I think the existing diagnostic is consistent with the corrected relation. The validity condition is P_phenotype ⊆ P_participants, while the failure condition is P_phenotype \ P_participants ≠ ∅. So saying that the phenotype TSV contains subjects absent from participants.tsv describes the error correctly.
The proposed alternative (“contains some or all participants listed in participants.tsv”) would also describe a valid file.

“A phenotype TSV contains participant_id values that are absent from participants.tsv.”
It may also be worth testing both directions explicitly, since equality would hide the original reversal: {01,02} ⊆ {01,02,03} should pass, whereas phenotype {01,03} against participants {01,02} should fail.

I think the existing diagnostic is consistent with the corrected relation. The validity condition is P_phenotype ⊆ P_participants, while the failure condition is P_phenotype \ P_participants ≠ ∅. So saying that the phenotype TSV contains subjects absent from participants.tsv describes the error correctly. The proposed alternative (“contains some or all participants listed in participants.tsv”) would also describe a valid file.

Ah, of course! Thanks for spelling that out.

Thanks for having a look, both!