Claude read it now. Same findings on the handshake (the fixed 18-value sequence ending in 2) and on the wasted cycles (in the port the ~1200-byte copies to and from shared RAM are patched out). The approach differs: you patch the jump table into direct calls, while the port leaves the 68000 code untouched and answers as the HuC would, including a 68000 rewrite of the 6280 collision routine, reading its data tables from the uploaded program image in ROM. It didn't consult your post while making the port, but it can't rule out having seen it during training.