SWE-bench-Science / tasks /task_005 /instruction.md
Kevin589981's picture
publish SWE-bench Science release
91895bb verified
|
Raw
History Blame Contribute Delete
1.11 kB

You are working with a pinned snapshot of Osprey/FID-A, a magnetic resonance spectroscopy processing pipeline. A collaborator is processing edited MRS data with multiple subspectra. In a reduced synthetic reproduction, one acquisition condition has a 180 degree polarity relationship to the others, and the current pipeline produces an inconsistent aligned/combined result.

Run python reproduce.py and repair the source under source/ so that edited subspectra are handled consistently before alignment and combination. The repair should be general across edited MRS datasets and ordinary load/process workflows.

The symptom may come from inconsistent semantics between data loading and later processing, not only from the final alignment routine.

Do not hard-code the public fixture, a particular sample count, a particular ppm grid, a particular subspectrum count, or fixed expected numbers from the public reproduction. Existing behavior for workflows that intentionally preserve the original acquisition polarity should remain distinct from workflows that need polarity-consistent spectra for alignment.