Added a preprocessing step in the decoder pipeline to discard any database rows containing fewer than two sign groups. By filtering out these inherently less informative or noisy sequences, the frequency analysis and random-init decipherment models can focus on higher-quality, richer data, reducing irrelevant processing.