Some fastqs (RNAseq) wont map (STAR) after fastp despite almost no trimming, with raw fastqs mapping well

[bmwx410]

I am downloading big datasets from the SRA and using fastp to QC the fastqs.

For some fastqs, despite over 99% of reads PF, looking good on QC etc, when I map them with STAR, they have a uniquely mapped read rate of ~0.5%.
When I map the raw fastqs (before fastp), the mapped rate is excellent at ~ 95% uniquely mapped reads.

Inspecting the fastqs, the sequences, sequence lengths look identical for all inspected reads before and after fastP (inspected read every million).

Any ideas what might be going on here?

Thank you