pirl-unc / hitlist / 25179647527
80%

DEFAULT BRANCH: main
Ran
Jobs 1
Files 26
Run time 1min
Badge
Embed ▾
README BADGES
x

If you need to use a raster PNG badge, change the '.svg' to '.png' in the link

Markdown

Textile

RDoc

HTML

Rst
coverage: 70.511% (+0.03%) from 70.482%
25179647527

push

github

web-flow 
v1.30.15: cross_reference emits needs_deconvolution sentinel for class-only PMIDs (#200)

Found while running v1.30.13's qc plan against the corpus:
Bassani-Sternberg 2017 (PMID 28832583) has 53,689 rows, ALL with
mhc_restriction = "HLA class I" (IEDB lost the per-peptide allele
attribution). The YAML correctly lists 32 alleles from donor
genotypes — but cross_reference reported each as a yaml_only
divergence, ballooning the report and inflating curation_plan
priorities.

The fix: when a PMID has data but ZERO 4-digit alleles in any
observation row, collapse the per-allele yaml_only emission into
a single needs_deconvolution sentinel row tagged with the YAML
allele count. The structural gap (paper-known allele lost on
import) is the same fact whether you list it 1 time or 32 times,
and the discrepancies report's class_pool_n / monoallelic_class_only_n
already surfaces these studies with row counts.

Effect on qc plan top 10: Bassani-Sternberg drops out entirely
(was #5 with priority_score 32000 from the phantom yaml_only
weight). Marcu 2021 also dropped earlier from the v1.30.14 fix.
Top 10 is now genuinely high-priority: Sarkizova 2020 leads with
real suspect_class_label and (pre-rebuild) PTM-annotation noise.

The yaml_only path still fires unchanged for PMIDs with at least
one 4-digit data allele — those are real curation gaps where
deconvolution isn't structurally blocked.

3589 of 5090 relevant lines covered (70.51%)

0.71 hits per line

Coverage Regressions

Lines Coverage ∆ File
8
93.29
 0.23% qc.py
Jobs
ID Job ID Ran Files Coverage
1 25179647527.1 26
70.51
 GitHub Action Run
Source Files on build 25179647527