Read an uncomputed annotation score from either Console notation

[htsugawa]

Consumer-side counterpart to systemsomicslab/MsdialWorkbench#785, which changes how MS-DIAL Console writes an annotation score that was never computed.

Problem

parse_mdpeak and parse_mdscan depended on the -1 sentinel:

if all(value is not None for value in (weighted, simple, reverse, matched_percentage, matched_count)):
    if all(value >= 0 for value in (...)):
        scored_count += 1
    scores.append({...})

So a precursor-only row was a candidate but not scored — which is what the tuning panel shows as MSP reference candidates versus MS/MS-scored candidates.

A Console that includes the shared AnnotationScoreFormat writes null into every score column of such a row instead of 0.000 dot products plus -1 matched-peak values. Read with the old rule those cells are None, the row drops out of both counts, and the two numbers collapse to the same value.

Change

• Candidate count now comes from the annotated name, which is independent of the score notation. Any name other than Unknown or empty is a reference candidate, including the no MS2: , w/o MS2: and low score: prefixes. This reproduces today's number against both an older and a newer Console, so the app works with an official Console package and a local source build alike.
• Scored count keeps its meaning — rows with a spectral comparison. A row whose comparison genuinely produced zero still counts as scored, because those zeros are measurements.
• Both parsers now share one _annotation_score_row helper, so they agree.

One corrected number

The .mdscan fixture's candidate count changes from 2 to 1, and that is a fix. GcmsAnalysisMetadataAccessor writes -1 into every score column when there is no match result at all, so the old rule counted an Unknown GC-MS row as a reference candidate.

Tests

python -m unittest discover -s tests — 66 pass.

• test_parse_ascii_mdpeak keeps its assertions, with the fixture given the Name column that a real .mdpeak always carries.
• test_parse_ascii_mdpeak_reports_same_counts_for_null_uncomputed_scores is new: the same counts from the new notation.
• test_parse_ascii_mdpeak_keeps_a_comparison_that_scored_zero is new: an all-zero row that a real comparison produced still counts as scored.

🤖 Generated with Claude Code