Variations of the standard genetic code are pretty rare, but as the cost of high-throughput genome sequencing continues to drop, there is a greater possibility of discovering additional exceptions. That being said, there is a clear emphasis in genome projects on nucleotide (genome and transcriptome) sequencing, with much less (if any) effort put into proteomics work (correct me if I'm wrong there).
Let's assume we're sequencing the genome of a new organism and we're focusing completely on genome and transcriptome sequencing--no proteomics. Let's also assume this organism has slight variations to the standard genetic code. Would it be possible to annotate this genome (for protein-coding genes) completely incorrectly since the gene prediction software does not take into account these variations, or would it be pretty obvious? What would you expect to see in this case?