Multi-task regression: every claim measured

Direct Algorithm 1 risks, three transfer exponents, exact low-rank balancedness, population comparability, bounded-domain logistic regression, and destructive controls.

1.4098

maximum linear safety-rate ratio over 6,480 task risks

−1.0087

measured task-sharing exponent; theory predicts −1

17.23×

median logistic transfer gain over independent fitting

0 / 1,200

population-comparability violations

14,197

exact finite low-rank balancedness pairs

3.67e9

unsafe pooling ratio at outlier radius 100,000

Four-panel claim-native audit
Boundary: finite native-scale execution reconstructs the registered mechanisms and failure controls. The pinned source—not these finite measurements—carries the universal high-probability statements.