What is the primary purpose of a convergence condition in a critic agent loop?
Answer: To prevent infinite loops by defining when the critique-revision cycle should stop
A convergence condition defines when the critic-revision cycle has done enough work and should terminate. Without it, the loop can oscillate indefinitely between critique and revision, or continue refining past the point of improvement. The condition typically combines a maximum iteration count (hard limit) with a quality-delta threshold (stop when improvement per round falls below epsilon). While limiting API costs is a side effect of having a convergence condition, it is not the primary purpose. Convergence conditions are orthogonal to parallelism and schema validation.