Assess performance under fixed hardware timing

Determine whether the learned Shellsort gap sequence retains its empirical advantage over the tested classical gap sequences when evaluated using a fixed hardware timing benchmark rather than comparison and move counts.

Background

The empirical evaluation measures exact comparisons and array writes, using equal-task geometric means over 25 large tasks. It does not measure wall-clock execution time, and the conclusion explicitly distinguishes the reported operation-count claim from hardware timing.

The unresolved question is whether the sequence's operation-count advantage translates into a real-time advantage under a controlled, fixed hardware and implementation environment. Such an evaluation would test the practical robustness of the learned prefix beyond the abstract cost metrics used in the paper.

References

And does the sequence keep its advantage under a fixed hardware timing test?

— A New Gap Sequence for Shellsort: RL-Driven Algorithm Discovery Beyond $N^{4/3}$  (2609.29881 - Liu, 24 Sep 2026) in Conclusion, Section 5