Personal Best (Gold) Question

Dear QHack team,

Our team Quantics ,has a question about Personal Best (Gold). Our valid submission uses exactly 4ⁿ − 1 rotations (cost 45,012 at 5 qubits, a 21% saving, error ~1e-14, no auxiliary wires) and we believe this is a hard lower bound: only RX/RY/RZ carry continuous angles, so a circuit with k rotations can only produce a k-dimensional family of unitaries, while a Haar-random 5-qubit unitary needs 1023 parameters beyond the free global phase. Gold’s 60% would need about 518 rotations or fewer; auxiliary wires only double the cost, and Clifford+T approximations need more than 44 T gates per rotation at this precision. Is 60% meant to be reachable with a circuit that honestly implements u_mat under the stated cost model, and if so, could you hint at which assumption we have wrong? Also, may code inside submission_circ modify the template’s global state (e.g. GATE_COSTS or the grading functions)?

Thanks!