The places I have worked have all emphasized goal-based metrics. The software developer lays out what they think their responsibilities should be and how they'll be measured on them. The manager then reviews the list with the developer and confirms the responsibilities and the measurements. The developer is then later graded upon his or her ability to meet the predetermined requirements.
I think the same thing could be used for teachers. Seems workable to me.
You argue in one place that scientific metrics are the only way to evaluate educators. And then here you are arguing that managerial judgment is a serious metric. Even though it's entirely non-numeric, personal, unstandardized, unrepeatable, and unreliable.
The same thing was used for teachers: it's called the judgment of their employers. If you're happy with that, then there's no particular need for all the NCLB standardized testing.
Do that successfully, and I'll happily give you a good metric for teachers.