test suite (cli if not web) could in principle keep track of the changelist of the last successful test for each test, and then do a differential test to find at what input/move regressions diverge.

[increpare]

or maybe the cli could write a more detailed trace to disk for the tests that the web tester could access

[increpare]

eh we can do that anyway. not sure it's worth 'caching', and not sure it's worth making a standalone tool to do the binary search.