Suggestion: allow repeating of randomised tests/benchmarks.

[nishihatapalmer]

Tests and benchmarks are often randomised.

SMART should log the seed used in experiments or tests somewhere, and provide a way to set the random seed via a --seed option.

This would massively simplifiy debugging algorithms and finding subtle errors when they occur.

[nishihatapalmer]

When an algorithm fails on a test or benchmark, it would also be good to log the pattern / algo that failed.