bruteforce incorrect ulp error report for cos(double)

Running 'bruteforce -rf1 cos' on my implementation produces:

1: cos fp64 ................passed 2.27 @ 0x1.d10000000074p+380

Now, for that particular input argument, my implementation produces:

0.46 ulp at (0x1.d100000000740p+380): *0x1.ffff1d06dd23f8a8p-8 vs 0x1.ffff1d06dd240p-8

Wolfram claims the correct answer for this argument is
0.0078124471536865077073474543921834934716746213988958700965071438...

Which in hex rounds to: 0x1.ffff1d06dd240p-8 in double precision and identical to what my implementation produces.

So why is bruteforce claiming my implementation has a 2.27 ulp error for this input?

Start by reproducing the report with `bruteforce -rf1 cos` and compare the displayed input, implementation result, and Wolfram reference for the reported double value. Trace how the bruteforce command computes and formats ULP error; done means the tool explains or corrects the discrepancy for this case.

Written by the indexing model from the issue text.