Table 1.

Comparison of different repair models.

	Change in error rate (%)
	Semantic	Total	Successful repairs (%)
Gain model (gain = 3)	–35.2	–31.0	85.5
Feature layer (mf = 3)	–33.8	–29.7	84.3
Target features (fr = .03)	–37.0	–32.2	86.7

Note: Criterion = 4.96, one extra time step. Each model run 10,000 times.