...down.

Other definitions are possible; e.g., the number of click-throughs for which A should be paid might not include those in which B responded with an error message. Nevertheless, in most cases the claims we make for our schemes continue to hold even under other reasonable definitions.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...7).

The scheduled delay (in Figure 7, of one second) before invoking notify avoids various race conditions in IE4, yielding a more robust implementation for this platform.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...toolbar,

This does not completely prevent a user from loading a different page into the nextpage window before pageB.html loads: e.g., the user may still use ``drag and drop'' features or keyboard shortcuts to load a different URL into the window. However, we expect that the percentage of users employing these mechanisms is small and thus that users loading different pages into the nextpage window will yield insignificant error in the lower bound, especially since there is typically a very limited time frame (i.e., before site B responds) in which the user would need to load the different page.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...Crowds [RR98]

See www.anonymizer.com, lpwa.com, and www.research.att.com/projects/crowds, respectively.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.