Sage fails silently if the protein database does not contain decoys in combination with `"generate_decoys": false`

[fabianegli]

Dear Michael

Thanks for this great and really fast search tool!
Today I spent quite some time figuring out why sage was failing on me and generating lfq.tsv files containing only the header.
In the hope the next person hitting this roadblock spending less time on it than I did today, I file this issue.

Observed behavior

If the FASTA protein database does not contain decoys and the config doesn't indicate that they should be generate, sage fails silently with only the following warning:

[2025-09-22T15:56:36Z WARN  sage] linear model fitting failed, falling back to heuristic discriminant score

The config would be the following:

{
  "database": {
    ...
    "generate_decoys": true,
  },
  ...
}

And the database one with no decoys.

Expected behavior

I would expect sage to fail in this case with a warning that there are no decoys in the database and that they either should be added to the added or the config changed so they are generated by sage.

[lazear]

I think it makes sense to issue a warning that FDR will not be calculated (and thus no LFQ enabled). I'm not sure if it is best to actually exit and fail the program though - sage will also 'successfully' complete when provided with empty spectra files, which I see as OK behavior (it is a functional mapping from spectra -> PSMs, and the empty list is a valid input).

[fabianegli]

I understand your viewpoint that it is convenient to get whatever is possible to get on the input available, but it also means that in some situations users may think they got what they wanted as stated in the config but actually didn't and will only at a later date realize it didn't work as they expected. I think this behavior has the potential cause users headaches.

One idea would be to introduce a --brittle to make sage fail and tell if there are inconsistencies between different configuration parameters or configuration parameters and file availability/contents. The alternative, and I would prefer that and believe it makes it more predictable to users would be a flag for the "I run through anything" like --terminator.

[fabianegli]

I am also thinking about weather it would be a good idea to also save a diff of the original configuration and the one that was actually used in the results folder. Otherwise it might become difficult to understand the discrepancy between the intent of the original config and what has been done if only the output folder is kept and the original config lost.