Junior has mentioned that I could also consider the leader sequence as an additional property of a true CRISPR system.

SignalP. Signal sequence in leader sequence.

• https://developer.nvidia.com/nsight-visual-studio-edition
• Leap checking rather than checking/creating every dyad/kmer?
• Only store/create kmer if dyad rather than a post-process filter?
• Windowing based fast detection of repeats (pass 1) followed by careful analysis of high-repeat domains for specific properties such as spacers (pass 2)?
• May not require iteration for Dyad predicate. Just check minimum provided length?
• As alternative to local alignment for Cas gene discovery, can do k-mer based similarity scores (higher-performance)
• Alternative to string comparison is XOR -> AND -> popcount of 3-bit encoded nucleotides (as opposed to 2-bit encoding to reduce an extra step associated with filtering down the popcount due to one of the 4 nucleotides having to be encoded with two 1s)
• Check for validation parameters only if the CRISPR loosely matches the definition. For example, maybe spacer sequences don't need to be aligned if the repeat array is perfect.
• Distributed computing via HPC.
• Usage of high physical memory consumption (RAM or VRAM) via HPC.
• CUDA best practices

Use contigs, not raw reads.

Closed in 472d511

Literature review

• Inclusion of literature on CRISPR structure ambiguity such as degeneration and tandem repeats.
• Inclusion of review of evaluation of precision and recalls of existing tools.

Just ended up deleting the old stuff and replacing it with a new build 472d511