TheWhippinpost - 1:37 am on Jun 8, 2005 (gmt 0)

What these docs seem to be spelling out are the apparent weaknesses (at least at the time of publication - and most likely still, in large part) of the algo(s).

I don't get the impression of direct human intervention being carried out, rather, either the delegation and investigation of submitted spam reports to paid staff, a "before and after" snapshot "between dances", or the flat "admission-by-inference" that automation alone cannot detect these techniques (yet!).

I mean, everything reported - apart from perhaps "white-on-white" text which is very highly probable to be spam - have (from memory) grey areas attached which automation alone cannot distinguish - Hidden text for instance; spam, or a means to save the speed-reader time? a'la MSDN which commonly use hidden text for hiding code examples?

[I just thought I'd drop that in 'cos as of page 5 everyone seemed more concerned with the rights & wrongs of human tweaking - Apologies if this has been raised in the interim but I guess not many will be reading - what's this? page 9 now? - this anyway!]