to be fair a Jev architecture would be better optimized for this particular workflow than an LLM.

    <think>\n\n</think>

but letting an LLM think would trade latency and performance for significant reliability above that of Jev.