Scientific figure registry · 2026-09-09
Every plot carries its unit, interval rule and source table.
Seven reusable views now cover all six priority figure classes: source effects, dataset-task-result coverage, reported sample/donor distributions, trajectory-time alignment, benchmark intervals and selective risk-coverage. Missing intervals remain visible as missing; effects on different scales are not pooled.
Target 360
Source effects remain separate
Three example genes cover a governed candidate, a broad atlas gene and a prospective design candidate. The frozen source summary has no effect intervals, so the plot marks that limitation and does not pool points.
Benchmark uncertainty
Four metrics, four axes, no cross-metric average
Data qualification
One dataset can qualify for one task and fail another
Result distribution
Improvement and failure counts use held-out targets as units
Reported sample units
Sample and donor scale stays visible before biological inference
Retained cell or spot counts are displayed for every reported unit in the three external disease-context datasets. Source metadata does not distinguish donor identity for every record, so the plot remains a QC distribution.
Trajectory and physical time
Unperturbed myogenic programs are aligned to 0, 24, 48 and 72 hours
The six frozen programs use mean per-gene z-scores. Cells are nested within one public differentiation experiment, and the view does not imply a perturbation-conditioned or DMD trajectory.
Selective risk and coverage
Support-aware utility changes across dataset groups and accepted coverage
The frozen ECDF policy is shown separately for KCP, McFarland and GSE314342. Values below zero favor the accepted subset relative to always-ridge; the curves are not pooled into a universal benefit.
