Compares each marker's threshold location against where the baseline put it, scaled by the baseline's own spread for that marker, so the tolerance adapts to how variable the marker legitimately is. The same comparison is made for population frequencies.
Usage
specification_conformance(
thr_all,
freq,
spec,
baseline,
transform = "arcsinh",
z_qualify = 3.5,
z_fail = 6,
transforms = NULL
)Arguments
- thr_all
this run's thresholds table
- freq
this run's frequency table
- spec
this run's population specification
- baseline
the list from
read_spec_baseline()- transform
this run's transform name
- z_qualify
robust z above which a marker is qualified
- z_fail
robust z above which a marker fails
- transforms
this run's per-panel transforms from
make_transform(), or NULL to compare on the analysis scale only when it is unchanged.
Details
THE SCALE IS FLOORED, for the reason it is floored in the within-run check: a marker whose baseline samples happened to agree closely would otherwise divide by near zero and fail on a difference of no consequence.
THE CUTS ARE COMPARED IN INSTRUMENT UNITS, on the baseline's scale. Thresholds are expressed on the analysis scale, and that scale is re-derived from each run's data: the arcsinh cofactor of the same 30 files moved from 261 to 403 between two versions of this package, which moves every cut on the transformed axis without moving it on the instrument's. Comparing the transformed values across runs therefore measured the change of ruler. Each marker's median cut is carried back to instrument units by this run's transform and forward by the baseline's, then compared there against the baseline's spread. A baseline written before transforms were stored is rebuilt from its arcsinh cofactor; where either transform cannot be rebuilt, a changed transform makes the comparison "not comparable", as before.