From email conversation with Rich Johnson. He was working with a file that looks like this:

file    scan    charge    adduct    inchikey    chemicalformula    moleculename    hmdb or cas    retentiontime    width    MW    MolecularIon    frag1comp    frag1calc    frag2comp    frag2calc
QEP1_2021_0823_RJ_21_2ab55prm.raw    3423    1    [M+H]    QQVDJLLNRSOCEL-UHFFFAOYSA-N      C2H8NO3P    (2-AMINOETHYL)PHOSPHONATE    HMDB11747    5    2    125.0241796    126.0320046    CH6O3P    97.0054556    H4O4P    98.9847214
QEP1_2021_0823_RJ_31_2ef55prm.raw    5909    1    [M+H]    XFNJVJPLKCPIBV-UHFFFAOYSA-N    C3H10N2    "1,3-DIAMINOPROPANE"    HMDB00002    8.9    2    74.0844    75.09222333    C3H8N    58.0656708    na    na

Note that the header line has two pairs of [fragment formula, fragment m/z] (obviously there could be any number of pairs in practice). Notice also that the second data line has "na" for its second pair values - only one fragment for that precursor.

It's **very** inconvenient to smith this kind of file into a "one transition per line" format that Skyline currently insists upon. We should be willing to consume more than one fragment per line, and I think we could reasonably insist that they be at the tail of the line, and that they each have a consistent set of descriptors. And of course we have to deal with different lines potentially leaving some of those sets empty since not all precursors have the same fragment count.