This paper analyzes faster-than-Nyquist (FTN) signaling within a consistent framework based on a fixed time-bandwidth product (TBP), resolving potential ambiguities present in finite blocklength analysis. A key feature of FTN is its ability to increase the number of transmitted symbols in a given time and frequency resource, which can lower the rate penalties inherent in short packet communications. We derive tight bounds on the maximum channel coding rate (MCCR) and demonstrate that FTN's rate gains over Nyquist signaling can be higher in the finite TBP regime than in the asymptotic case. Performance is benchmarked against the theoretical optimum of transmitting prolate spheroidal wave functions, showing that a well-designed FTN system can closely approach this limit. We present practical design criteria, including the optimal time-acceleration factor for maximizing signaling dimensions, an optimized pulse shape that meets strict out-of-band constraints, and a turbo-equalization-based coding scheme that performs near the derived MCCR bounds. These findings establish FTN as a practical and near-optimal technique for enhancing the rate and reliability of latency-constrained communications.