Explaining runtime scaling differences via time limits (conjecture)

Ascertain whether the observed discrepancy in runtime scaling for geometric complete-graph instances between the current manuscript and Stein–2023 is caused by different runtime limits (60 seconds versus 1000 seconds), by conducting controlled experiments and analyses that isolate the effect of early-size instances on fitted scaling trends.

Background

In reproducing prior experiments on new hardware with different time limits, the paper reports a worse apparent scaling exponent for the geometric data set on complete graphs than previously published.

The authors conjecture that the difference is due to the shorter runtime cap disproportionately affecting small-size instances that influence the fitted trend; verifying this conjecture would clarify the methodology for reporting scaling behavior.