Bonferroni treatment of correlated self-evolution attempts

Determine whether the live Bonferroni correction for the Rank-IC gate should count each self-evolution mutation separately in the cumulative candidate count or discount correlated mutation attempts.

Background

SEADS applies a live Bonferroni correction to the Rank-IC admission threshold using the cumulative number of candidate trials. Self-evolution generates mutations from the same parent candidate rather than fresh independent trials, creating dependence among the associated statistical tests.

The paper explicitly leaves unresolved how these correlated mutation attempts should enter the multiple-testing correction: counting every mutation separately is conservative, whereas discounting correlated attempts could better reflect their effective number of independent tests. The choice affects the admission threshold and therefore the evaluated productivity and performance of the discovery system.

References

Its mutations are correlated draws from the same parent candidate rather than fresh independent trials; whether the rank-IC gate's live Bonferroni correction (Appendix~\ref{app:metrics}) should count each mutation separately toward the cumulative candidate count or discount correlated attempts is left for future work.

— Agentic Empirical Asset Pricing: Methodological Foundations  (2609.00731 - Pan et al., 1 Sep 2026) in Appendix, Section “Pipeline Internals,” paragraph “Self-evolution”