Daniel's testing shows a 1% difference between the tools. Any given model of impact is going to have at least a 10% swing in performance over a large sampling of tools. The testing of extended vs non-extended will show a difference in the average when testing a large enough sample, however, it was not enough to differentiate in the rating of the tools in the catalog (although more than 1%). Daniel has done a nice job with the test, however, for the sake of having the most valid experiment possible, in the future, it would make sense to eliminate as many variables as possible. I have no vested interest in the results, just pointing this out.