The encoder has strange PSNR drop in individual parameters?

Hello.

I used audio-codec-benchmark to test the performance curve of the exhale and found that there was a strange PSNR drop at 44.1KHz, 48KHz with the encoding parameter f. (PSNR on the left, A-weighted PSNR on the right)

I examined the error spectrum and found more observable errors across the band when the coding parameter f is used for 44.1KHz, 48KHz audio.

After timeline testing, the whole sample without any delay but looks like caused by frequency domain errors. I don't know if this is due to a special quantisation used for that parameter or some kind of bug?

Thanks again.