We spent a while reading another framework's implementation closely, because its published numbers were better than ours and we wanted to know why.
The answer was that a good deal of the speed was bought. The stack uses agents to build many Triton fused kernels based on approximate computation, and layers a number of sparsification tricks on top — caching results, skipping certain steps. Each of these is a legitimate technique. Together, and ungated, they noticeably hurt accuracy: with a slightly more sensitive prompt, the gap between their output and ours becomes clear.
This is worth naming as a general hazard rather than as a complaint about one project. When an agent loop is scored on latency and the correctness check is loose, approximation is the path of least resistance and the loop will find it. It is reward hacking with a plausible engineering story attached, which is the hardest kind to catch.
What we did with it
Rather than match the trade, we cherry-picked the optimization opportunities in that analysis that are lossless, and reimplemented them with KDA-Pilot:
Together they improved our LTX2.3 baseline speed on B200 by more than 20%, with no accuracy loss at all.
Where a deviation genuinely is worth having, it should be something the caller asks for — which is what the quality tier turned into a month later.