Signal-to-Noise is too noisy

I was at Web2.0Expo last week, and noted a lot of talk about "Signal-to-Noise."

The whole concept of Signal-to-Noise is really becoming stretched beyond utility now. You know how it started: signal was the waveform that carried information, and noise was any other changes to the waveform - the ones that didn't carry information.  So if your noise had too high of an amplitude, you couldn't figure out what the signal was. It was like the ability to see in the dark... either you could see the sofa -- maybe even just barely -- or you couldn't because it was too dark.

But now, we have something more like gold prospecting, but for any given river, people are looking for different kinds of rocks.  Sure, some are looking for gold - and few would turn it away - but others are looking for zinc and still other want stones of any composition, so long as they're very flat.

So it's more like mining from a slurry with a constant sluice possibly carrying away something you want.

Not really signal to noise, more like "yield."  What's my yield on Twitter? On Facebook's "news feed"? 

And the other component: does Twitter need my undivided attention to get a good yield?

This depends on, first, how time-dependent the tool is: Twitter doesn't have great search so unless you find a 3rd party tool that works for you, you have to look at it all day.... and, second, how fast the decay rate of the information typically is.

For example if I don't look at it for 6 hours because I'm busy, is that ok, because I wasn't going to meet one friend at Peet's and answer another's time-sensitive question?  In that case, the time decay is fast but that's fine given the attention required. However if I am using Twitter to capture information that occurs suddenly but I need to reference for a while... well, then I still need to hover over it all the time if there's no available way to easily search or aggregate responses.

So I am thinking: most web2.0 tools, can they be better described using the metric "Yield-Per-Amount of Attention Spent," rather than "Signal-to-Noise"?

I propose: "yield per glance" and "optimal glance frequency" for passive monitoring; and "yield per interaction" and "optimal interaction frequency" for non-passive monitoring.