WBC standards don't reflect real world usage - Page 3

Want to talk espresso but not sure which forum? If so, this is the right one.
gscace

#21: Post by gscace »

another_jim wrote:I guess I'm not getting a lot of feedback on the profile independent error measures. I wonder if it's worth working on.

To Espressoperson. With serial correlation (i.e if one observation is X degrees, the next one is going to be X or very close to X), measures of time on or off target can become deceptive as well.

As a correction to myself, on second thought, a standard deviation will work unmodified as a measure of accuracy. However, because of the correlated observations, calculations on how repeatable that standard deviation will be on subsequent tests will be off in an optimistic direction (that is, the number of independent readings one has gathered is less than the number of correlated observations, so any confidence interval derived using calculations based on independent observations will be too optimistic).
Howdee Jim:

Don't let my silence stop you. I've got several fish frying at the same time right now and I'm not a prolific as i could be. I think that the analysis that you propose solves a problem because it quantifies reproducibility in profiles that are not flat. It's certainly worth doing. The trick will be to get people to accept the analysis and to actually perform it. The standard won't be any good at all if people don't use it. Personally I think it supplies additional useful info, so it's worth investigating.

-Greg

Ken Fox

#22: Post by Ken Fox replying to gscace »

The only way any profile studies will bear any fruit will be to pair them with taste tests. A high degree of reproducibility of a shot temperature profile is of importance only if it tastes good. Otherwise, the information is irrelevant.

Every machine design out there has its own temperature profile, plus its own ability to produce the same profile on a repeatable basis. I think other than the basic ability to be temperature stable, which is fairly easily tested, the rest of this puts us very much into "cart in front of the horse" territory.

I may not be understanding what Jim proposes well enough, but my impression is that some very simple statistical measure that takes a few data points into consideration, such as peak temperature at the "hump," at shot initiation, trough temperature, and maybe lowpoint temperature would be enough to find out if a machine is temperature stable. Once we could say it was so, then we should taste the results. Once we know which machines pull the tastiest shots, then we can go back and try to figure out if the temperature curve is closely related to the temperature or if it is due to some other factor such as pressure or whatever.

ken