> You can decide for yourself whether an error like this—7 minutes out of 9 hours—makes the real data “close” to a sine wave.

I think the most interesting part of this article is the bit at the end. What really is 'close'? We have so many 'rules of thumb' here but a real definition to target is elusive. Do you go off of pure utility? 'using this definition achieves my requirement to be corret xxx% of the time and now I make money using it' Or do you go off of something more like information content: 'when plotting the error it conforms to a normal curve...' Anyone got a good rule for 'close'?