Hybrid Spiking Neural Networks (SNN) and Artificial Neural Networks (ANN) combine the energy efficiency of SNNs with the high detection accuracy of ANNs, forming a leading approach for event‑based object detection.\ \ Existing hybrid SNN‑ANN systems rely on static inference and select the SNN‑ANN boundary solely based on accuracy and energy consumption, ignoring dynamic inference and system reliability.\ \ This paper introduces BLADE, the first reliability‑aware dynamic boundary selection methodology that supports ANN early‑exit.\ \ The framework jointly optimizes the SNN‑ANN boundary and ANN early‑exit configuration with respect to reliability, detection accuracy, execution time, and energy consumption, incorporating hierarchical statistical fault injection during design‑space exploration to assess reliability.\ \ Experiments on an event‑based detector achieve $\text{mAP}_{0.5}=0.691$, and when the ANN early exit fires, inference compute energy drops to $15.82\,\text{mJ}$.\ \ Reliability analysis reveals that the most significant floating‑point exponent bit is the dominant source of catastrophic failures, causing significant or worse accuracy degradation in $58.8\%$ of fault injections. Protecting this single bit with roughly $3\%$ storage overhead eliminates catastrophic failures under realistic technology fault rates.\ \ Increasing the proportion of SNN computation improves fault tolerance; a fully SNN configuration retains reliability of $0.965$ under aggressive fault conditions.\ \ In summary, jointly optimizing reliability, accuracy, latency, and energy enables more dependable deployment of dynamic hybrid SNN‑ANN systems for safety‑critical edge AI applications.\ \ Review