I am working on a machine-learning invention that can be applied in the narrow field in which I work (mainframe system programming). My English is not supreme but I need to define what I am working on (for marketing and documentation purposes) very business-y and correctly.
Let me explain what is the thing I am looking a word (or expression) for: We have 500 tests that are run in parallel. They aim at simulating a typical mainframe load while the testware sniffs for a system failure. Once a system failure occurs, the current modus operandi is that I, MANUALLY, try to reduce the 500 test set to a subset (I call it "the special mix") that will reproduce the problem but with less noise and less functionality being tried (leading to faster root cause analysis).
Hence, I am looking for the "test or tests that, when run, causes the system failure"... Well, the test doesn't "cause" the failure... The system developer allowed the logical error into the code... "Relevant" also doesn't work... All the tests are relevant :)
The best I have right now: "exact scope subset that reproduces the failure".