"... In PAGE 9: ...f trail users on the natural environment ................................................................... 68 Table10 : Importance of surfacing considerations to trail managers.... In PAGE 79: ...69 The construction cost for different surfaces is a strong consideration for many trail managers. In the manager survey, cost was tied for third overall ranking behind suitability for user requirements and site constraints ( Table10 ). With tighter park funding for many jurisdictions over the past decade, both the cost of trail construction and maintenance requirements have become heavily scrutinized.... ..."

"... In PAGE 10: ... TRAIL also excluded from the test sets molecules whose fragments were not presented in the compounds of corre- sponding training sets. The results given in Table5 show that average statistical criteria obtained at the training (R 2 ) 0.... In PAGE 12: ... Indeed, differentiation of bonds in cycles and in chains (see the Results section) led to an increase of the number of variables for the whole set, but the statistical criteria of fitting (R 2) and LeaVe One Out prediction (Rcv2) become better (Table 4; Figure 2). Similar conclusions can be drawn when comparing statistical criteria of predictions for test sets 1-3 ( Table5 ; Figures 3 and 4). Another possibility to improve the robustness of the TRAIL model is to change the threshold for acceptance of molecules having fragments of rare occurrence .... In PAGE 12: ...re 0.780 and 4.76 (CODESSA-PRO, Table 3), 0.733 and 8.80 (TRAIL, Table5 ), and 0.866 and 3.... ..."

"... In PAGE 13: ... In the other benchmarks a much smaller fraction of unifications can be improved due to the heavy use of non-deterministic predicates. Table2 presents the timing results of each benchmark for the given number... ..."

"... In PAGE 12: ...68 58 C3 static 12,361 24,493 9.36 58 Table2 . Finding a safety violation with A* and several reduction methods.... In PAGE 13: ... In other words, the bigger the reduction is, the longer the stuttering equivalent executions and, therefore, the longer the expected trail lengths become. Table2 also shows that the overhead introduced by partial order reduction and heuristic searchdoesnot avoid time reduction. 4.... ..."

"... In PAGE 13: ...able 3. Finding a safety violation with IDA* with and without reduction. the two C3 conditions C3 duplicate and C3 static that can be applied jointly with A*. Table2 shows the e ect of applying C3 duplicate and C3 static in conjunction with A*. The table has the same format as the previous one, but this time the length of the error trail is included.... In PAGE 13: ... In other words, the bigger the reduction is, the longer the stuttering equivalent executions and, therefore, the longer the expected trail lengths become. Table2 also shows that the overhead introduced by partial order reduction and heuristic searchdoesnot avoid time reduction. 4.... ..."

"... In PAGE 32: ...3 Effect of the improved trailing scheme combined with trailing analysis Ltrail in the Mercury back-end of HAL The trailing analysis presented in Section 4 and implemented in HAL, was modified, as proposed in Section 8, to deal with the improved trailing scheme. Table7 presents the timing and maximal trail for the HAL benchmarks obtained under the improved scheme with the information inferred by the modified analysis, and compares the results obtained under the same scheme without any analysis information. Table 7.... ..."

"... In PAGE 32: ...3 Efiect of the improved trailing scheme combined with trailing analysis Ltrail in the Mercury back-end of HAL The trailing analysis presented in Section 4 and implemented in HAL, was modifled, as proposed in Section 8, to deal with the improved trailing scheme. Table7 presents the timing and maximal trail for the HAL benchmarks obtained under the improved scheme with the information inferred by the modifled analysis, and compares the results obtained under the same scheme without any analysis information. Table 7.... ..."

"... In PAGE 10: ...4 Comparison of the classic and improved scheme with respective trailing analyses [13] proposes a trailing analysis on top of the classic PARMA trailing scheme. Table5 compares the current work with the results obtained in [13]. The timing differences between the two systems with anal- ysis show a similar pattern as those without analysis: for some benchmarks the improved systems is better because more trailings were avoided, for others it is worse because of additional instruction overhead.... ..."

"... In PAGE 23: ... Similarly, node 7 is the parent of two tip nodes, numbered 11 and 12. Table2 shows the tips, their induced trails and the score of these trails... ..."