A sampling technique involving more samples but allowing missing segments. Another technique for incrementally increasing the sample size if the test fails.
This attempts to tackle the reality that in networks there are transient errors, and we might obtain false negatives. In our case, we have extra redundancy (the subnets), which also helps, but mechanisms like this would add an additional layer of resiliency and potentially latency concluding the DAS test.
Finally, it proposes a better way to select than random (distinct rows and distinct columns), which is designed for Ethereum's 2D structure but perhaps can be adapted.