Better averaging of judges ratings

[epugh]

Is your feature request related to a problem? Please describe.
Today we just straight up average. @david-fisher doesn't love this!

Describe the solution you'd like
if i have 3 or more, then I take the three highest ratings.

if all thee agree, i am happy.

if all three do NOT agree, then I take the min of all three.

[jvia]

Curious why this would be a better approach.

[david-fisher]

When combining the judgments from multiple judges, we can make a variety of assumptions. If we assume:

1. Judges are generally subject matter experts (SMEs)
2. Judges are rational
3. Judges have individual biases when interpreting relevance

Appealing to 1, we take the three highest out of all the judgments there are. Note, many additional judgments could all agree with these first three. Here we are being optimistic.
If the judges do not agree, we appeal to numbers 2 and 3. If a rational judge deems the candidate to be poorer than the other judge or judges, we will trust them, and use their value. Here we are both being pessimistic, concerning the individual judgment, and optimistic, concerning the direction of the bias. That is, we expect judges to overrate, in the general case.

As cases grow, and the relative quality of the judges becomes clearer, alternative combinations could be used.