"... In PAGE 10: ...iterature (c.f. difference T4). The prioritizations at both levels ( Table1 ) were done with the same technique. The prioritization was conducted by the chief architect as he is the person making choices between metamodel extensions and profiles in most cases.... ..."

"... In PAGE 5: ... 5.2 Force Resolution The context controls which forces are most impor- tant, as shown in Table1 . The \Innovation quot; entries denote problematic contexts where good performance requires either ad-hoc locking primitives or a redesign to achieve good performance.... ..."

"... In PAGE 9: ... The ran- dom and additional functional coverage techniques, however, presented great variation in results that cannot be explained based solely on the increase in granularity. The Anova presented in Table7 con#0Crms these observa- tions relative to optimal and random techniques. #28Anovas for all pairs of techniques are given in the Appendix; we omit the other two here because they are similar to the one in Ta- ble 7.... ..."

"... In PAGE 9: ... The ran- dom and additional functional coverage techniques, however, presented great variation in results that cannot be explained based solely on the increase in granularity. The Anova presented in Table7 con rms these observa- tions relative to optimal and random techniques. (Anovas for all pairs of techniques are given in the Appendix; we omit the other two here because they are similar to the one in Ta- ble 7.... ..."

"... In PAGE 5: ... As well as being tested against each other, we also compare these algorithms to Prioritized Sweeping which we showed in Uther amp; Veloso (1997) was one of the better non-generalizing algorithms. In Table1 Fitted Prioritized Sweeping is compared to standard Prioritized Sweeping. It learns signi - cantly faster than standard Prioritized Sweeping, and keeps performing better even after the initial games.... ..."