LURCH (Menzies et al. 2004) uses random search to explore a state machine’s state space. Because the search is random, it gives no guarantee that the state space has been exhaustively explored, so LURCH cannot be used for verification. However, any errors it finds are real errors, so it can be used for debugging. It can find assertion violations, deadlock, and cycle-based errors (no-progress cycles, and violations of temporal logic properties specified as acceptance cycles); finding cycle-based errors requires a state hash table. LURCH’s input format is concurrent finite-state machines, where transitions may be specified by C code.
In a case study of running model checkers to find seeded errors in a flight guidance system model, NuSMV (a model checker not described in this report) had a mean runtime of 2 hours where LURCH had a mean runtime of 3 minutes, yet LURCH found 40 of the 44 seeded errors that NuSMV found. In other demonstration problems, LURCH found all the errors that SPIN found when SPIN could terminate with the given resources. LURCH is claimed to scale to larger problems, and tends to use a stable amount of resources.