Determine whether deficit-repair prioritization can redistribute corpus coverage

Determine whether modifying the DDSA repair loop to prioritize under-filled strata over overall recovery yield can redistribute the resulting corpus toward underrepresented strata.

Background

The paper’s ablation finds that the human-in-the-loop repair stage adds pairs without measurably changing the corpus distribution across the sampling frame. The authors therefore attribute stratum balance primarily to the deficit-driven planning stage rather than to repair.

A possible alternative design would prioritize retrying failures from under-filled strata even when that reduces aggregate recovery yield. The paper does not evaluate whether such a policy would produce meaningful redistribution, leaving the question unresolved.

References

Whether a repair loop can be made to redistribute --- by prioritising under-filled strata over overall yield when choosing which failures to retry --- is a question this design does not answer.

— Deficit-Driven Stratified Acquisition for Building STRATA-S12: An Open SAR-Optical Corpus for Self-Supervised Learning  (2610.17680 - Sanchez-Leon et al., 13 Oct 2026) in Section 8, Discussion, Section 8.2 Limitations, item (vii)

The formal guarantee of Section~\ref{sec:formulation} covers the planning stage only: the greedy plan is $(1-1/e)$-optimal for the coverage objective, but the deficit-repair loop that acts on the realised (post-gate) coverage is characterised operationally --- as a bounded iteration measured by its recovery efficiency $\eta$ --- rather than with a coverage guarantee of its own. Establishing under what availability conditions the loop provably closes the residual deficit is left as future work.

— Deficit-Driven Stratified Acquisition for Building STRATA-S12: An Open SAR-Optical Corpus for Self-Supervised Learning  (2610.17680 - Sanchez-Leon et al., 13 Oct 2026) in Section 8, Discussion, Section 8.2 Limitations, item (ix)