RF ensemble classifiers often abstain on very shortIQ sequences (N < 32), forfeiting coverage under burst or bandwidth constraints. We present a policy-driven framework with alightweight learned short-signal head and show it improves theaccuracy-coverage frontier versus strict abstention and paddingbaselines. On synthetic truncations of four modulations, the headattains higher utility across N ∈ [12, 28) … Continue reading Short-Signal Resilience: Learned Heads and Policy Boundaries for N < 32 IQ Classification