mean_after - (-60.0)) < 0.1
+
+    def test_sample_count(self):
+        """generate_samples should produce exactly rate * duration samples."""
+        c = SimulatedCollector(seed=0, sample_rate_hz=20.0)
+        samples = c.generate_samples(3.0)
+        assert len(samples) == 60
+
+
+# ===========================================================================
+# Feature extraction tests
+# ===========================================================================
+
+class TestFeatureExtractor:
+    def test_time_domain_from_known_sine(self):
+        """
+        A pure sinusoid at -50 dBm baseline with 2 dBm amplitude should
+        produce known statistical properties.
+        """
+        sample_rate = 100.0
+        rssi = make_sinusoidal_rssi(
+            freq_hz=1.0, amplitude=2.0, baseline=-50.0,
+            duration_s=10.0, sample_rate=sample_rate,
+        )
+
+        ext = RssiFeatureExtractor(window_seconds=30.0)
+        features = ext.extract_from_array(rssi, sample_rate)
+
+        # Mean should be close to -50
+        assert abs(features.mean - (-50.0)) < 0.1
+
+        # Variance of A*sin(x) is A^2/2
+        expected_var = 2.0**2 / 2.0  # = 2.0
+        assert abs(features.variance - expected_var) < 0.2
+
+        # Skewness of a pure sinusoid is ~0
+        assert abs(features.skewness) < 0.2
+
+        # Range should be close to 2*amplitude = 4.0
+        assert abs(features.range - 4.0) < 0.2
+
+    def test_frequency_domain_dominant_frequency(self):
+        """
+        A 0.3 Hz sinusoid should produce a dominant frequency near 0.3 Hz.
+        """
+        sample_rate = 10.0
+        rssi = make_sinusoidal_rssi(
+            freq_hz=0.3, amplitude=3.0, baseline=-50.0,
+            duration_s=30.0, sample_rate=sample_rate,
+        )
+
+        ext = RssiFeatureExtractor(window_seconds=60.0)
+        features = ext.extract_from_array(rssi, sample_rate)
+
+        # Dominant frequency should be close to 0.3 Hz
+        assert abs(features.dominant_freq_hz - 0.3) < 0.1, (
+            f"Dominant freq {features.dominant_freq_hz} != ~0.3 Hz"
+        )
+
+    def test_breathing_band_power(self):
+        """
+        A 0.3 Hz signal should produce significant power in the breathing
+        band (0.1-0.5 Hz) and negligible power in the motion band (0.5-3 Hz).
+        """
+        sample_rate = 10.0
+        rssi = make_sinusoidal_rssi(
+            freq_hz=0.3, amplitude=3.0, baseline=-50.0,
+            duration_s=30.0, sample_rate=sample_rate,
+        )
+
+        ext = RssiFeatureExtractor(window_seconds=60.0)
+        features = ext.extract_from_array(rssi, sample_rate)
+
+        assert features.breathing_band_power > 0.1, (
+            f"Breathing band power too low: {features.breathing_band_power}"
+        )
+        # Motion band should have much less power than breathing band
+        assert features.motion_band_power < features.breathing_band_power, (
+            f"Motion band ({features.motion_band_power}) should be less than "
+            f"breathing band ({features.breathing_band_power})"
+        )
+
+    def test_motion_band_power(self):
+        """
+        A 1.5 Hz signal should produce significant power in the motion
+        band (0.5-3.0 Hz) and negligible power in the breathing band.
+        """
+        sample_rate = 10.0
+        rssi = make_sinusoidal_rssi(
+            freq_hz=1.5, amplitude=3.0, baseline=-50.0,
+            duration_s=30.0, sample_rate=sample_rate,
+        )
+
+        ext = RssiFeatureExtractor(window_seconds=60.0)
+        features = ext.extract_from_array(rssi, sample_rate)
+
+        assert features.motion_band_power > 0.1, (
+            f"Motion band power too low: {features.motion_band_power}"
+        )
+        assert features.motion_band_power > features.breathing_band_power, (
+            f"Motion band ({features.motion_band_power}) should dominate over "
+            f"breathing band ({features.breathing_band_power})"
+        )
+
+    def test_band_isolation_multi_frequency(self):
+        """
+        A signal with components at 0.2 Hz AND 2.0 Hz should produce power
+        in both bands, each dominated by the correct component.
+        """
+        sample_rate = 10.0
+        n = int(30.0 * sample_rate)
+        t = np.arange(n) / sample_rate
+        # 0.2 Hz component (breathing) + 2.0 Hz component (motion)
+        rssi = -50.0 + 3.0 * np.sin(2 * np.pi * 0.2 * t) + 2.0 * np.sin(2 * np.pi * 2.0 * t)
+
+        ext = RssiFeatureExtractor(window_seconds=60.0)
+        features = ext.extract_from_array(rssi, sample_rate)
+
+        # Both bands should have significant power
+        assert features.breathing_band_power > 0.05
+        assert features.motion_band_power > 0.05
+
+    def test_constant_signal_features(self):
+        """A constant signal should have zero variance and no spectral content."""
+        rssi = np.full(200, -50.0)
+        ext = RssiFeatureExtractor()
+        features = ext.extract_from_array(rssi, 10.0)
+
+        assert features.variance == 0.0
+        assert features.std == 0.0
+        assert features.range == 0.0
+        assert features.iqr == 0.0
+        assert features.total_spectral_power < 1e-10
+
+    def test_too_few_samples(self):
+        """Fewer than 4 samples should return empty features."""
+        rssi = np.array([-50.0, -51.0])
+        ext = RssiFeatureExtractor()
+        features = ext.extract_from_array(rssi, 10.0)
+        assert features.n_samples == 2
+        assert features.variance == 0.0
+
+    def test_extract_from_wifi_samples(self):
+        """Test extraction from WifiSample objects (the normal path)."""
+        collector = SimulatedCollector(
+            seed=42, sample_rate_hz=10.0,
+            baseline_dbm=-50.0, sine_freq_hz=0.3,
+            sine_amplitude_dbm=2.0, noise_std_dbm=0.1,
+        )
+        samples = collector.generate_samples(10.0)
+
+        ext = RssiFeatureExtractor(window_seconds=30.0)
+        features = ext.extract(samples)
+
+        assert features.n_samples == 100
+        assert abs(features.mean - (-50.0)) < 1.0
+        assert features.variance > 0.0
+
+
+# ===========================================================================
+# CUSUM change-point detection tests
+# ===========================================================================
+
+class TestCusum:
+    def test_step_change_detected(self):
+        """CUSUM should detect a step change in the signal."""
+        signal = make_step_signal(
+            baseline=0.0, step_value=5.0,
+            step_at_sample=100, n_samples=200,
+        )
+        target = float(np.mean(signal))
+        std = float(np.std(signal, ddof=1))
+        threshold = 3.0 * std
+        drift = 0.5 * std
+
+        change_points = cusum_detect(signal, target, threshold, drift)
+
+        assert len(change_points) > 0, "No change points detected for step change"
+        # At least one change point should be near the step (sample 100)
+        nearest = min(change_points, key=lambda x: abs(x - 100))
+        assert abs(nearest - 100) < 20, (
+            f"Nearest change point at {nearest}, expected near 100"
+        )
+
+    def test_no_change_point_in_constant(self):
+        """A constant signal should produce no change points."""
+        signal = np.full(200, 0.0)
+        change_points = cusum_detect(signal, 0.0, 1.0, 0.1)
+        assert len(change_points) == 0
+
+    def test_multiple_step_changes(self):
+        """CUSUM should detect multiple step changes."""
+        n = 300
+        signal = np.zeros(n, dtype=np.float64)
+        signal[100:200] = 5.0
+        signal[200:] = 0.0
+
+        target = float(np.mean(signal))
+        std = float(np.std(signal, ddof=1))
+        threshold = 2.0 * std
+        drift = 0.3 * std
+
+        change_points = cusum_detect(signal, target, threshold, drift)
+        # Should detect at least the step up and the step down
+        assert len(change_points) >= 2, (
+            f"Expected >= 2 change points, got {len(change_points)}"
+        )
+
+    def test_cusum_with_feature_extractor(self):
+        """Feature extractor should detect step change via CUSUM."""
+        signal = make_step_signal(
+            baseline=-50.0, step_value=-60.0,
+            step_at_sample=150, n_samples=300,
+        )
+
+        ext = RssiFeatureExtractor(cusum_threshold=2.0, cusum_drift=0.3)
+        features = ext.extract_from_array(signal, 10.0)
+
+        assert features.n_change_points > 0, (
+            f"Expected change points but got {features.n_change_points}"
+        )
+
+
+# ===========================================================================
+# Classifier tests
+# ===========================================================================
+
+class TestPresenceClassifier:
+    def test_absent_when_low_variance(self):
+        """Low variance should classify as ABSENT."""
+        features = RssiFeatures(
+            variance=0.1,
+            motion_band_power=0.0,
+            breathing_band_power=0.0,
+            n_samples=100,
+        )
+        clf = PresenceClassifier(presence_variance_threshold=0.5)
+        result = clf.classify(features)
+
+        assert result.motion_level == MotionLevel.ABSENT
+        assert result.presence_detected is False
+
+    def test_present_still_when_high_variance_low_motion(self):
+        """High variance but low motion energy should classify as PRESENT_STILL."""
+        features = RssiFeatures(
+            variance=2.0,
+            motion_band_power=0.05,
+            breathing_band_power=0.3,
+            n_samples=100,
+        )
+        clf = PresenceClassifier(
+            presence_variance_threshold=0.5,
+            motion_energy_threshold=0.1,
+        )
+        result = clf.classify(features)
+
+        assert result.motion_level == MotionLevel.PRESENT_STILL
+        assert result.presence_detected is True
+
+    def test_active_when_high_variance_high_motion(self):
+        """High variance and high motion energy should classify as ACTIVE."""
+        features = RssiFeatures(
+            variance=3.0,
+            motion_band_power=0.5,
+            breathing_band_power=0.1,
+            n_samples=100,
+        )
+        clf = PresenceClassifier(
+            presence_variance_threshold=0.5,
+            motion_energy_threshold=0.1,
+        )
+        result = clf.classify(features)
+
+        assert result.motion_level == MotionLevel.ACTIVE
+        assert result.presence_detected is True
+
+    def test_confidence_for_absent_decreases_with_rising_variance(self):
+        """
+        When classified as ABSENT, confidence should decrease as variance
+        approaches the presence threshold (less certain about absence).
+        """
+        clf = PresenceClassifier(presence_variance_threshold=10.0)
+
+        clearly_absent = clf.classify(RssiFeatures(
+            variance=0.5, motion_band_power=0.0, n_samples=100
+        ))
+        borderline_absent = clf.classify(RssiFeatures(
+            variance=9.0, motion_band_power=0.0, n_samples=100
+        ))
+
+        assert clearly_absent.motion_level == MotionLevel.ABSENT
+        assert borderline_absent.motion_level == MotionLevel.ABSENT
+        assert clearly_absent.confidence > borderline_absent.confidence, (
+            f"Clearly absent ({clearly_absent.confidence}) should have higher "
+            f"confidence than borderline absent ({borderline_absent.confidence})"
+        )
+
+    def test_confidence_bounded_0_to_1(self):
+        """Confidence should always be in [0, 1]."""
+        clf = PresenceClassifier()
+
+        for var in [0.0, 0.1, 1.0, 10.0, 100.0]:
+            result = clf.classify(
+                RssiFeatures(variance=var, motion_band_power=var, n_samples=100)
+            )
+            assert 0.0 <= result.confidence <= 1.0, (
+                f"Confidence {result.confidence} out of bounds for var={var}"
+            )
+
+    def test_cross_receiver_agreement_boosts_confidence(self):
+        """Matching results from other receivers should boost confidence."""
+        clf = PresenceClassifier(presence_variance_threshold=0.5)
+        features = RssiFeatures(variance=2.0, motion_band_power=0.0, n_samples=100)
+
+        result_solo = clf.classify(features)
+
+        # Other receivers also report PRESENT_STILL
+        other = [
+            SensingResult(
+                motion_level=MotionLevel.PRESENT_STILL,
+                confidence=0.8,
+                presence_detected=True,
+                rssi_variance=1.5,
+                motion_band_energy=0.0,
+                breathing_band_energy=0.0,
+                n_change_points=0,
+            )
+        ]
+        result_agreed = clf.classify(features, other_receiver_results=other)
+
+        assert result_agreed.confidence >= result_solo.confidence
+
+    def test_result_dataclass_fields(self):
+        """SensingResult should contain all expected fields."""
+        clf = PresenceClassifier()
+        features = RssiFeatures(
+            variance=1.0,
+            motion_band_power=0.2,
+            breathing_band_power=0.3,
+            n_change_points=2,
+            n_samples=100,
+        )
+        result = clf.classify(features)
+
+        assert hasattr(result, "motion_level")
+        assert hasattr(result, "confidence")
+        assert hasattr(result, "presence_detected")
+        assert hasattr(result, "rssi_variance")
+        assert hasattr(result, "motion_band_energy")
+        assert hasattr(result, "breathing_band_energy")
+        assert hasattr(result, "n_change_points")
+        assert hasattr(result, "details")
+        assert isinstance(result.details, str)
+        assert len(result.details) > 0
+
+
+# ===========================================================================
+# Backend tests
+# ===========================================================================
+
+class TestCommodityBackend:
+    def test_capabilities(self):
+        """CommodityBackend should only report PRESENCE and MOTION."""
+        collector = SimulatedCollector(seed=0)
+        backend = CommodityBackend(collector=collector)
+
+        caps = backend.get_capabilities()
+        assert Capability.PRESENCE in caps
+        assert Capability.MOTION in caps
+        assert Capability.RESPIRATION not in caps
+        assert Capability.LOCATION not in caps
+        assert Capability.POSE not in caps
+
+    def test_is_capable(self):
+        collector = SimulatedCollector(seed=0)
+        backend = CommodityBackend(collector=collector)
+
+        assert backend.is_capable(Capability.PRESENCE) is True
+        assert backend.is_capable(Capability.MOTION) is True
+        assert backend.is_capable(Capability.RESPIRATION) is False
+        assert backend.is_capable(Capability.POSE) is False
+
+    def test_protocol_conformance(self):
+        """CommodityBackend should satisfy the SensingBackend protocol."""
+        collector = SimulatedCollector(seed=0)
+        backend = CommodityBackend(collector=collector)
+        assert isinstance(backend, SensingBackend)
+
+    def test_full_pipeline(self):
+        """
+        End-to-end: SimulatedCollector -> features -> classification.
+
+        With a 0.3 Hz sine and some noise, the pipeline should detect
+        presence (variance > threshold).
+        """
+        collector = SimulatedCollector(
+            seed=42,
+            sample_rate_hz=10.0,
+            baseline_dbm=-50.0,
+            sine_freq_hz=0.3,
+            sine_amplitude_dbm=3.0,
+            noise_std_dbm=0.3,
+        )
+        backend = CommodityBackend(
+            collector=collector,
+            extractor=RssiFeatureExtractor(window_seconds=10.0),
+            classifier=PresenceClassifier(
+                presence_variance_threshold=0.5,
+                motion_energy_threshold=0.1,
+            ),
+        )
+
+        # Pre-fill the collector buffer with generated samples
+        samples = collector.generate_samples(10.0)
+        for s in samples:
+            collector._buffer.append(s)
+
+        result = backend.get_result()
+        features = backend.get_features()
+
+        # With amplitude 3 dBm, variance should be about 4.5
+        assert features.variance > 0.5, (
+            f"Expected variance > 0.5, got {features.variance}"
+        )
+        assert result.presence_detected is True
+        assert result.motion_level in (MotionLevel.PRESENT_STILL, MotionLevel.ACTIVE)
+
+    def test_absent_with_constant_signal(self):
+        """
+        A collector producing a near-constant signal should result in ABSENT.
+        """
+        collector = SimulatedCollector(
+            seed=0,
+            sample_rate_hz=10.0,
+            baseline_dbm=-50.0,
+            sine_amplitude_dbm=0.0,
+            noise_std_dbm=0.05,  # very low noise
+        )
+        backend = CommodityBackend(
+            collector=collector,
+            extractor=RssiFeatureExtractor(window_seconds=10.0),
+            classifier=PresenceClassifier(presence_variance_threshold=0.5),
+        )
+
+        samples = collector.generate_samples(10.0)
+        for s in samples:
+            collector._buffer.append(s)
+
+        result = backend.get_result()
+        assert result.motion_level == MotionLevel.ABSENT
+        assert result.presence_detected is False
+
+    def test_repr(self):
+        collector = SimulatedCollector(seed=0)
+        backend = CommodityBackend(collector=collector)
+        r = repr(backend)
+        assert "CommodityBackend" in r
+        assert "PRESENCE" in r
+        assert "MOTION" in r
+
+
+# ===========================================================================
+# Band power helper tests
+# ===========================================================================
+
+class TestBandPower:
+    def test_band_power_single_frequency(self):
+        """Power of a single frequency should concentrate in the correct band."""
+        sample_rate = 10.0
+        n = 300
+        t = np.arange(n) / sample_rate
+        signal = 5.0 * np.sin(2 * np.pi * 0.3 * t)
+
+        # Apply window and compute FFT
+        window = np.hanning(n)
+        windowed = signal * window
+        from scipy import fft as scipy_fft
+        fft_vals = scipy_fft.rfft(windowed)
+        freqs = scipy_fft.rfftfreq(n, d=1.0 / sample_rate)
+        psd = (np.abs(fft_vals) ** 2) / n
+
+        # Skip DC
+        freqs_no_dc = freqs[1:]
+        psd_no_dc = psd[1:]
+
+        breathing = _band_power(freqs_no_dc, psd_no_dc, 0.1, 0.5)
+        motion = _band_power(freqs_no_dc, psd_no_dc, 0.5, 3.0)
+
+        assert breathing > motion, (
+            f"0.3 Hz signal should have more breathing band power ({breathing}) "
+            f"than motion band power ({motion})"
+        )
+
+    def test_band_power_zero_for_empty_band(self):
+        """Band with no frequency content should return ~0 power."""
+        freqs = np.array([0.1, 0.2, 0.3, 0.4, 0.5])
+        psd = np.array([1.0, 0.0, 0.0, 0.0, 1.0])
+
+        # Band 0.21-0.39 has no power
+        p = _band_power(freqs, psd, 0.21, 0.39)
+        assert p == 0.0